Wiki Index

Last updated: 2026-06-27

Entities (404)

• 9elements — German firmware and cybersecurity company (9elements GmbH), founding member of the Open Source Firmware Foundation and ISV partner for post-warranty firmware support.
• A. Stryker — Storage technologist at Solidigm presenting on liquid cooling challenges for SSDs at OCP Global Summit 2025.
• ADTechnology — Samsung Foundry Design Solution Partner responsible for hardening and implementing the Arm Neoverse V3 CSS CPU chiplet on Samsung 2nm GAA for the four-party AI CPU chiplet platform.
• AKCP — AKCP is a US environmental monitoring and data center infrastructure management (DCIM) company founded in 1981. Its flagship Quicklime DCIM platform, now in its fourth generation with 26 years of continuous development, targets AI data center operators who need to reconcile real-time sensor data with computational fluid dynamics models and AI-driven operational insights.
• AMD — AMD (Advanced Micro Devices) is a major semiconductor company designing CPUs, GPUs, and networking silicon for data center and AI workloads.
• AMI — AMI (American Megatrends International) is a firmware company whose products span the full data center infrastructure stack: UEFI firmware (Aptio), baseboard management controllers (MegaRAC BMC), and platform security (Tektagon).
• ASE — Advanced Semiconductor Engineering (ASE Group) is the world's largest provider of independent semiconductor packaging and testing services, with over 100,000 employees and operations spanning multiple countries.
• ASHRAE — ASHRAE (American Society of Heating, Refrigerating and Air-Conditioning Engineers) is a global professional society that develops standards, guidelines, and research for HVAC, refrigeration, and building systems.
• AWS — Amazon Web Services; developed the proprietary SRD (Scalable Reliable Datagram) protocol as a custom alternative to RoCEv2 for large-scale cloud networking.
• AbbVie — AbbVie is a large pharmaceutical company that has built one of the most structured AI business development functions in the industry.
• Abzena — CDMO specializing in complex biologics and advanced modalities, where Campbell Bunce serves as Chief Scientific Officer.
• Accelink Technologies — Chinese optical transceiver manufacturer providing DSP, LPO, LRO, and CPO components for AI-scale data center interconnects.
• Accelsius — Accelsius manufactures QTQD (quick thermal quick disconnect) connectors for pumped two-phase (P2P) direct liquid cooling TCS systems.
• Aivres — AI server vendor offering OCP-compliant full-stack AI infrastructure solutions based on NVIDIA GPUs, including liquid-cooled rack systems, OAM module servers, and PCIe card GPU servers.
• Aldevron — Aldevron is a Danaher company specializing in nucleic acid and plasmid DNA manufacturing.
• Alessandro — Gene therapy pioneer at Ospedale San Raffaele in Milan who led development of treatments for ADA-SCID and metachromatic leukodystrophy from academic discovery through regulatory approval and commercialization.
• Alfa Laval — Alfa Laval is a Swedish company specializing in heat transfer, fluid handling, and separation technologies.
• Alibaba — Alibaba Cloud is a hyperscale cloud provider and one of the largest operators of AI training infrastructure globally.
• Alistair — Partner at Faculty AI helping bioprocessing organizations build workforce capability in AI, data science, and life sciences transformation.
• Alitex — Alitex Data Center Design Services is a data center design consultancy founded approximately 29 years before 2023. The company provides facility design, engineering, and consulting services for data center infrastructure.
• Alnylam Pharmaceuticals — Pioneer siRNA drug company that established GalNAc liver targeting as the dominant delivery mechanism for siRNA, building a portfolio of approved drugs targeting liver-produced proteins after 20+ years of development.
• Amy Walker — CEO of Forbase Bio, UK-based pioneer in cell-free enzymatic DNA manufacturing at GMP grade for cell and gene therapy, advocating scale-out over scale-up for personalized medicine.
• Andrew Heron — CEO and founder of Portal Biotech, biotech entrepreneur with background in Oxford Nanopore next-generation sequencing, now applying molecular biosensing to bioprocessing.
• Ann Lee — Senior executive at Prime Medicine; member of the National Academy of Engineering; contributed to commercialization of 25+ medicines including multiple CAR-T therapies across BMS, Celgene, Juno, Genentech, and Roche.
• Anthony Constantine — EDSFF specification contributor who reviewed the past year's connector, cable and power spec changes and the upcoming liquid-cooling and PCIe 7.0 challenges.
• Anthropic — Anthropic is a frontier AI safety company that develops the Claude family of large language models and Claude Code, an agentic coding tool.
• Aogenics Group — Taiwan-based CDMO and biosimilar developer founded in 2012 with eight biosimilars in pipeline and two on market, deriving over 65% of revenue from Japan and international clients.
• Arc Institute — Arc Institute is a nonprofit research organization pursuing foundational questions at the intersection of biology, computation, and AI.
• Arista Networks — Arista Networks is a network equipment vendor specializing in high-performance Ethernet switches and routers for data center and cloud environments.
• Arm — Arm Holdings is a semiconductor IP company whose processor architectures power a growing share of data center compute, from edge inference to cloud-native servers.
• Arthur — Product manager at cable supplier MOSE who presented a hybrid orthogonal cable-plus-connector solution for EDSFF backplanes.
• Asperitas — Asperitas is a Dutch immersion cooling company that develops and manufactures natural convection immersion cooling systems for data centers.
• Astera Labs — Astera Labs is a purpose-built connectivity IC company designing retimers, switches, and smart cable modules for rack-scale AI infrastructure.
• Astra Networks — Astra Networks is a TU Eindhoven spinout developing photonic-native network fabrics for AI data centers.
• AstraZeneca — AstraZeneca is a global pharmaceutical company with over 100,000 full-time employees.
• Auras Technology — Auras Technology is a Taiwan-based cooling solutions company offering a complete liquid cooling ecosystem from chip-level cold plates to rack manifolds to facility-level CDUs for AI and HPC data centers.
• Austin Labs — Services team of Teledyne LeCroy providing third-party OCP testing, protocol training, and an OCP Experience Center in Austin, Texas.
• Avayla Consulting — Avayla Consulting is a data center standards consultancy led by Dr. Kelley Mullick.
• Avdash Chhodavdia — Astera Labs engineer presenting on PCIe 6 security — end-to-end trust and security considerations for PCIe 6 and switch deployments — at OCP Global Summit 2025.
• Avidity Biosciences — Biotech using full monoclonal antibody conjugated to oligonucleotides via TFR1 targeting for neuromuscular diseases, acquired by Novartis for $12 billion in 2026, validating commercial potential of antibody-oligonucleotide conjugate platforms.
• Axiom Math — Axiom Math is an AI company focused on formal mathematics and formal verification, founded by Carina Hong and based in downtown Palo Alto.
• Baby KJ — Baby KJ is an infant cured of a fatal metabolic disorder by a personalized CRISPR therapy made by Danaher companies.
• Bhavesh Patel — Dell Fellow in the CTO office at Dell Technologies, co-presenting the PowerEdge XC7740 server with Intel Gaudi 3 at OCP Global Summit 2025.
• Bicycle Therapeutics — Clinical-stage biotech using phage display-derived bicyclic peptides as targeting vehicles for radiopharmaceutical payloads, including lead-212 targeted alpha therapy.
• Big Hat Biosciences — Big Hat Biosciences is an AI-driven protein and antibody design company.
• Bill Brennan — President and CEO of Credo Semiconductor, co-developer of ZeroFlap optics for AI cluster networks.
• BioVerative — Rare disease company spun out of Biogen under CEO John Cox; grew to over $1 billion in annual revenue before being acquired by Sanofi; notable for rapidly acquiring True North for $800 million just two months after launch.
• Biocon Biologics — Indian biosimilars company with manufacturing in Bangalore and Johor Bahru managing complex supply chains for biologics including insulin and biosimilar mAbs, pursuing a scale-out strategy and SAP transformation for supply reliability.
• Biogen — Large biopharmaceutical company where John Cox held senior leadership and implemented a unified organizational model consolidating process engineering, technical development, cell culture, manufacturing, and engineering under a single reporting structure with monthly cross-functional meetings.
• Bob Weng — General Manager, Cloud Infrastructure SBU at LITEON Technology, presenter of LITEON GPU power solutions at OCP Global Summit 2025.
• Boyd — Boyd Corporation is a thermal management and engineered materials company that designs and manufactures cooling solutions, sealing systems, and other precision components for demanding thermal environments.
• Brad Rampelberg — Chief Sales Officer for Intelligent Infrastructure at Jabil, long-time OCP participant who sold servers to Facebook starting in 2007 and attended the inaugural OCP Summit in 2011.
• Brent Cheng — Senior manager in the thermal design department at Inventec, presenting on dielectric coolant system design and Intel SuperFluid integration at OCP Global Summit 2025.
• Brian Zonstack — Attendee from Power Rocks who asked questions about the GPU power stabilization bounding box mechanism at OCP Global Summit 2025.
• Brijesh Warrier — VP of Engineering at Microsoft who led the cross-company V-team with NVIDIA and OpenAI on AI training data center power stabilization.
• Bristol-Myers Squibb — US pharma major that announced a $15.2 billion asset-exchange collaboration with Henlius in May 2026, combining immuno-oncology and autoimmunity programs and five joint discovery projects.
• Broadcom — Broadcom is a leading semiconductor company whose switching ASICs (Tomahawk, Jericho, Ramon families) underpin a large share of data center Ethernet networking infrastructure.
• ByteDance — ByteDance is a major internet technology company that operates large-scale AI training infrastructure.
• C.K. Peng — C.K. Peng is Senior Marketing Director at MediaTek, leading the marketing team for MediaTek's data center business unit and presenting the company's ASIC and advanced packaging strategy.
• CBTX — Biotech founded by Hendrik Dietz developing in vivo CAR-T products using immune-quiet single-stranded DNA delivered by targeting-ligand-decorated LNPs.
• CEJN — CEJN is a Swedish manufacturer of high-performance quick connect couplings and fittings for industrial fluid transfer applications.
• CHIPS Alliance — Open-source hardware organization that delivers the canonical OCP lock implementation as part of Caliptra; only its implementation may be called 'lock'.
• CJ Newburn — Data center architect focused on IO who co-presented the GPU-initiated storage and SCADA effort at OCP.
• CS Kim — Chunsung Kim, SVP of eSSD Product Development at SK hynix, who presented the AIN Family at OCP Global Summit 2025.
• Calyos — Calyos is a Belgian cooling technology company developing loop heat pipe (LHP) passive two-phase cooling systems for high-performance computing hardware.
• Campbell Bunce — Chief Scientific Officer of Abzena, 25 years translating complex science into manufacturable therapeutics with a focus on advanced modalities.
• Castrol — Castrol, a BP group company, is a global lubricants and specialty fluids manufacturer that has expanded into data center cooling fluids.
• Celestica — Celestica is a Canadian ODM/OEM that designs and manufactures open networking switch platforms for data center deployments, reporting $12.39B in revenue for 2025. The company is an active contributor to the Open Compute Project, with Jeff Catlin serving as co-chair of the OCP steering committee.
• Cell and Gene Therapy Catapult — UK independent not-for-profit accelerating cell and gene therapy delivery to patients worldwide through early-stage translation, manufacturing development with automation partners, and a hospital network for clinical implementation standards.
• Cellular Origins — Biotech enabling physical, fluidic, and digital connectivity between instruments in cell therapy manufacturing to replace unscalable manual human-mediated steps with an automated platform.
• Charles — Google security expert who presented the OCP L.O.C.K. and TCG MEK-MPA use-case material at the 2025 OCP Global Summit storage session.
• Charles Chen — Representative at MiTAC Computing Technology Corp. presenting AI cluster liquid cooling solutions at OCP Global Summit 2025.
• Charlie Ho — VP of Strategy at Micas Networks, leading CPO product strategy and ecosystem development.
• Chemours — Chemours (The Chemours Company) is a global specialty chemical company and a major manufacturer of fluorinated compounds, including the Opteon and Vertrel fluid families used as two-phase immersion cooling fluids in data center applications.
• Chen (Dr. Chain) — Chief AI Scientist at Hon Hai (Foxconn) Research Institute, lead technical presenter on FoxBrain model training at OCP Global Summit 2025.
• Chong — Senior Product Manager from Delta's Thermal Management Business Group with nine years at Delta, presenting the liquid cooling portfolio at OCP Global Summit 2025.
• Chris Abel — Google storage engineer on the OCP panel, focused on single-firmware convergence, QoS/latency, and OCP L.O.C.K. security.
• Chris Blackburn — Astera Labs engineer presenting on active cabled backplane approaches for next-generation AI infrastructure at OCP Global Summit 2025.
• Chris Butler — Leader running the embedded and critical power business at Flex, co-presenter at OCP Global Summit 2025 on the 2030 solid-state transformer data center architecture.
• Chris Coates — Chris Coates is Principal Solutions Architect at Nscale, responsible for the technical design and deployment architecture of Nscale's liquid-cooled AI data center sites across Europe and North America.
• Chris Collins — Engineering leader at Credo who will co-chair the OCP Optics Reliability Project alongside Oracle's Steve Manley.
• Chris Marriott — Chris Marriott is VP & GM of Enterprise Platforms at NVIDIA, responsible for NVIDIA's enterprise infrastructure product line including GPU-accelerated computing platforms and data center reference architectures.
• Chris Petersen — Astera Labs engineer and UALink Consortium board of directors member presenting on UALink applications, use cases, and production deployment at OCP Global Summit 2025.
• Christian Walter — Managing Director of 9elements GmbH (firmware/cybersecurity), co-founder of Open Source Firmware Foundation.
• Chugai — Japanese pharmaceutical company that co-developed emicizumab (Hemlibra) with Roche, a bispecific antibody replacing factor VIII cofactor function in hemophilia A and B, with $5.4B in annual sales as of 2025.
• Cisco — Cisco Systems is a global networking infrastructure company that designs switches, routers, and silicon for enterprise and hyperscale data centers.
• Coefficient Bio — Coefficient Bio was a clinical development reasoning AI tool co-founded by Samuel Stanton and Joyce Hong.
• Coherent — Coherent (formerly II-VI Incorporated, which merged with Finisar and later rebranded) is a diversified photonics company producing lasers, optical components, transceivers, and optical circuit switches.
• Connor Malone — Connor Malone is VP & GM of Cloud Service Provider (CSP) at Lenovo, responsible for Lenovo's infrastructure business with hyperscale and cloud customers, and a moderator and panelist at OCP industry events.
• CoolIT Systems — CoolIT Systems is a direct liquid cooling (DLC) solutions provider specializing in cold plate technology, CDUs, and rack-scale cooling systems for high-density data centers.
• Credo — Credo is a full-stack connectivity semiconductor company offering products spanning active electrical cables, micro-LED interconnects, and optical transceivers under the ZeroFlap brand.
• Credo Semiconductor — Semiconductor company specializing in high-speed connectivity solutions; developer of ZeroFlap AEC and optical module technology for AI cluster networks.
• Cultivarium — Cultivarium is a nonprofit research institution based in Boston, MA, focused on expanding the microbial diversity available for scientific study and engineering.
• Curtis Stevens — Strategist at Seagate Technology specializing in HDD fault tolerance, drive regeneration, and storage element depopulation technologies.
• Cytiva — Cytiva is a Danaher bioprocessing company spanning monoclonal antibodies, viral vectors, and cell therapy.
• DMTF — DMTF (Distributed Management Task Force) is the industry standards organization that owns and publishes the Redfish hardware management standard (ISO/IEC 30115-1:2022).
• DS Automotion — DS Automotion is an Austrian company specializing in automated guided vehicle (AGV) systems for industrial environments.
• Dan Bracewell — UCL academic and bioprocessing researcher focused on next-generation talent development and industry-connected education.
• Dan Helmick — Engineer at Samsung working on NVMe QoS standardization (TP4176) for shared SSD performance management.
• Dan McGrath — Head of engineering and technical services at Sarepta Therapeutics, with over three decades of tech transfer and advanced modality manufacturing experience.
• Danaher — Danaher is a science and technology conglomerate whose operating companies span the bioprocessing value chain.
• Dane Wittrup — MIT bioengineering professor who co-authored with Adimab a landmark study of 137 clinical-stage antibodies demonstrating that cheap in vitro developability assays reliably predict clinical attrition.
• Danfoss — Danfoss is a Danish industrial technology company known for its expertise in heating, cooling, drives, and motion technologies — including heat exchanger components and fluid control systems.
• Daniel Nishbal — Director of Research at SemiAnalysis focusing on networking and total cost of ownership across the full AI infrastructure stack.
• Dave Landsman — Dave Landsman is a Western Digital representative who manages the OCP data center SAS specification feedback process.
• David Bach — Lead author of the OCP Data Center NVMe SSD specification at Dell who praised Micron's pioneering OCP telemetry implementation and clarified the spec's human-readable output definition.
• David Black — Dell Technologies storage technologist on the OCP storage spec panel, advocating an industry baseline and generalized event injection.
• David Maltz — Technical Fellow and CVP at Microsoft; long-time SONiC-on-routing collaborator with Cisco and on-stage partner at Cisco's OCP Global Summit 2025 session on scale-across networking.
• DeepSeek — Chinese AI lab whose open-source model and published distillation methodology demonstrated that smaller organizations can build capable LLMs without massive data-center training infrastructure.
• Dell Technologies — Dell Technologies is a global server and infrastructure OEM that designs and manufactures integrated rack-scale systems for AI data centers.
• Dell'Oro Group — Market research firm cited for projecting global data center capex will exceed $1 trillion by approximately 2028, driven by AI infrastructure investment.
• Delta Electronics — Global leader in power and cooling solutions for over five decades, co-developing 800 VDC power distribution and liquid cooling products for AI data centers with NVIDIA, Meta, Microsoft, and Google.
• Denvr Dataworks — Denvr Dataworks is a cloud service provider specializing in AI infrastructure built on single-phase immersion cooling deployed in modular container form factors.
• Dine Therapeutics — Late-stage rare disease biotech led by CEO John Cox; underwent a two-year organizational transformation from early-stage to launch-ready and is considering continuous manufacturing as a strategic cannonball investment.
• Divya Ramakrishnan — Sr. Director of Hardware Engineering at NVIDIA who presented the GPU Min Power Floor (MPF) feature deployed on Blackwell products for AI training data center power stabilization.
• Dow Chemical — Dow Chemical contributed PG25 (propylene glycol 25%) coolant guidelines to the OCP Cold Plate Sub-Project and provided an ASTM-based propylene glycol compatibility testing methodology for direct liquid cooling systems.
• Dyne Therapeutics — Clinical-stage biotech founded 2019 developing targeted nucleic acid medicines using the FORCE platform (Fab-TFR1 conjugate) for neuromuscular diseases; pursuing first BLA approval for DMD exon 51 in early 2027.
• Eaton — Eaton is a diversified power management company whose data center solutions span medium-voltage solid-state transformers (MVSST), UPS systems, DC power conversion (sidecar architecture), and facility management software (Brightlayer DCIM).
• Eddie Ruan — Director of ALiNOS at Alibaba Group; co-announced the QuantumCat switch built on Cisco Silicon One P200 at OCP Global Summit 2025.
• Edison Scientific — Edison Scientific is a for-profit company spun out of Future House, a nonprofit research organization, in November 2025. Co-founded by Andrew White, the company is headquartered in the Dog Patch neighborhood of San Francisco, where it operates a wet lab.
• Edwin Stone — CEO of Cellular Origins, focused on physical, fluidic, and digital instrument integration to replace unscalable human intermediaries in cell therapy manufacturing, advocating a capital-efficient scale strategy gated on clinical proof.
• Envicool — Chinese liquid cooling solutions company founded in 2005 with 5,000+ employees, HQ in Shenzhen, R&D in Fremont CA, and 2 GW+ of deployed liquid cooling products.
• Equinix — Equinix is a global colocation data center operator with more than 250 data centers across 70+ countries, providing interconnection and co-location services to enterprises, cloud providers, and network operators.
• Eridu — Eridu is a fabless semiconductor startup developing an extreme-radix switch ASIC targeting AI cluster networking.
• Esha Choukse — Principal Researcher at Microsoft who developed and presented the Firefly software mitigation for AI training data center power stabilization.
• Estel — Estel is a Belgian-headquartered startup and the only European manufacturer of pluggable optical transceivers as of April 2026.
• Eureka Labs — Eureka Labs is an AI education company founded by Andrej Karpathy after his departures from OpenAI and Tesla.
• ExxonMobil — ExxonMobil is a major integrated oil and gas company that has extended its specialty chemicals and lubricants expertise into the data center immersion cooling fluids market.
• FDA — US regulatory agency that mandates prospective process performance qualification for gene therapy manufacturing components including mRNA and guide RNA; released a new draft guidance on gene therapy and gene editing on June 10, 2026.
• FUCHS — FUCHS (Fuchs Petrolub SE) is a German specialty lubricants and fluid technology company that has entered the data center immersion cooling fluids market.
• Facebook — Hyperscaler that originated Project Freedom and subsequently the Open Compute Project in 2011, establishing the template for cost-optimized, workload-specific open-source data center hardware.
• Faculty — UK AI and data science consultancy helping life sciences organizations build AI capability, with a team of 40 hiring talent across the UK.
• Flex — Global electronics manufacturing services company offering the complete data center vertical stack: utility power pods, switchgear, busway, BBUs, PSUs, IT rack hardware, rack integration, and liquid cooling CDUs.
• Fondazione Telethon — Italian not-for-profit charity that co-founded TIGET with San Raffaele Hospital, funded rare disease gene therapy research, and became the first charity to achieve full marketing authorization for gene therapy products in both the US and EU.
• Forbase Bio — UK biotech producing enzymatic (cell-free) GMP-grade DNA to displace plasmid DNA as a critical starting material in cell and gene therapy, vaccines, and gene editing.
• Formerica Optoelectronics — Formerica Optoelectronics is a Taiwan-based optical transceiver manufacturer that leads the OCP Immersion Component Validation workstream.
• Fortum — Fortum is a Finnish energy company and major operator of district heating networks in the Nordic region.
• Foxconn — Hon Hai Precision Industry (Foxconn) — the world's largest electronics manufacturer, now pursuing an AI platform strategy with FoxBrain foundation models powering Smart Manufacturing, Smart EV, and Smart City platforms.
• Frank Frankovsky — The public face and founder of the Open Compute Project (OCP), hired by Jonathan Hiliger at Facebook; previously at Dell where he recruited Brad Rampelberg.
• Frigel — Frigel is an Italian thermal management company and OCP Bronze member that specializes in industrial cooling systems.
• Furukawa Electric — Furukawa Electric is a Japanese optical and electrical company advancing VCSEL-based co-packaged optics for energy-efficient AI datacenter interconnects.
• GSK — Pharmaceutical company that entered a landmark 2010 alliance with the San Raffaele/Telethon institute to co-develop gene therapies across five rare diseases, contributing GMP, regulatory filing, and commercial expertise.
• Ganesh Srinivasan — Ganesh Srinivasan is VP of Product Management for AI Cloud & Enterprise at TE Connectivity, leading the CPC, OTB, backplane, bus bar, and liquid-cooled I/O product lines for high-density AI data centers.
• Genentech — Biopharmaceutical company where Ann Lee held a senior role; planned a major CCP2 capacity expansion with 8 × 25,000-liter bioreactors at Vacaville — the largest scale available at the time — before Lee drove a titer improvement campaign that reduced the need for that capacity.
• Generalist — Generalist is a general physical intelligence and robotics company co-founded by Pete Florence, who serves as CEO.
• Georg Fischer — Georg Fischer (GF) is a Swiss industrial group founded in 1802, with its GF Flow Solutions division (approximately CHF 1.1B revenue) specializing in polymer and metal piping systems for high-purity industrial applications.
• George Tchaparian — CEO of the Open Compute Project (OCP), publicly praised AMI's firmware contributions at the 2025 OCP Global Summit.
• Ginkgo Bioworks — Ginkgo Bioworks is a Boston-based company that builds autonomous lab infrastructure and offers both cloud lab services and R&D services to the life sciences industry.
• Goodfire — Goodfire is a mechanistic interpretability company applying interpretability methods to biology AI models.
• Google — Google is one of the world's largest hyperscale data center operators and a founding member of the Open Compute Project community.
• Gordian Biotechnology — Gordian Biotechnology is a longevity biotech company co-founded by Martin Jensen (CSO).
• Green Revolution Cooling — Green Revolution Cooling (GRC) is an immersion cooling tank manufacturer known for its ICEraQ series of single-phase immersion systems.
• Grok — AI company whose large language model (Grok 3) was trained on the world's largest liquid-cooled AI supercluster, deployed by Supermicro in August 2024.
• Grundfos — Grundfos is a Danish multinational pump manufacturer and one of the world's largest pump companies.
• HPE — Hewlett Packard Enterprise (HPE) is an OCP member company, co-creator of the DC-MHS modular server specification, and co-leader of the Open Platform Firmware work group, focused on enterprise system management, multi-architecture support, and server modularity.
• Hal Loket — Hal Loket is a Principal R&D DVL Engineer at TE Connectivity focused on OCP ORv3 power delivery, DC-MHS specifications, and next-generation rack power architecture standards.
• Hancock Chang — Director of Solution Enabling at MiTAC Computing, open-source firmware ecosystem advocate.
• Handy Dao — Business development lead at Multilane, specializing in production test equipment for CPO and LPO technologies in hyperscaler data center deployments.
• Hardeep Singh — Thermal engineer at Solidigm who demonstrated liquid cooling solutions for SSDs including cold plate integration and cooling energy analysis.
• Harry Petty — Representative from NVIDIA's Accelerated Computing Group presenting AI infrastructure trends and ROI economics at OCP Global Summit 2025.
• Hasan Siraj — Hasan Siraj is Head of Software Products and Ecosystem at Broadcom, responsible for networking silicon product strategy and ecosystem partnerships, including co-presenting Broadcom's Tomahawk 6 launch and ESUN initiative at OCP 2025.
• Hendrik Dietz — Founder/CEO of CBTX, former academic researcher pioneering in vivo CAR-T therapy using immune-quiet single-stranded DNA delivered by targeted LNPs.
• Henlius Biopharmaceutical — 16-year-old Hong Kong-listed China biopharmaceutical company that pioneered biosimilars in China, earned the first EMA approval from China, adopted single-use technology early, and is executing a dual-engine biosimilar-and-innovation strategy.
• Hyper-tech — Hyper-tech is an OCP community member that participated in the first seismic certification of a liquid immersion cooling system in collaboration with METAseismic, Sub-mere, and Wyoming Hyperscale.
• Hyve Solutions — ODM and systems integrator focused on hyperscale infrastructure; presented on next-generation AI switch architecture at OCP Global Summit 2025 and is a participating member in OCP's OSAI initiative.
• IBET — Portuguese bioprocess research institute (Instituto de Biologia Experimental e Tecnológica) pioneering perfusion-based and continuous integrated manufacturing processes for viral vectors and cell therapies since the early 1990s.
• IBM — IBM is a multinational technology and consulting corporation with a significant research arm, IBM Research, operating labs across the world including Europe (Zurich, Rüschlikon, Switzerland).
• ICFO — ICFO is a Barcelona-area research institute coordinating the PIXEurope pilot line, a EU Chips Joint Undertaking worth 400 million euros that spans 11 countries and 20 partners to bridge photonic integrated circuits from lab to fab.
• IDC — IDC (International Data Corporation) is a global market research and advisory firm whose 2024 Worldwide AI Infrastructure Forecast projected global AI infrastructure spending growing from ~$42B (2023) to ~$90B by 2028.
• IDT — IDT (Integrated DNA Technologies) is a Danaher company producing nucleic acid components.
• IEEE — Standards body developing IEEE 802.3dj for 224G per lane signaling on Ethernet optical interfaces.
• IOWN Global Forum — Global industry consortium driving full-stack, end-to-end infrastructure innovation using optical communication and device technologies, with 168 members as of July 2025.
• Iceotope — Precision immersion cooling company that uses environmentally safe dielectric mineral oil to cool entire IT stacks, enabling up to 950+ kW per rack.
• Impleon — Impleon is a data center cooling technology company focused on bio-based coolant solutions for European AI data centers.
• Impon — Impon is a French technology company co-founded by David Garon and Franck (surname not disclosed in available materials), focused on developing "IT boiler" solutions that repurpose server waste heat for productive use in public and commercial facilities.
• Industrial Filtration Limited — Industrial Filtration Limited presented on filtration challenges for liquid-cooled data center systems at the 2024 OCP Global Summit.
• Intel — Intel Corporation is a semiconductor company that designs and manufactures CPUs, networking silicon, and accelerator products for data center workloads.
• Inventec — Taiwan-based ODM founded in 1975 with USD 20.1B annual revenue, 30,000+ employees, producing 200M+ devices annually including cloud servers, PCs, and smart devices; expanding into Physical AI edge cluster solutions.
• Jabil — US-founded contract manufacturer and electronics manufacturing services company (founded 1966, Michigan) with $29.8B FY25 revenue, 140,000+ employees in 25+ countries; Intelligent Infrastructure segment ships 750,000+ L10 servers and 25,000+ L11 racks annually.
• James Zo — Research fellow at Aivres with nearly 30 years of experience in server design and manufacturing.
• Javier Gonzalez — Samsung storage engineer and Linux contributor who has championed NVMe data-placement technologies including FDP.
• Jean-Marie Verdun — Jean-Marie Verdun is a Senior Distinguished Technologist at HPE, co-creator of the DC-MHS specification, and co-leader of the OCP Open Platform Firmware work group, focused on modular server architecture and system-level firmware management.
• Jeff — Google security expert who fielded deep-dive Q&A on OCP L.O.C.K. key management, keeper.io support, and security boundary layers at the 2025 OCP Global Summit.
• Jen-Yao Chung — VP of Business Development at Inventec, presenting on workload-centric edge AI cluster architecture at OCP Global Summit 2025.
• Jensen Huang — Jensen Huang is President and CEO of NVIDIA, the dominant GPU and AI accelerator company. He appears across multiple industry events as a key voice on AI infrastructure trends, scaling laws, and the future of computing.
• Jesse Koh — Hardware Systems Engineer at Meta who led the hardware implementation of Meta's first-generation QLC flash server and the Jabil OEM partnership.
• Jet Cool — Liquid cooling CDU supplier whose unit was featured in Flex's 1 MW rack demonstration at OCP Global Summit 2025.
• Jiajia — ScaleFlux engineer presenting resource grant and isolation designs for FDP overprovision management.
• Jimmy Lee — CEO of WuXi XDC, bioprocess engineer turned ADC CDMO leader who built an integrated four-component manufacturing platform compressing DNA-to-IND timelines by half.
• John Cox — CEO of Dine Therapeutics; former CEO of BioVerative and senior leader at Biogen; expert on biotech organizational design and supply chain decision-making.
• John Kaufman — AstraZeneca platform development and digital transformation lead for drug substance development, champion of FAIR data, AI interoperability standards, continuous processing, and 80/20 automation philosophy
• John Maraganore — Founder of Alnylam Pharmaceuticals who sustained the company through 20+ years of extreme valuation cycles to achieve the first approved siRNA drugs before reaching cash-flow break-even profitability.
• John Matulsky — Independent biologics consultant advising on capex strategy and manufacturing due diligence, with prior CMO leadership at Lonza (US and Singapore scale-up), Sanofi, Catalent, and Cytiva (formerly GE Life Sciences).
• Johnson Jia — Staff Growth Product Manager at Supermicro, AI cluster specialist who led deployment of the world's largest liquid-cooled supercluster (Grok 3) in August 2024.
• Jonathan Balance — CEO of Iceotope, presenting on precision immersion cooling for the entire IT stack including storage at OCP Global Summit 2025.
• Jonathan Hiliger — VP of Technical Operations and CTO at Facebook who initiated Project Freedom (the precursor to OCP), advocating cost-optimized, workload-specific hardware over Swiss-army-knife OEM servers.
• Jonathan Hughes — SSD architect at Solidigm advocating multiple reclaim group FDP for hyperscale storage.
• Juniper Networks — Juniper Networks, now part of HPE (acquired 2024), is an AI-native networking vendor whose QFX switches, PTX routers, and operational platforms (Apstra, Mist/Marvis AI) serve AI data center fabrics.
• Keith D Underwood — Senior Distinguished Technologist at HPE; presented UEC's 'Escaping the Chaos of Order' talk at OCP Global Summit 2025.
• Kevin Brandt — Systems Architect at Micron Technology who led Micron's OCP NVMe SSD telemetry implementation and contributed code to the nvme-cli OCP plugin.
• Keysight — Keysight Technologies is a test and measurement company providing traffic generation and network analysis solutions for data center and AI infrastructure validation.
• Kioxia — NAND flash and SSD manufacturer that built the open-source U-Frog RocksDB plugin to demonstrate FDP benefits.
• Kishore Atreya — Senior Director, Cloud Platform at Marvell, specializing in co-packaged optics deployment strategy for AI scale-up infrastructure.
• Kite Pharma — Gilead subsidiary and CAR-T cell therapy manufacturer; its Amsterdam commercial facility producing 4,000 doses/year with ~1,000 staff is cited as evidence that current autologous cell therapy manufacturing is too labor-intensive to be a viable industry.
• LC — Founder and Chairman of Aogenics Group, who returned to Taiwan in 2012 to establish a biosimilars-focused CDMO drawing explicit parallels between Taiwan biotech and the TSMC semiconductor trajectory.
• LITEON Technology — Taiwan-based power electronics manufacturer and key supplier of GPU power solutions for AI data centers, including embedded PSUs, power shelves, and next-generation 800 VDC power racks.
• LPO MSA — Multi-Source Agreement body for Linear Pluggable Optics standardizing interoperable LPO specifications at 112G and 200G per lane.
• Lattice Semiconductor — Lattice Semiconductor — mentioned as an FPGA-based security solution provider in AMI's cross-vendor modular AI infrastructure diagram.
• Lee Prewitt — Microsoft storage architect who frames the hyperscaler live-migration problem driving NVMe virtual-SSD standardization.
• Lee Prudhomme — Lee Prudhomme is a co-lead of the OCP storage project alongside Ross Stenfort.
• Lenovo — Lenovo is a multinational technology company and the world's largest PC manufacturer, with a significant data center infrastructure division offering AI-ready servers, storage, and networking under the ThinkSystem and Neptune brands.
• LightCounting — Market research firm providing optical transceiver and photonics market forecasts widely cited in the industry.
• Linux Foundation — Linux Foundation — hosts the OpenBMC open-source project that AMI contributes to and builds its management firmware upon.
• Lonza — Swiss CDMO that evolved from a chemical company into a biologics manufacturing powerhouse, cited as a model of CMO partnership maturity and for its capacity-sharing joint venture with Sanofi related to Regeneron products.
• Lumentum — Lumentum is an optical and photonic products company that serves as an incumbent vendor of MEMS-based optical circuit switches and a laser supply partner in NVIDIA's Spectrum-6 co-packaged optics ecosystem.
• Lundbeck — Danish neurology-focused pharmaceutical company serving 10 million daily patients, transitioning from pure small-molecule CNS development to biotherapeutics including bispecifics, blood-brain barrier shuttles, and oligonucleotides via full CMO outsourcing.
• Luxshare-Tech — Global connector and cable backplane manufacturer with 10+ years of OEM experience and 300 million differential pair shipments, presenting the Intrepid NEXUS 224G+ backplane system at OCP 2025.
• METAseismic — METAseismic is a seismic isolation technology company founded by Dr. Noemi Bonessio, originating from the University of California seismic research laboratory.
• MIT — Research institution whose 'GenAI Divide: State of AI in Business 2025' report found 95% of generative AI pilots at companies are failing.
• MLCommons — AI benchmarking organization that publishes MLPerf industry standard benchmarks, including storage performance benchmarks showing direct linear correlation between storage bandwidth and GPU utilization.
• MOSE — Cable and connector supplier offering a mass-production hybrid orthogonal cable solution for EDSFF backplanes across E1, E2 and E3.
• Madhavan Ravi — Hardware Systems Engineer at Meta who co-designed and deployed Meta's first QLC flash storage server.
• Manuel Corando — VP at IBET (Portugal), bioprocess engineering pioneer who developed the first perfusion system for retroviruses (published 2002) and continuous integrated gene therapy processes published 2010-2015.
• Mark Orthodox — Senior Director at AMD's Data Center GPU business unit, leading OEM and ODM partnerships globally.
• Marvell — Marvell Technology is a semiconductor company that designs switching ASICs, optical interconnect DSPs, and data infrastructure silicon.
• Masahisa Kawashima — Technology Working Group Chair at IOWN Global Forum, presenter at OCP Global Summit 2025 on open optical acceleration and OCP-IOWN collaboration.
• Matt Bromage — Matt Bromage is an ARM engineer who presented on Cortex-R82 adoption for NVMe SoC designs, with prior roles at Western Digital, Seagate, and Sandforce.
• Matthew Dy — CEO of Cell and Gene Therapy Catapult (UK), challenging the 'process is the product' axiom and building rare disease manufacturing platforms through aggregated scale and prior knowledge reuse.
• McKinsey — McKinsey & Company is a global management consulting firm whose data center power consumption forecast was cited at OCP 2025, projecting US data centers will account for more than 12% of national electricity by 2030.
• MediaTek — MediaTek is a Taiwanese fabless semiconductor company offering ASIC design services to major cloud service providers (CSPs), with a portfolio spanning advanced process nodes (2nm/A16), packaging (2.5D/3.5D CoWoS), SerDes IP (224G/448G), custom memory interfaces, and Co-Packaged Optics.
• Medra — Medra is a Physical AI scientist company founded by Michelle Lee.
• Merck KGaA — German pharmaceutical and life sciences company where Michelangelo Canzaneri led digital transformation, deploying digital twins and predictive maintenance across CMC and manufacturing plants.
• Meta — Meta Platforms is a hyperscale data center operator and one of the founding forces behind the Open Compute Project, which originated from Meta's (then Facebook's) decision to open-source its data center designs in 2011. Meta operates a global network of data centers supporting its social media platforms and increasingly large AI training and inference workloads.
• MiTAC Computing — OEM server/datacenter hardware vendor (MiTAC Computing Technology Corp.) actively contributing to OpenBMC and OCP Open Platform Firmware.
• MiTAC Computing Technology — MiTAC Computing Technology Corp. is an OEM and ODM providing end-to-end AI rack design, integration, and manufacturing with 3,000 racks/month capacity and global footprint.
• Micas Networks — Micas Networks is a CPO (Co-Packaged Optics) switch manufacturer that has achieved hyperscaler-proven deployment of liquid-cooled switches with integrated optics.
• Michael Lane — VP of Networking at Hyve Solutions, presenting on next-generation switch architecture for AI hyperscale infrastructure at OCP Global Summit 2025.
• Michelangelo Canzaneri — Bioprocess engineer with 12 years at Sanofi and 6 years at Merck KGaA, leading digital transformation in pharmaceutical CMC and manufacturing operations.
• Micron — Micron Technology is a US memory and SSD manufacturer — a key HBM/custom-memory partner in MediaTek's ASIC supply chain and an adopter of OCP NVMe SSD telemetry standards and GPU-initiated storage.
• Microsoft — Microsoft operates one of the largest hyperscale cloud infrastructures in the world through its Azure platform, running data centers across dozens of regions globally.
• Mike Allison — Mike Allison is the OCP storage project's steering committee representative who runs the data center NVMe SSD spec CLA feedback process.
• Mike Hendricks — Associate VP of Solutions and Ecosystem at Astera Labs, co-presenting on UALink and the Astera Labs–Intel partnership at OCP Global Summit 2025.
• Mike Henney — Chief Product and Supply Chain Officer at Bicycle Therapeutics, leading the company's entry into radiopharmaceutical manufacturing with lead-212 alpha therapy.
• Minh Hoang — Storage engineer at Meta driving the open-sourcing of SSD qualification tests via the OCP Diag AutoVal framework on GitHub.
• Moderna — Biotechnology company whose mRNA personalized cancer vaccine (PCV) sequences each patient's tumor to identify up to 34 unique neoantigens and manufactures a custom construct in six weeks, with ASCO 2026 data showing 49% reduction in recurrence/death with pembrolizumab.
• Multilane — Test equipment company specializing in production testing for CPO and LPO optical technologies in hyperscaler data center deployments.
• Murata — Murata is a Japanese electronics manufacturer known for passive electronic components and power supply solutions.
• Myberk Ozerk — Microsoft storage reliability engineer on the OCP panel, detailing fleet telemetry, panic recovery, and ML-driven failure prediction and self-healing.
• NIST — National Institute of Standards and Technology, currently reviewing and recommending post-quantum cryptographic methods to replace quantum-vulnerable encryption standards like RSA.
• NOETIK — NOETIK is a biotechnology company based in South San Francisco that builds multimodal foundation models of human biology with a primary focus on oncology.
• NTT — NTT (Nippon Telegraph and Telephone) is a Japanese telecommunications company whose NTT Devices division is developing Co-Packaged Optics (CPO) switch technology and contributing to the SONiC ecosystem.
• NVIDIA — NVIDIA is the dominant designer of GPUs and accelerated computing platforms for AI training and inference.
• NVMe (NVM Express) — Standards body developing TP4176, the Quality of Service specification for PCIe bandwidth and IOPS rate limiting per SSD controller; still in active discussion as of October 2025.
• NXP Semiconductors — NXP Semiconductors is a Dutch-founded edge semiconductor company with deep roots in automotive, industrial, and infrastructure markets, now repositioning as a physical AI infrastructure company.
• Nan Zhang — EVP of Henlius (Hurray Group) overseeing IND organization for 16 years, architect of landmark licensing deals with BMS and GSK and advocate of China's sustainable innovation thesis.
• Nathan Tracy — Nathan Tracy is President of the OIF (Optical Internetworking Forum) and a DVL Engineer at TE Connectivity, bridging standards work and product engineering for high-speed interconnect.
• Net Zero Innovation Hub — Net Zero Innovation Hub is a clean energy innovation organization co-issuing a joint RFI with OCP to identify future clean standby power technologies as replacements for diesel generators in data centers.
• NewLimit — NewLimit is a startup focused on epigenetic reprogramming to treat age-related diseases, co-founded by Jacob Kimmel approximately four years before UNLOCK 2026.
• Nick Krosinski — VP of Products and Services at Austin Labs (the services arm of Teledyne LeCroy) and a panelist at the OCP SSD test panel.
• Ning Liu — Product Manager for liquid cooling solutions at Envicool, presenter at the 2025 OCP Global Summit.
• Novartis — Pharmaceutical leader in radiopharmaceuticals via Pluvicto for prostate cancer, having invested tens of billions in supply chain infrastructure and providing the only existing playbook for global radiopharmaceutical delivery.
• Novo Nordisk — Danish pharmaceutical company developing two next-generation hemophilia treatments: a high-affinity anti-TFPI antibody (daily dosing) and Mim8, a bispecific bridging factor IX and X projected at 20-50 mg once-monthly pen injector dosing and BLA submission within 5 years of Phase 1 start.
• Nscale — Nscale is a global hyperscaler focused on generative AI compute infrastructure, operating GPU cloud capacity in Europe and North America using modular, reference-architecture-based liquid-cooled data center sites.
• OCP — The Open Compute Project (OCP) is an industry collaborative founded in 2011 to share data center hardware and infrastructure designs as open specifications.
• OCP CL Lab — OCP testing lab that ran early Sprandom validation on a 64TB drive.
• ODCA — The Open Direct Current Alliance (ODCA) is an approximately 90-member industry consortium registered in Germany under ZVEI (German Electrical and Electronic Manufacturers' Association) that promotes DC power distribution in industrial facilities and data centers.
• OIF — Optical Internetworking Forum, a productive optical transport standards organization whose work IOWN Global Forum acknowledges while arguing existing optical standards bodies remain layer-bound and cannot fully unlock optical transport potential.
• OVHcloud — OVHcloud leads the system-level test workstream for aluminum heat exchanger integration in the OCP Door HX sub-project, collaborating with Valeo and Meta.
• OakGate Technologies — Provider of the desktop and rack-mount platforms on which Austin Labs runs the OCP compliance test suite.
• Omar Baldonado — Senior Director of Network Infrastructure at Meta, responsible for AI and data center networking strategy; co-founded the ESUN (Ethernet for Scale-Up Networking) initiative within OCP at the 2025 Global Summit.
• OmniOn Power — OmniOn Power is a power supply company with over 50 years of heritage in telecom power systems, having entered the data center power market approximately 9–12 months before the 2026 OCP EMEA Summit.
• Open Source Firmware Foundation — Independent non-profit foundation (co-founded by 9elements and Siemens) providing governance, funding, and standardization for open-source firmware projects.
• OpenAI — OpenAI is a frontier AI research lab and company that develops large language models (GPT series), the ChatGPT consumer product, and the Codex agentic coding platform.
• Oracle Cloud Infrastructure — Oracle Cloud Infrastructure is Oracle's hyperscale cloud infrastructure division operating large-scale GPU clusters for AI workloads.
• Orchard Therapeutics — Biotech company that continued GSK's rare disease gene therapy programs before discontinuing commercialization of the ADA-SCID therapy for commercial reasons, prompting Fondazione Telethon to take over the license.
• Oriole Networks — Oriole Networks is a London-based startup developing the PRISM photonic network, which uses a passive glass photonic core to achieve nanosecond-scale optical packet switching for AI data centers.
• PCI-SIG — Standards body governing PCIe specifications, including PCIe 6 (64 GT/s), which Astera Labs has production-deployed in data centers following cloud-scale interop testing.
• Parker Hannifin — Parker Hannifin is a global industrial motion and control technology company that manufactures precision-engineered solutions for a wide range of mobile, industrial, and aerospace markets.
• Partha Ranganathan — Partha Ranganathan is a VP and Fellow at Google, responsible for data center systems architecture and a keynote speaker at OCP Global Summit 2025 presenting Google's strategy for the open data center ecosystem.
• Paul Caylor — Hewlett Packard Enterprise storage architect on the OCP panel, leading post-quantum security, SATA-replacement NVMe, and power-efficiency requirements.
• Pegatron — Taiwan-headquartered ODM/OEM manufacturer (est. 2008, ~100,000 employees, $40.3B 2023 revenue, Fortune Global 500 #375) offering end-to-end AI server, rack, and liquid-cooling integration services.
• Perplexity — Perplexity is an AI company best known for its AI-powered search product, and increasingly for Perplexity Computer — an AI operating system launched in February 2026 that creates a team of up to 20 AI agents, orchestrates across models, tools, and files, and operates as a model-agnostic agentic harness.
• Perstorp — Swedish specialty chemicals company, wholly-owned subsidiary of Petronas Chemicals Group, with 140+ years of history; offers PFAS-free biodegradable synthetic ester cooling fluids (SYNMERSE portfolio) for data center direct-to-chip and immersion cooling.
• Peter Troy — Manager of the drive platform team at Pure Storage, leading its high-capacity QLC drive form factor effort.
• Petronas Chemicals Group — Malaysian integrated chemicals provider (PCG) and parent company of Perstorp, operating under the broader Petronas Group.
• Pfizer — Global biopharmaceutical company whose bioprocess R&D spans multiple acquisition legacy systems, exemplifying the industry challenge of data heterogeneity and AI readiness gaps at large scale
• Pharmaceutical Supply Chain Initiative — Industry initiative with 85 member pharma companies (~70% of global pharma turnover, ~$1.2T aggregate) building responsible supply chains through common audit frameworks, decarbonization, and human rights programs.
• Phil Sharp — Nobel Prize-winning MIT scientist, co-founder of Biogen and founder of Alnylam, who predicted in 2008-2009 that an explosion of therapeutic modalities would define the third revolution of biotech.
• Portal Biotech — London-based biotech founded by Andrew Heron, applying Oxford Nanopore sequencing lessons to develop accessible molecular biosensing tools for biopharma and bioprocessing.
• Pranav Garg — Pranav Garg is VP of Strategy, Business Development, and Marketing at TE Connectivity, responsible for AI data center interconnect market positioning and the five-trend thesis shaping TE's AI rack portfolio.
• Pravin Patel — Pravin Patel is Distinguished Engineer and Chief System Architect at Lenovo, responsible for the architecture of Lenovo's server and infrastructure platforms including the Neptune liquid cooling portfolio.
• Prime Medicine — Gene editing company founded on the prime editing platform; cleared its first IND for a CD34 prime edited cell therapy in 2024 and subsequently pivoted the pipeline after achieving proof-of-concept clinical data in two patients.
• Promersion — Promersion is a data center cooling analytics firm led by CEO Rolf Brink.
• Pure Storage — Storage vendor supplying Direct Flash Modules (DFMs) and a user-space FTL driver for Meta's first QLC storage server deployment.
• QCT — QCT (Quanta Cloud Technology), a subsidiary of Quanta Computer, is a data center solution provider offering servers, storage, switches, and integrated rack solutions for hyperscale and enterprise AI deployments.
• Qualcomm — Qualcomm is a US semiconductor company best known for its Snapdragon system-on-chip (SoC) family, which powers the majority of premium Android smartphones and a rapidly growing share of Windows PCs, connected cars, wearables, and industrial devices.
• Quside — Quside is a spin-off from the ICFO research institute that commercializes photonic quantum random number generation technology for post-quantum cryptography applications.
• RISE — RISE (Research Institutes of Sweden) is a Swedish state-owned research institute that provides applied research, testing, and certification services.
• Radical AI — Radical AI is a materials science company that combines closed-loop AI with robotics to accelerate materials discovery.
• Rainer Blair — Rainer Blair is President and CEO of Danaher, leading its bioprocessing strategy.
• Raj Kapoor — Fellow and Chief Firmware Architect at AMD, co-chair of OCP Open Platform Firmware project.
• Rakesh Chopra — SVP and Fellow, Hardware Architecture at Cisco; lead architect of the Silicon One family and presenter of Cisco's scale-across networking strategy at OCP Global Summit 2025.
• Rakuten Mobile — Rakuten Mobile is a Japanese mobile network operator with approximately 10 million subscribers, operating an end-to-end cloud-native network since 2020. Rakuten Mobile is a pioneer in open networking adoption in the telecom sector, having built its network on virtualized infrastructure from inception.
• Ralph Peeper — R&D Director of the Custom Design Business Unit (CDBU) within Delta's Power and Systems business group, presenting 800 VDC power solutions at OCP Global Summit 2025.
• Ranga Godavarti — Head of Bioprocess R&D at Pfizer with 25+ years spanning cell line development, process development, and clinical manufacturing of biologics and vaccines
• Rebellions — South Korean AI chip startup designing chiplet-based AI accelerators; products include the ION (2022), ATOM (2023), REBEL-Q (2024, 2 PFLOPS FP8 on Samsung 4nm), and REBEL-IO (2025), targeting LLM inference workloads.
• Rick Chai — Rick Chai is Vice Chairman and CEO of MediaTek, the Taiwanese fabless semiconductor company. He was shown wearing Meta AR glasses at OCP Global Summit 2025 to demonstrate AI-driven spatial computing becoming commercially real.
• Rick Walsh — Senior VP at Sanblaze, a protocol-level NVMe test-tool vendor, and a panelist at the OCP SSD test panel.
• Rob Campbell — VP/GM running the IT hardware, integration, and liquid cooling business at Flex, co-presenter at OCP Global Summit 2025 on power-compute convergence for AI data centers.
• Rob Williamson — Senior director of sustainable procurement at AstraZeneca and board member/past chair of PSCI, driving supplier science-based target commitments and clean heat programs.
• Robbie Xu — Director of Product Management and Marketing at Luxshare-Tech, presenter of the Intrepid NEXUS 224G+ backplane solution at OCP Global Summit 2025.
• Robert Barrentine — Sustainability manager at Saint-Gobain Life Sciences, expert on LCA, mass balance approach, and polymer circularity in biopharma.
• Roche — Swiss multinational pharmaceutical company re-entering the obesity and cardiovascular space with major GLP-1 manufacturing investment including a new North Carolina facility with digital twin and automation capabilities.
• Ronda Duffy — EVP and Chief Operations Officer at Biocon Biologics, responsible for biosimilar supply reliability across manufacturing sites in Bangalore and Johor Bahru with a focus on inventory strategy and logistics.
• Rory Bolt — Senior Fellow at Kioxia who develops FDP use cases and the U-Frog RocksDB plugin.
• Ross Stenfort — Ross Stenfort is a Meta storage engineer and co-lead of the OCP storage project.
• Royal — TCT engineer who presented on water-based coolant technology for ICT liquid cooling systems at OCP Global Summit 2025, concluding from a five-year China pilot that pure water coolant is the best choice for MLCP deployments.
• SK hynix — South Korean semiconductor company developing the AIN Family of AI-optimized NAND storage products targeting inference and large-scale AI workloads.
• SNIA — Storage Networking Industry Association — standards body currently reviewing Solidigm's draft specification for liquid-cooled SSD mechanical and thermal interfaces.
• SONiC Foundation — The SONiC Foundation is a Linux Foundation project governing the open-source SONiC (Software for Open Networking in the Cloud) network operating system, originating from Google and spanning dozens of ecosystem partners.
• STMicroelectronics — STMicroelectronics (ST) is a Franco-Italian semiconductor manufacturer operating foundry and fab capacity across Europe.
• Sachin Katti — Intel executive who delivered the keynote at OCP Global Summit 2025, announcing the Crescent Island next-generation GPU.
• SafeWay Custom Fluid Transfer — SafeWay Custom Fluid Transfer (a brand of The Specialty Manufacturing Company) is a US manufacturer of fluid transfer components and quick-disconnect valve systems.
• Saint-Gobain Life Sciences — Manufacturer of single-use polymer and plastic critical fluid path components for biopharma, pursuing LCA, mass balance feedstocks, and polymer circularity.
• SambaNova — SambaNova Systems is an AI chip company that designs the Reconfigurable Dataflow Unit (RDU), a hardware accelerator purpose-built for AI inference with a fundamentally different architecture from GPU-based approaches.
• Samsung — Samsung is a South Korean conglomerate with major semiconductor divisions producing DRAM, NAND, and HBM memory; listed as a key memory ecosystem partner in MediaTek's ASIC supply chain.
• Samsung Foundry — Samsung's semiconductor foundry division offering advanced process nodes from legacy to 2nm GAA, HBM memory, advanced packaging (2.5D/2.3D/3D), and the SAFE ecosystem of 100+ partners — positioning itself as a one-stop-shop for AI chiplet designs.
• San Raffaele — San Raffaele is a research partner collaborating with Danaher companies on lentiviral vector production.
• SanDisk — Storage vendor that developed Sprandom and contributed it to the open-source FIO benchmarking tool.
• Sanblaze — Protocol-level NVMe test-tool vendor offering the 'Certified by Sanblaze' suite, scaled racks, and desktop units for early SSD/controller validation.
• Sanjoy Maity — CEO of AMI (American Megatrends International), presenting on open-source firmware for rack-scale AI infrastructure at OCP Global Summit 2025.
• Sanofi — Global biopharmaceutical company (formerly including Bayer Vaccines) operating a pilot lights-out synthetic lab and advancing continuous downstream processing, with patient affordability as a stated development obligation
• Sarepta Therapeutics — Gene therapy company where Dan McGrath heads engineering and technical services, applying decades of tech transfer expertise to advanced modality manufacturing.
• Saurabh Kulkarni — VP of AI Technical Product Management at Intel, leading Intel's data center AI product portfolio and presenting the heterogeneous hardware strategy for agentic AI at OCP Global Summit 2025.
• ScaleFlux — ScaleFlux is a fabless semiconductor company producing NVMe and CXL SoCs, transitioning from ARM A53 to Cortex-R82 for its Gen 5/6 storage controllers.
• Schneider Electric — Power and energy management company cited as the source for data-center cooling power overhead data: air-cooled facilities spend ~37% of total power on HVAC vs ~2% for liquid-cooled facilities.
• Scott Shadley — Solidigm presenter known for immersion cooling demonstrations, referenced as speaking at OCP Global Summit 2025.
• Seagate — Seagate Technology is a leading hard disk drive (HDD) manufacturer and storage systems company.
• Sean Lim — Sean Lim is a Manager at Auras Technology who presented on AI infrastructure and advanced thermal solutions at OCP Global Summit 2025, arguing that cooling is a design constraint — not merely an operational consideration — at every layer from chip to facility.
• SemiAnalysis — Technology research and analysis firm covering the full AI infrastructure stack from systems to chip design, with a focus on networking and TCO.
• Shell — Shell is a global energy and chemicals company that has entered the data center cooling fluids market with both single-phase immersion cooling fluids and direct-to-chip cooling coolants.
• Shubha Mitra Basu — Representative at Perstorp specializing in synthetic ester-based thermal management fluids for data center liquid cooling, presenter at OCP Global Summit 2025.
• Siemens — Industrial technology company, co-founder of the Open Source Firmware Foundation alongside 9elements.
• Silka Maul — VP and Head of Global MSAT, Engineering and Sustainability at Roche, bridging technical development and manufacturing with focus on GLP-1 capacity expansion and manufacturability assessments.
• Solidigm — SSD manufacturer focused on data center storage, developing liquid-cooled SSD solutions and industry standards for thermal management.
• Southwest Research Institute — Southwest Research Institute (SwRI) is an independent, nonprofit applied research and development organization based in San Antonio, Texas, with approximately 3,000 employees.
• Stephen Manley — Senior Principal Network Engineer at Oracle Cloud Infrastructure (OCI), technical lead for RDMA fabric operations and co-developer of ZeroFlap optics with Credo.
• Steve Bates — UK life sciences policy leader shaping sector competitiveness, investment, and long-term industrial strategy for the government.
• Steven Sprouse — SanDisk engineer who developed Sprandom, a fast pseudo-random SSD preconditioning method.
• Submer — Submer is an immersion cooling tank manufacturer and technology company that develops single-phase immersion systems for data center applications.
• Sumitomo Electric — Sumitomo Electric (SEI) is a Japanese conglomerate with a major fiber optics division operating as Sumitomo Electric Lightwave.
• Super Micro — Super Micro Computer (Supermicro) is a US server and IT infrastructure manufacturer that designs and manufactures high-performance server platforms, including GPU servers, storage servers, and high-density blade systems.
• Sustainable Markets Initiative — CEO-level cross-industry collaboration for sustainable supply chains, set up by King Charles with Health Systems Task Force chaired by AstraZeneca CEO Pascal Soriot, publishing joint supplier targets.
• Suzanne Farid — Head of UCL biochemical engineering department and professor of bioprocess systems engineering, researching sustainability metrics, LCA tools, and green bioprocess curricula.
• TCT (Thermal Control Technology) — Chinese thermal management company specializing in water-based coolant systems and medium purification modules (MPM) for ICT liquid cooling infrastructure, advocating pure water coolant over formulated alternatives.
• TE Connectivity — TE Connectivity is a global industrial technology company specializing in connectors, sensors, and electronic components for data center power and cooling infrastructure.
• TSMC — TSMC is the world's largest contract chip manufacturer, fabricating processors for virtually every major AI and semiconductor company including NVIDIA, Arm, Qualcomm, and Marvell on leading-edge 3nm and 2nm process nodes.
• Taeksang Song — Corporate VP of DRAM Solution Engineering at Samsung Semiconductor, presented Samsung's full next-generation memory portfolio at OCP Global Summit 2025.
• Tar Samad — Head of Global Research and Corporate Patents at Lundbeck, leading the company's expansion from small-molecule neurology into biotherapeutics including bispecifics, blood-brain barrier shuttles, and oligonucleotides.
• Tareq Bustami — Tareq Bustami is VP of Market Technology at Celestica, responsible for defining next-generation data center networking platforms and co-presenting Celestica's AI fabric strategy at industry events.
• Teledyne LeCroy — Protocol-analyzer company (PCIe/NVMe analysis, error injection, oscilloscopes) whose services arm is Austin Labs.
• Tencent — Tencent is a Chinese technology conglomerate and hyperscaler operating large-scale AI training infrastructure.
• Terra Therapeutics — Terra Therapeutics is an AI-first small molecule biotech founded approximately 7.5 years before UNLOCK 2026. The company's core platform, Emmy (Experimentation Meets Machine Intelligence), is built on 16 billion precise target-ligand binding measurements growing at roughly 1 billion data points per quarter.
• Texas Instruments — Texas Instruments (TI) is a major semiconductor company specializing in analog and embedded processing chips.
• Thomas Mermans — Business strategy and business development manager at Samsung Foundry, co-presenter at OCP Global Summit 2025 on the AI CPU chiplet platform and Samsung's one-stop-shop foundry value proposition.
• Thomas Saur — Sanofi global CMC lead for microbial biologics and genomic medicine with 29 years in biologics development, advocate for process simplification, continuous downstream processing, and patient affordability as a core obligation
• Tiger Ninomiya — Optical engineer at Accelink Technologies presenting on AI-era transceiver technologies at OCP Global Summit 2025.
• Tim Amundsen — Tim Amundsen is Senior Business Development Director at ScaleFlux, focused on the company's NVMe SoC line of business.
• Trusted Computing Group — The Trusted Computing Group (TCG) is an industry consortium that publishes hardware security specifications; it recently published the host interface specification for OCP Lock, extending Caliptra's open-source hardware security root of trust.
• UALink Consortium — Industry standards body developing UALink, the open scale-up interconnect for AI accelerators; Intel and Astera Labs are promoter (board-level) members actively driving the specification.
• UCL — University College London, a leading UK research university with bioprocessing education programs including the 20-year industry-facing MBI program.
• UNLOCK 2026 — UNLOCK 2026 was the first annual UNLOCK conference, held on April 22, 2026, in San Francisco.
• Ultra Ethernet Consortium — The Ultra Ethernet Consortium (UEC) is an industry consortium standardizing Ethernet enhancements specifically for AI backend networks, addressing congestion signaling, link-layer recovery, and flow control beyond PFC limitations.
• Unicom Engineering — Unicom Engineering is an IT hardware services company specializing in server modifications and deployment support for immersion cooling environments.
• University of Illinois — Academic institution where the GPU-initiated storage research originated under Wen-mei Hwu.
• Upscale AI — Upscale AI is a networking startup founded in Q3 2025 building purpose-built switching infrastructure for AI data centers.
• Uptime Institute — Uptime Institute is a global data center advisory organization known for its Tier Standard (Tier I through Tier IV) for data center infrastructure.
• Valeo — Valeo is a EUR 20.9B automotive thermal systems supplier that leads the aluminum heat exchanger corrosion testing workstream in the OCP Door HX sub-project.
• Vertiv — Vertiv is a global infrastructure technology company providing power, thermal management, and IT infrastructure solutions for data centers.
• Victaulic — Victaulic is a global leader in mechanical pipe joining systems, specializing in grooved-end pipe connections, flow control, and liquid distribution infrastructure.
• Vikram Mailthody — Researcher from the University of Illinois, student of Wen-mei Hwu, driving the research side of the GPU-initiated storage project.
• Vineet Parekh — Storage engineer at Meta driving OCP NVMe Latency Monitoring spec adoption and fleet-wide SSD debugging capability at hyperscale.
• Way Huang — President of Henlius Biopharmaceutical who led the company from biosimilar pioneer to global biologics innovator with products approved in 70+ countries.
• Wen-mei Hwu — University of Illinois professor and research advisor whose student Vikram Mailthody drove the GPU-initiated storage research.
• Wesco — Wesco International is a distributor and integrator of electrical and industrial products, including cooling and data center infrastructure solutions.
• Western Digital — Western Digital (WDC) is a storage vendor whose Dave Landsman manages the OCP data center SAS specification feedback.
• Wewin — Wewin is a hardware partner in the development of Active Air Liquid Cooling (AALC) sidecar cooling systems with AMD.
• Wieland — Wieland is an engineering and manufacturing company that contributes technical leadership to OCP immersion cooling standardization efforts.
• Wilo SE — Wilo SE is a German pump manufacturer and OCP Platinum sponsor that produces high-efficiency circulation pumps for industrial and data center fluid management applications.
• Wistron — Taiwanese ODM and parent company of Wiwynn, co-developing GB200 NVL72-based rack systems and co-exhibiting at OCP Global Summit 2025.
• Wiwynn — Wiwynn Corporation is a Taiwan-based cloud infrastructure server and storage solutions provider that designs and manufactures high-density rack systems, power shelves, and cooling solutions for hyperscale data centers.
• Wonguk Kim — Hardware tech team leader at Rebellions, co-presenter at OCP Global Summit 2025 on chiplet-based AI accelerator design strategy.
• WuXi XDC — China-based ADC and bioconjugate CDMO founded approximately five years ago with integrated four-component manufacturing at a single site, expanding into Singapore and the US to address geopolitical supply chain risk.
• Wyoming Hyperscale — Wyoming Hyperscale is a data center company and OCP community member that participated in the first seismic certification of a liquid immersion cooling system in collaboration with METAseismic, Hyper-tech, and Sub-mere.
• X-Cell — X-Cell is Xaira Therapeutics' virtual cell model — a diffusion language model trained on perturbative causal single-cell data.
• Xaira Therapeutics — Xaira Therapeutics is an AI drug discovery company that builds therapeutic programs and AI models simultaneously.
• YTL — Malaysian conglomerate and emerging sovereign neocloud operator that partnered with Wiwynn and Nvidia to deploy a Level 12 AI cluster architecture, representing the growing class of infrastructure-owning AI cloud operators.
• Yong Lee — Director of AI Research Center at Hon Hai (Foxconn) Research Institute, co-presenter of FoxBrain at OCP Global Summit 2025.
• Yuliya Tarnikova — Engineer at Google presenting multi-tenant SSD QoS requirements and use cases at OCP Global Summit 2025.
• ZutaCore — ZutaCore is an OCP Startup Member specializing in two-phase direct-to-chip liquid cooling technology.
• eBay — eBay is a global e-commerce company that operates large-scale network infrastructure.
• iMason — iMason is a data center industry association that partnered with OCP and other industry players to develop the embodied carbon disclosure specification for hardware vendors.
• iPronics — iPronics is a Barcelona-based silicon photonics startup founded after ten years of research, specializing in programmable optical circuit switches (OCS) for AI datacenter scale-up networking.
• imec — imec is Belgium's leading nanoelectronics and digital technology research institute, operating a 300 mm silicon photonics pilot line and conducting advanced research into 3D assembly, hybrid bonding, and wafer-scale integration.

Concepts (434)

• 224G PAM4 Interconnect — The current generation high-speed copper interconnect standard for AI scale-up rack interconnects, operating at 224 Gbps per lane using PAM4 modulation, with bump-to-bump channel budgets constrained to ~24 dB at 53 GHz.
• 800 VDC Power Architecture — Next-generation data center power distribution using 800 V DC bus between power rack and IT racks, reducing current distribution losses at megawatt-scale GPU infrastructure.
• 800 VDC Power Distribution — High-voltage DC power distribution architecture for AI data centers co-developed by Delta, NVIDIA, Meta, Microsoft, and Google to replace 50 V systems for rack loads exceeding 250 kW.
• 800V DC Sidecar Architecture — The 800V DC Sidecar Architecture is a data center power distribution approach that centralizes AC-to-DC power conversion in a dedicated "sidecar" rack, distributing high-voltage DC (+/-400V, totaling 800V) to adjacent IT racks.
• AAV Gene Therapy — Adeno-associated virus gene therapy delivering factor VIII or IX genes to liver cells via IV infusion; one product approved for hemophilia B achieves annualized bleeding rate of 0.4/year but faces gradual expression decline and low commercial uptake (~$90M 2025 sales) from a cautious patient community.
• ADA-SCID — Adenosine deaminase deficiency causing severe combined immunodeficiency; treated by the first hematopoietic stem cell gene therapy approved globally, with 50 patients treated, all alive, and durable immune reconstitution confirmed at 10-15+ years.
• ADC Manufacturing — Multi-component biomanufacturing discipline for antibody-drug conjugates requiring integrated antibody, linker, payload, conjugation, and drug product capabilities with complex scheduling, high-containment facility demands, and escalating molecular complexity.
• ADP Parameter Files — Parameter (ADP) files that let OCP tests match a drive's firmware-reported test points, required for fully compliant results.
• AI Agents for Scientific Discovery — AI agents for scientific discovery are AI systems that autonomously execute multi-step scientific workflows — including literature search, hypothesis generation, experimental design, and data analysis — with minimal human intervention per step.
• AI Cluster Deployment — End-to-end service covering design, installation, and validation of a complete AI compute cluster in a customer data center, spanning hardware architecture, network topology planning, on-site integration, and LLM workload validation.
• AI Cluster Networking — The set of bandwidth, latency, telemetry, security, and topology requirements specific to interconnecting GPU clusters for AI training and inference workloads, fundamentally different from traditional data center networking.
• AI Cluster Scaling — The empirical trend that training state-of-the-art AI models requires exponentially more GPUs and compute every few years; from ~100 GPUs for GPT-2 (2019) to 50,000–100,000 GPUs for GPT-5 (2025), a 5,000x increase in GPU count with only 2.5x improvement in MMLU reasoning performance.
• AI Data Center Networking — The network fabric architecture — scale-up, scale-out, and scale-across — connecting GPUs and accelerators in large AI training and inference clusters.
• AI Data Center Power Density — The rapid escalation in per-rack and per-chip power consumption driven by AI accelerators, with leading GPUs exceeding 1,000W TDP and racks trending from 20–30 kW toward over 1 MW within a few years.
• AI Data Center Power Scaling — The rapid escalation of data center power requirements from tens of megawatts to gigawatt scale, driven by AI workload density, identified as the top infrastructure challenge in 2025.
• AI Fabrics — AI Fabrics are the specialized network interconnect layers that tie GPU/XPU accelerators together within and across racks, requiring fundamentally different design principles from traditional data center networking to maintain 80%+ accelerator utilization.
• AI Inference Architecture — AI Inference Architecture encompasses the hardware, software, and system-level design choices specific to deploying large language models for production inference, distinct from training infrastructure.
• AI Inference Memory Hierarchy — The layered memory and storage architecture for LLM inference — combining HBM, DRAM, and AI-optimized NAND — co-optimized to maximize tokens per second per watt.
• AI Inferencing Infrastructure — The compute layer for serving pre-trained AI models to end users, distinct from training clusters; built on PCIe-based GPU/NPU servers and the primary profit-generating phase of an enterprise AI deployment.
• AI Infrastructure — AI Infrastructure is the full system stack supporting AI training and inference workloads, comprising GPU/ASIC compute engines, high-speed DDR memory, SSD storage, high-speed networking, power conversion, and thermal management; global spending projected by IDC to grow from ~$42B (2023) to ~$90B by 2028.
• AI Model Scaling — The trend of exponentially increasing AI model parameter counts and GPU cluster sizes, growing at 14x per year, driving infrastructure requirements from 20K to 100K to 1M GPU clusters.
• AI Network Telemetry — AI Network Telemetry encompasses the monitoring and diagnostic approaches specifically designed for AI training cluster fabrics, where traditional telemetry methods (SNMP polling, coarse counters, In-band Network Telemetry) have proven fundamentally inadequate.
• AI Networking Bottleneck — The AI networking bottleneck describes the structural mismatch between GPU compute scaling and switching infrastructure scaling in large AI training clusters.
• AI Open Rack — Astera Labs' open rack architecture for AI Infrastructure 2.0, featuring a Compute Platform (any CPU/GPU/XPU), Scale-Up Switch Platform (Scorpio PCIe/UALink switches), and Active Cabled Backplane, collaboratively built with 12+ ecosystem partners.
• AI Rack Design Differentiation — Engineering differences between ODM implementations of the same NVIDIA reference design — in power architecture, hose materials, cable management, and manufacturing scale — that determine rack reliability and TCO at the $3M GB200 price point.
• AI Scaling Laws — Three distinct scaling laws—pre-training, post-training, and test-time (long thinking)—describing how applying more compute yields higher AI model capability, collectively driving 100x compute demand growth.
• AI Sustainability and Carbon Emissions — The carbon and energy footprint of large-scale AI inference and training; rapid AI growth drove a 150% rise in tech-sector CO2 emissions from 2020–2023, and GPT-5 queries consume ~18 Wh each — 8.6× more than GPT-4.
• AI Traffic Engineering — AI Traffic Engineering encompasses the techniques for optimizing network traffic distribution and congestion management specifically for AI/ML training workloads.
• AI Training Data Center Power Stabilization — Full-stack engineering challenge of preventing GPU collective power oscillations (ranging from near-idle to gigawatt scale) from destabilizing utility grids, addressed via hardware, software, and energy storage mitigations.
• AI Workload Power Dynamics — AI Workload Power Dynamics describes the distinctive electrical load behavior of AI training and inference clusters, which differs fundamentally from traditional compute, storage, and networking workloads.
• AI and Machine Learning in Bioprocessing — Application of AI/ML across the bioprocessing lifecycle from discovery to commercial manufacturing, enabling faster development timelines, data-driven decision making, and cross-organizational knowledge synthesis
• AI-Driven Infrastructure Management — AI-driven infrastructure management applies machine learning to data center operations — fault prediction and detection, energy and environmental monitoring, analysis and recommendation, and self-healing remediation — managing complexity that manual processes cannot scale to address.
• AI-Enabled Bioprocessing — The application of artificial intelligence and advanced analytics to bioprocessing development, manufacturing, and decision-making, transforming workforce requirements and scientific practice.
• AIN Family — SK hynix's three-product family of AI-optimized NAND solutions — AIN P (ultra-high IOPS), AIN B (high bandwidth flash), AIN D (high capacity/low cost) — designed to address inference bottlenecks in the AI data pipeline.
• AMD Pensando Pollara 400 — AMD's shipping 400 Gbps AI NIC based on P4 programmable architecture, supporting RoCEv2, Ultra Ethernet features, and multi-plane networking for both front-end and back-end AI cluster connectivity.
• AMD Pensando Vulcano — AMD's next-generation 800 Gbps AI NIC on 3nm process, supporting UAL and PCIe Gen6, delivering 8x scale-out bandwidth improvement over Pollara 400 per GPU, announced for 2026.
• ARM Cortex-R82 — ARM Cortex-R82 is ARM's latest 64-bit real-time CPU core for Gen 5/6/7 NVMe controllers, delivering an 81% perf gain over R5, 1 TB addressing, CHI coherency, and an optional ML engine.
• ARM Server Architecture — ARM-based server CPU architecture (including Ampere and NVIDIA Grace/Vera options) under active evaluation by hyperscalers and cloud customers as an alternative to x86 (Intel/AMD) for data center workloads.
• ASHRAE A2 Data Center Cooling Standard — ASHRAE A2 specifies a maximum inlet air temperature of 35°C for data center equipment — a threshold that air-cooled SSDs above ~25W TDP cannot meet without active cooling modifications.
• Academic-Industry Partnerships — Collaborative frameworks between academic research institutions and pharmaceutical or biotech companies for translating scientific discoveries into commercial gene therapies, requiring clear shared goals, strong governance, transparency, and mutual flexibility.
• Access Conditions — Abstract condition objects added by MEK-MPA that an entity validates before enabling data access; they gate read/write lock changes and encapsulate the secrets used to enable access.
• Active Cabled Backplane — Backplane design using active copper or optical cables to boost signal integrity and extend reach for scale-up interconnects within and across racks, enabling larger GPU scale-up pod configurations than passive backplane approaches.
• Adaptive Routing For AI — Adaptive Routing for AI refers to the traffic engineering techniques that dynamically distribute AI training traffic across available fabric paths to maximize bandwidth utilization and maintain performance under link failures.
• Advanced Packaging — Multi-die integration technologies (2.5D CoWoS, 3.5D stacking, hybrid packaging) used to exceed reticle size limits and integrate compute, memory, IO, and optical dies in a single package; 3.5D is identified as the successor once 2.5D CoWoS hits its physical boundary at 9–10x reticle size.
• Advanced Therapy Medicinal Products — Regulatory category encompassing gene therapies, cell therapies, and tissue-engineered products; facing development bottlenecks around data reuse and regulatory pathways that EU initiatives like Joint for ATMP are working to address.
• Agentic AI — Agentic AI refers to AI systems that observe, reason, plan, use tools, and take autonomous multi-step actions — as distinct from conversational AI that responds to a single prompt and stops.
• Agentic Engineering — Agentic Engineering is a term coined by Andrej Karpathy to describe the engineering discipline of coordinating AI agents to produce professional-quality software without sacrificing the quality bar.
• Analytical Technology in Cell and Gene Therapy — The persistent analytical technology gap in cell and gene therapy — where product characterization tools lag far behind the biologics space — drives regulatory delays via complete response letters, perpetuates the process-freeze problem, and must be resolved alongside automation for the field to achieve commercial viability.
• Annual Interruption Rate — Azure reliability metric (AIR) tracking customer VM disruptions; its low target is the business reason live migration is required.
• Antibody Developability — Framework of in vitro biophysical assays — selectivity, HIC behavior, insect cell binding, long-term stability — used to predict clinical attrition from discovery; a Wittrup/Adimab study of 137 clinical-stage antibodies demonstrated approved drugs are nearly always flag-free while Phase 2 candidates routinely carry multiple flags.
• Antibody-Drug Conjugates — Antibody-drug conjugates (ADCs) are targeted biologics that place unique demands on manufacturing infrastructure.
• Antisense Oligonucleotides — Single-stranded synthetic nucleic acid analogs used to modulate gene expression by binding mRNA or nuclear RNA; the first oligonucleotide modality with a 40+ year development arc from 1978 discovery to a broad clinical pipeline.
• Arm Neoverse V3 CSS — Arm's Neoverse V3 Compute Subsystem — a pre-validated 64-core server CPU chiplet platform contributed to the AI CPU chiplet collaboration; being hardened on Samsung Foundry 2nm targeting 3.5–4.0 GHz.
• Asset Lifecycle Management — The practice of tracking equipment performance data and proactively scheduling capital reinvestment to avoid situations where aging manufacturing assets collide with increased portfolio capacity demands, preventing forced CDMO reliance during transitions.
• Automation in Cell Therapy Manufacturing — Integration of physical, fluidic, and digital connectivity between instruments in cell therapy workflows to replace unscalable human-mediated steps, enabling the analytical platforms and manufacturing productivity needed for commercial viability.
• Autonomous Laboratory Operations — Self-directed automated laboratory systems capable of executing experimental workflows with minimal human intervention, from high-throughput automated platforms to lights-out facilities; achieved in synthetic chemistry but not yet in biologics as of 2026
• Battery Energy Storage Systems — Battery Energy Storage Systems (BESS) are grid-interactive power buffering technologies being evaluated by Google alongside microgrids to manage AI training workload power fluctuations and enable clean standby power, with standardized APIs under development.
• Bioprocessing — Bioprocessing is the manufacture of biologic therapies, the central focus of the Danaher Summit.
• Biosimilar Development — Development of bioequivalent versions of branded biologics that built foundational CMC and regulatory competency in Chinese and Taiwanese manufacturers, with a major patent-expiry cliff now driving fresh demand.
• Biosimilar Supply Chain Management — Managing supply reliability for biosimilars requires balancing forecast volatility, long CMO lead times, multi-stage inventory positioning, cold-chain logistics, and regulatory release timelines in import markets such as the EU.
• Born-In-Immersion Hardware — Born-In-Immersion Hardware refers to IT equipment designed natively for fluid immersion environments rather than converted from air-cooled designs.
• Build-vs-Buy Manufacturing Strategy — The strategic decision framework for biopharmaceutical companies on whether to build internal manufacturing capacity, outsource to CDMOs, partner, or adopt hybrid models, driven by cost, scale, flexibility, and pipeline modality complexity.
• C9h Strings File — Vendor-provided binary file paired with NVMe SSD telemetry data by the nvme-cli OCP Plugin to produce human-readable debug output per the OCP specification.
• CAR-T Cell Therapy Manufacturing — Autologous CAR-T manufacturing is constrained by 2-7+ week vein-to-vein times, up to 31% patient dropout, inherited 20-30-year-old processes, and ~4 doses per operator per year productivity, creating urgent pressure to transform the manufacturing paradigm.
• CDMO Partnerships — Contract development and manufacturing relationships central to advanced modality programs; frequently fail due to incentive misalignment, insufficient modality-specific expertise, and absence of risk-sharing or transparency.
• CDU (Coolant Distribution Unit) — A Coolant Distribution Unit (CDU) is the facility-level component that manages the interface between the data center's technology cooling system (TCS) loop and the facility water or heat rejection system.
• CLA Process — The CLA process is an OCP process letting non-authors give feedback on draft specs before contribution, at the authors' discretion.
• CXL (Compute Express Link) — Open interconnect standard for memory expansion and pooling built on PCIe; Astera Labs demonstrates CXL memory controllers at OCP 2025 running AI workloads for larger context windows and faster inference responses.
• Cable Backplane and Cable Cartridge — High-density copper interconnect assembly used in AI scale-up racks to connect GPU and switch nodes via a midplane, evolving from 112G to 224G+ data rates.
• Cable Cartridge — A cable cartridge is a modular cable-and-connector assembly serving as the top-to-bottom signal backbone of AI racks, replacing traditional backplane PCBs with active or passive cable bundles that provide higher bandwidth density, improved signal integrity, and serviceability.
• Caliptra — Caliptra is an open-source hardware root-of-trust specification and implementation jointly developed through OCP; version 2.1 adds post-quantum cryptography, OCP Lock key management, and manufacturing debug capabilities.
• Canopy BMC — Cross-vendor OpenBMC distribution focused on upstream-first development, long-term stability, and CI/CD validation across multiple hardware platforms.
• Capacitor Buffer Unit — Power capacitor shelf deployed in data center power systems to absorb high-dynamic GPU load transients, smoothing AC grid demand spikes during AI compute bursts.
• Cell and Gene Therapy — Cell and gene therapies are advanced modalities requiring entirely new manufacturing paradigms.
• Cell and Gene Therapy Manufacturing — CMC (Chemistry, Manufacturing, and Controls) development for cell and gene therapies, requiring standardization and reproducibility for patient-specific living cell products across a wide range of batch sizes and unpredictable patient scheduling.
• Cell-Free Enzymatic DNA Manufacturing — A cell-free alternative to plasmid DNA production using enzymatic synthesis to manufacture GMP-grade DNA starting materials for viral vector and mRNA therapy manufacturing, offering potential safety, speed, and cost advantages over incumbent plasmid fermentation.
• Centers of Excellence for ATMPs — EU-promoted hub model where ATMP research, manufacturing, and clinical expertise are concentrated regionally, enabling rapid technology transfer and knowledge sharing across rare disease programs and countries.
• Certified Device List — Microsoft's approved-device list (CDL); tool vendors build tests to support CDL and tools are approved by Microsoft for CDL testing.
• China Biomanufacturing Innovation — The emergence of China as a biopharmaceutical innovation hub, evidenced by $130 billion in license-out value in 2025 and landmark multinational partnerships, marking a shift from capacity center to global innovation partner.
• Chiplet Architecture — Chiplet Architecture is the design methodology of decomposing a monolithic SoC into smaller disaggregated dies (chiplets) that are co-packaged using advanced interconnects; enables independent optimization of compute, IO, memory, and optical dies across different process nodes.
• Chiplet Design — An approach to semiconductor design composing a package from multiple smaller dies connected by high-speed die-to-die interconnects, enabling modular, cost-effective, and scalable AI chip development that can adapt to rapid AI innovation cycles.
• Circular Economy in Bioprocessing — Closing material loops in biopharma manufacturing to reclaim and reuse every polymer and plastic input; described as the 'last frontier' after decarbonization, requiring industry-wide collaboration to establish the recovery business case.
• Cisco Silicon One P200 — Cisco's 51.2 Tbps full-duplex routing ASIC announced at OCP 2025; delivers router-level capabilities at switch-level power efficiency via a fully shared deep packet buffer and run-to-completion engine.
• Clos Topology — Hierarchical leaf-spine network topology used in traditional and AI data centers; supports ECMP load balancing but is radix-limited without adding tiers, expected to remain dominant through approximately 2027.
• Co Packaged Optics — Co-Packaged Optics (CPO) is a technology that integrates optical transceivers directly into the switch ASIC package, eliminating thousands of pluggable transceiver modules.
• Co-Packaged Copper — Co-Packaged Copper (CPC) is a high-density substrate-based interconnect technology routing 1,024+ differential pairs on a 110×110mm package from chip to backplane, supporting 224G to 448G per lane — one of the densest copper interconnect implementations in the industry.
• Compute Express Link — Compute Express Link (CXL) is a PCIe-based open interconnect standard enabling memory expansion, pooling, and near-data computation; Samsung is mass-producing CMM-D 2.0 (128–256 GB, 36 GB/s) with CXL 3.1/PCIe Gen 6 CMM-D planned for Q4 2025.
• Confidential Computing — Mutual VM/host distrust model that NVMe is addressing with TDISP support and an NVMe interface report so a VM can verify untampered hardware.
• Congestion Signaling — Congestion Signaling (CSIG) is an OCP/UEC network load-balancing mechanism that uses explicit congestion signals to improve traffic distribution in AI cluster fabrics; proven in Google production and targeting standardization in UEC 1.1 with Broadcom support.
• Continuous Bioprocessing — Manufacturing approach where upstream and downstream operations run in a connected, uninterrupted flow rather than discrete batch cycles, enabling real-time feedback control, higher throughput per unit, and cost-efficient scale-out for distributed manufacturing
• Continuous Capability Investment — The principle that workforce capability development in bioprocessing must be an ongoing sustained commitment rather than a one-time training event, matching the continuous pace of technological change.
• Contract Development and Manufacturing Organization — CDMOs provide both manufacturing and process development services to biopharmaceutical innovators; the industry has shifted from treating CDMOs as transactional vendors to strategic partners with shared development capabilities for titer improvement and process optimization.
• Conventional Magnetic Recording — The traditional HDD recording technology where sectors are randomly and equally writable; Seagate is actively researching three approaches to extend depopulation-based fault tolerance to CMR drives.
• Cooling Distribution Unit — Core component of direct-to-chip liquid cooling systems that conditions and distributes coolant to server cold plates, available in liquid-to-liquid and liquid-to-air configurations from 100 kW to 2.5 MW.
• Crescent Island GPU — Intel's next-generation inference-optimized GPU, featuring up to 160 GB LPDDR5x memory and 1.5 TB/s bandwidth on Xe 3P IP, in a PCIe card form factor targeting low-cost inference with H2 2026 sampling.
• Custom Silicon (ASIC) — Application-Specific Integrated Circuits (ASICs) designed to customer specifications for AI accelerators; the custom AI accelerator TAM is projected to grow from $6.1B in 2023 to over $40B in 2028 at a CAGR above 46%.
• DC Automation and Robotics — DC Automation and Robotics encompasses the deployment of autonomous machines -- mobile robots, automated guided vehicles (AGVs), drones, and robotic arms -- to perform physical operations in data center facilities that have traditionally required human labor.
• DC Telemetry — DC Telemetry refers to standardized data center telemetry guidelines being co-developed by Google and multiple hyperscalers to securely monitor mechanical systems (generators, chillers) in real time against rapidly fluctuating AI workloads.
• DCMHS — DCMHS (Data Center Modular Hardware System) — OCP modular server architecture enabling hot-swappable Host Processor Modules (HPM) and Data Center Secure Control Modules (DCSCM) across different vendor processors with dynamically built firmware.
• DPU SmartNIC Integration — DPU/SmartNIC Integration refers to the convergence of networking and compute at the switch level, exemplified by Microsoft's Smart Switch concept.
• DSP-Based Optical Transceivers — Fully retimed pluggable optical transceivers using digital signal processing for equalization — the proven standard for robust multi-vendor AI interconnects.
• Data Center Commissioning — Data Center Commissioning is the process of systematically validating that a data center's cooling, power, and controls infrastructure operates correctly and meets design specifications before production IT hardware is deployed.
• Data Center Energy Consumption — Aggregate electricity and water use by data centers; US data centers consumed 176 TWh in 2023 (4.4% of US electricity), with cooling accounting for ~50% of total energy, and are projected to reach 12% of US electricity by 2028.
• Data Center Migration — The geographic shift of hyperscale AI data centers away from high-cost urban real estate toward cheaper-power inland regions, driven by AI training cluster energy demands; Meta's planned 5GW Hyperion facility exemplifies the scale.
• Data Center NVMe SSD Specification — The OCP data center NVMe SSD specification (version 2.6 in 2024) defines data center NVMe SSDs and drives the marketplace's SSD listings.
• Data Center Power and Busway Infrastructure — Power distribution architecture in AI data halls; 75% of global racks today support under 20 kW, while GB200/300 NVL72 deployments require four busway systems — a configuration rarely available outside purpose-built facilities.
• Data Center SAS Specification — The OCP data center SAS specification, version 1.5, was contributed in 2025 by Microsoft, Dell, HPE, and Meta.
• Data Center Sustainability — Data Center Sustainability refers to the discipline of minimizing the environmental impact of data center infrastructure across its entire lifecycle -- from raw material selection and manufacturing through operational use to decommissioning and disposal.
• Data Object Exchange — Proposed PCIe mailbox mechanism to transfer NVMe SSD telemetry binary data from drives that appear as PCIe but not NVMe devices, enabling OCP telemetry when standard NVMe get-log commands are unavailable.
• Deep-Buffer Networking — Network switching/routing architecture using large shared packet buffers to absorb traffic bursts during congestion and link failures; Cisco argues it is necessary for AI scale-across workloads where distance makes reactive congestion control alone insufficient.
• Deterministic Networking — IOWN Global Forum's packet forwarding architecture using a hub-and-spoke topology over Open APN to deliver deterministic QoS, reducing hop counts and enabling service-level guarantees analogous to parcel tracking.
• Developability — Drug design philosophy that evaluates manufacturability and scale-up potential from the earliest stage through iterative design-make-test-redesign cycles, treating process feasibility as a priority design criterion.
• Digital Bioprocessing — The end-to-end digitization of biopharmaceutical manufacturing using AI, real-time sensors, digital twins, and data infrastructure to reduce failures, accelerate development, and improve product quality.
• Digital Twin — A digital twin is a physically accurate simulation of a physical system; in data centers, digital twins model airflow, thermal dynamics, power delivery, grid interaction, and water table impacts before and during physical deployment.
• Digital Twins — Software models that mirror real bioprocessing operations in real time, enabling predictive process control, experimentation reduction, and learning from both process and product-level data.
• Digital Twins in Bioprocessing — Computational models of bioprocessing unit operations or whole plants used to simulate process behavior, align development and manufacturing constraints before physical experiments, and provide simple real-time deviation alerts in commercial operations
• Direct Flash Module — Pure Storage's U.2-long flash module combining NAND with a user-space FTL, enabling software-controlled performance tuning and cross-drive RAID.
• Direct-to-Chip Cooling — Direct-to-Chip (DTC) Cooling, also called Direct Liquid Cooling (DLC), uses cold plates mounted directly on processor dies to transfer heat to a circulating liquid coolant (typically propylene glycol solutions such as PG25 or bio-based alternatives).
• Direct-to-Chip Liquid Cooling — Liquid cooling technology that removes heat directly from processor packages via cold plates and coolant distribution, mandatory for AI racks above ~120 kW and projected necessary for 1 MW/rack by 2027.
• Disaggregated Scheduled Fabric — Meta's scale-out AI network architecture that disaggregates intra-chassis switching across a full data hall using scheduled traffic management; evolved from 4K GPUs (DSF v1) to 20K GPUs per building (2-stage DSF).
• Discovery-Manufacturing Collaboration — Drug discovery framework that aligns antibody discovery with manufacturing from the outset — shared purification platforms, CHO cell lines with single-site integration, miniaturized manufacturing assays, selectivity panels, and mechanistic PK/PD modeling — to prevent late-stage attrition and deliver patient-friendly subcutaneous formulations.
• Dissolved Oxygen Control — Water quality management technique in pure water coolant systems that inhibits aerobic bacteria proliferation and metal corrosion by reducing dissolved oxygen via oxygen-adsorption resin or membrane degassing.
• Distributed Manufacturing — Strategy using multiple smaller regional plants rather than centralized large-scale facilities, offering supply chain resilience and proximity to patients but challenged by loss of the economies of scale that have traditionally justified large-batch biologics manufacturing
• Distributed Training Parallelism — Combination of data parallelism, tensor parallelism, and pipeline parallelism used to shard large language models across GPU clusters for efficient training.
• Door Heat Exchanger — Door Heat Exchangers (Door HX, also called Rear Door Heat Exchangers or RDHx) are rack-mounted liquid-to-air heat exchangers that capture the hot exhaust air from IT equipment as it exits through the rear of the rack.
• Dragonfly Topology — Emerging data center network topology interconnecting clusters of servers via a circular multi-connected fabric; enables low-latency node-to-node communication and easy horizontal expansion, projected to become required for large-scale AI clusters from ~2027.
• Drive Self-Reported Power — NVMe device self-reported power with interval power measurement (TP 4199 thresholds) enabling sustainability reporting and real-time power management.
• Dual ToR Redundancy — Dual ToR Redundancy is Microsoft's server-level network redundancy approach for Azure data centers, providing continuous connectivity when a Top-of-Rack switch fails.
• Duchenne Muscular Dystrophy — X-linked recessive progressive muscle-wasting disease caused by dystrophin gene mutations with 100% mortality from cardiac or diaphragm failure in late 20s-30s; Dyne Therapeutics' lead asset Dyne-251 targets exon 51 skipping.
• Dupixent — Dupixent is a biologic from the Sanofi-Regeneron collaboration, cited as an early manufacturing-and-partnership bet.
• E1.S Form Factor — The dense, high-performance EDSFF compute-storage form factor projected to exceed 40% of unit volume by 2029 and the focus of EDSFF liquid-cooling standardization.
• E2 Drive Form Factor — EDSFF E2 enterprise SSD form factor with a single-PCB design and superior cooling; Meta's planned platform for next-generation high-density QLC storage servers.
• E2 Form Factor — A newly ratified EDSFF high-capacity SSD form factor (SFF-1042) enabling petabyte-class drives with an 80W connector and thermally optimized single-PCB enclosure.
• E3 Form Factor — An established EDSFF form factor extended in 2025 to be supported by the orthogonal connector and new liquid-cooling spec work.
• ECMP Packet Spraying — A multipathing technique used by Ultra Ethernet to distribute individual packets across all available network paths using Equal-Cost Multi-Path routing with many entropy values, eliminating flow-level load balancing limitations.
• EDSFF — The Enterprise and Datacenter SSD Form Factor family (E1.S, E1.L, E2, E3) standardized through SFF/OCP for datacenter flash storage.
• ELSFP — External Laser Small Form-factor Pluggable (ELSFP); a proposed pluggable laser module form factor for co-packaged optics that keeps the laser field-replaceable outside the compute package.
• Edge AI Cluster — A rack-based, on-premises AI computing cluster deployed at the edge for real-time inference, simulation, and model refinement, combining industrial-grade GPU nodes (IGX), OVX simulation servers, storage, and a Kubernetes-managed control plane.
• Edge Computing — Edge computing brings compute infrastructure closer to data sources — away from centralized cloud data centers — enabling AI inference and data processing at the point of generation; enterprise edge infrastructure is forecast to grow at 19.1% CAGR from 2024 to 2028.
• Edge-Cloud Convergence — IOWN Global Forum's vision for unifying edge and cloud into a single seamless accelerated computing zone via Open APN and deterministic networking, enabling RDMA across data centers and bypassing TCP/IP overhead at the edge.
• Embodied Carbon Disclosure — Embodied Carbon Disclosure is an OCP-standardized reporting framework for hardware embodied carbon, developed with iMason, enabling consistent apples-to-apples comparisons across vendors from component to server to company level.
• Emicizumab — Bispecific antibody (brand Hemlibra) co-developed by Chugai and Roche that bridges factor IX and factor X at micromolar affinity to replace factor VIII cofactor activity; approved for all hemophilia types with and without inhibitors; annualized bleeding rate 2.4; $5.4B in 2025 sales.
• End-to-End IO Prioritization — Single-tenant SSD QoS extension ensuring high-priority IOs are expedited through all internal SSD stages (submission queues, NAND Die, DMA) to achieve ultra-low latency for critical commands.
• Epoch Protection Key — Root key owned by the key management block and shared by all MEKs on a device, derived via a KDF from a firmware-managed soft epoch key and a fuse-held hard epoch key that is never visible to firmware.
• Ethernet For AI Training — Ethernet for AI Training refers to the emerging approach of using standard Ethernet with merchant silicon and open networking stacks (SONiC, SAI) as a viable alternative to NVIDIA InfiniBand for large-scale AI training cluster interconnects.
• Ethernet for Scale-up Networking — Ethernet for Scale-up Networking (ESUN) is an OCP work stream launched at the 2025 Global Summit with 12 founding industry members to standardize open Ethernet as the consensus approach for AI accelerator scale-up interconnects, complementing the Ultra Ethernet Consortium's scale-out work.
• Event Injection — Generalization of air injection in the OCP spec: injecting misbehaviors and specifying drive responses so host code can be validated before production failures.
• Exon Skipping — Therapeutic RNA splicing modulation strategy using ASOs or PMOs to force pre-mRNA splicing to skip mutant exons, enabling production of truncated but partially functional protein such as dystrophin in DMD.
• FAIR Data Principles — Data management framework (Findable, Accessible, Interoperable, Reusable) identified as the foundational prerequisite for AI adoption in bioprocessing; companies achieving FAIR data early gain decisive competitive advantage in process development speed
• FBOSS — FBOSS (Facebook Open Switching System) is Meta's open-source network operating system that manages Ethernet switches via OCP SAI; now extended to support NVIDIA Spectrum-4, Broadcom, and Cisco ASICs at gigawatt scale.
• FIO — Open-source flexible I/O benchmarking tool into which SanDisk integrated Sprandom as a new job type.
• FORCE Platform — Dyne Therapeutics' proprietary Fab-oligonucleotide conjugate delivery technology that targets TFR1 via a hit-and-run mechanism for broad, deep delivery to skeletal muscle, cardiac muscle, and CNS.
• FPGA Pre-Silicon Validation — Running OCP/FTP test suites against an FPGA board emulating an early ASIC to catch issues before tape-out.
• Fabric Grooming — Proactive and reactive process of identifying and removing degraded or suspect optical links from an AI cluster fabric to maintain overall stability at the cost of reduced capacity.
• Federated AI Learning — A distributed machine learning approach where organizations contribute to shared model improvement without exposing proprietary data, proposed as the future model for CDMOs and cross-industry biopharma collaboration.
• Federated Learning — Federated learning is a machine learning approach that enables model training across multiple data-holding institutions without requiring direct transfer of the underlying data.
• Field Serviceability — Design principle requiring that CPO systems support component replacement by data center technicians without significant downtime, covering laser field-replaceable units, simplified tray racking, and blind-mate fiber backplane connections.
• Firefly — Microsoft software mitigation that runs a power-hungry secondary workload during AI training communication phases to flatten GPU power swings to below 5%, using NVIDIA MPS for low-overhead context switching.
• Firefly Clock Synchronization — Firefly is Google's ultra-accurate clock synchronization protocol for data centers, designed to enable precise time distribution across distributed components for ML scheduling and financial-exchange-grade accuracy, with details published in a SIGCOMM paper.
• Fitusiran — siRNA therapeutic (brand Qfitlia) silencing antithrombin expression to rebalance hemostasis in hemophilia; dosed every couple of months but creates a 3-4 month irreversible reduction in antithrombin, raising surgical thrombosis management concerns for a well-informed patient population.
• Flexible Data Placement — NVMe Express feature letting the host place data into separate physical SSD locations to reduce write amplification, ratified in NVMe and adopted by OCP.
• Formulated Coolant — Proprietary chemical mixture coolants for ICT liquid cooling that use additive-based corrosion and microbial inhibition; each vendor's formula is incompatible with others and requires periodic lab sampling rather than real-time monitoring.
• Foundation Models for Biology — Foundation models for biology are large neural networks pre-trained on biological data — genomics, transcriptomics, imaging, patient records, or combinations thereof — and fine-tuned for downstream scientific and clinical tasks.
• FoxBrain — Foxconn's proprietary foundation model family (flagship FoxBrain-70B) trained on manufacturing-domain and synthetic data, underpinning Smart Manufacturing, Smart EV, and Smart City platforms.
• Full-Chain Liquid Cooling — End-to-end liquid cooling strategy covering the complete supply chain from chip provider to data center operator, integrating cold plates, QDs, manifolds, CDUs, piping, coolants, and control systems under unified vendor responsibility.
• Fungible Data Center — A data center architecture paradigm enabling late-binding deployment decisions — such as switching between GPU and TPU within a 30-day window — through standardized power, cooling, mechanical, and telemetry interfaces across first- and third-party facilities.
• GLP-1 — GLP-1s are a peptide therapeutic class proving mature manufacturing can scale to population-level impact.
• GLP-1 Manufacturing Capacity — Rapid scale-up of manufacturing capacity for GLP-1 receptor agonists (obesity and metabolic disease drugs) is driving major capital investment in bioprocessing, with demand forecast volatility and delivery device configuration flexibility as key operational challenges.
• GPU Load Transient Management — Power electronics techniques — bulk capacitor sizing, digital PFC control, and firmware — used to absorb GPU dynamic load fluctuations and prevent transients from propagating to the AC input.
• GPU Min Power Floor — NVIDIA hardware feature (Blackwell GB200+) that constrains GPU power within a programmable band using a bounding box controller and ramp rate controller to prevent grid-destabilizing power swings, achieving 20% swing at data center level.
• GPU Power Shelf — Centralized rack-level power conversion unit that moved out of the server in GB200/GB300 architectures to supply DC power to GPU compute trays at 33–110 kW per shelf.
• GPU-Initiated Storage — A model where the GPU directly initiates fine-grained, massively concurrent storage access to NVMe, independent of the CPU.
• GalNAc Delivery — N-acetylgalactosamine sugar ligand conjugated to oligonucleotides for targeted hepatocyte delivery via ASGPR receptor, enabling the portfolio of approved siRNA and ASO drugs targeting the liver and liver-produced proteins.
• Gate-All-Around GAA Transistor — Next-generation transistor architecture beyond FinFET, used in Samsung Foundry's 2nm and 3nm process nodes; Samsung's 2nm GAA is described as the company's third iteration of the technology.
• Gene Therapy Manufacturing — Manufacturing of gene and cell therapy products, characterized by small patient populations, complex regulatory requirements for prospective process validation, and the challenge of aligning CMC timelines with breakthrough clinical data in ultra-rare disease programs.
• Gene Therapy Platform Technologies — Modular approach to gene therapy development using shared manufacturing infrastructure, analytical methods, and regulatory packages across multiple disease programs, typically changing only the transgene while keeping the vector backbone and process constant.
• Gene Therapy for Rare Diseases — Use of viral vectors to deliver corrective genes into patient cells (especially CD34 hematopoietic stem cells) to treat genetic rare diseases, requiring translational expertise from academic discovery through manufacturing, regulation, and commercialization.
• Gigawatt-Scale Data Centers — Gigawatt-Scale Data Centers refers to the industry transition from data center campuses sized at a few megawatts to deployments measured in hundreds of megawatts and ultimately several gigawatts of IT power.
• Global Data Center GPU Market — The global data center GPU market is projected to grow at a 14.2% CAGR from 2025 to 2034, reaching $183B by 2034, with data center energy consumption projected to reach 945 TWh by 2030.
• HBM High Bandwidth Memory — Stacked DRAM memory standard providing very high memory bandwidth for AI accelerators; Rebellions' REBEL-Q integrates four HBM3E (12H) stacks delivering 4.8 TB/s bandwidth and 144 GB capacity.
• HVDC Power Architecture — HVDC (High Voltage DC) Power Architecture refers to the use of high-voltage direct current distribution -- typically at 800V or +/-400V -- within data center facilities as a replacement for conventional AC power distribution to server racks.
• HVDC Power Distribution — High-voltage DC power delivery architecture for AI data center racks, enabling higher power density and efficiency compared to traditional AC distribution, now considered standard for next-generation AI rack deployments.
• Half-Life Extended Clotting Factors — Modified recombinant factor VIII and IX products with extended plasma half-life achieved via Fc fusion, albumin fusion, or PEGylation, reducing infusion frequency below the 2-3x/week required for unmodified factors; factor IX is more amenable due to simpler clearance biology.
• Heat Reclamation — Heat reclamation is the practice of capturing waste heat from data center cooling systems and redirecting it to commercial or industrial processes; commercially viable when tied to always-on industrial uses rather than seasonal district heating.
• Heat Reuse — Heat Reuse refers to the capture and productive use of waste heat generated by data center IT equipment, typically by raising coolant temperatures high enough (50-70°C) to feed district heating networks, industrial processes, or agricultural applications.
• Hematopoietic Stem Cell Gene Therapy — Ex-vivo gene therapy approach that corrects or adds functional genes into hematopoietic stem cells to treat blood disorders, immunodeficiencies, and neurodegenerative diseases; three products approved as of 2026 with durable benefits lasting 10-15+ years.
• Hemophilia — Inherited bleeding disorder caused by factor VIII deficiency (Hemophilia A) or factor IX deficiency (Hemophilia B), affecting ~10,000 people in the UK; characterized by joint bleeds, CNS bleeds, and historically shortened life expectancy; treated by factor replacement, gene therapy, or hemostasis rebalancing.
• Heterogeneous Hardware — Data center infrastructure strategy combining CPUs, GPUs, and XPUs from multiple vendors across a unified fabric, enabling cost-efficient agentic AI workload placement.
• High Bandwidth Memory — High Bandwidth Memory (HBM) is stacked DRAM technology co-packaged with AI accelerators to provide the extreme memory bandwidth required by large model training and inference; MediaTek is developing Custom HBM with non-standard base dies to exceed JEDEC HBM5 bandwidth limits.
• High Speed Copper Interconnects — High-Speed Copper Interconnects refers to the ecosystem of copper-based cabling and connector technologies carrying data at 224 Gbps and 448 Gbps per lane within and between AI data center racks.
• High-Throughput Experimentation — Miniaturized automated experimental platforms used in bioprocess development to rapidly characterize large design spaces, reducing timelines for purification development and generating large datasets required for model training
• High-Voltage DC Power Architecture — High-Voltage DC (HVDC) power architecture uses ±400V or 800V DC distribution in data centers, reducing transmission losses by 10–20% and saving up to 50% rack space versus 48V DC, and is increasingly adopted for AI rack power densities above the 48V ceiling.
• Human Phage Display — Antibody discovery technology that incorporates human volunteer-derived antibody sequences into bacteriophage particles displaying antibody fragments on their surface; incubation with a target antigen selects binders that are sequenced, generating large initial hit sets screened by activity and biophysical assays.
• Hybrid Orthogonal Connector — A cable-plus-connector EDSFF backplane approach that routes high-speed signals through cable while a cheap FR4 backplane carries sideband, clock and power.
• Hybrid Public Key Encryption — RFC 9180 scheme used by MEK-MPA to encapsulate access-enabling secrets, combining a KEM, a key derivation function, and an encryption algorithm, all declared in an X.509 certificate extension.
• Hybrid Quantum-Classical Computing — Computing paradigm (HQCC) that combines classical HPC infrastructure with quantum processing units to deliver practical quantum advantage today by leveraging each technology for its strengths.
• ICT Liquid Cooling — Liquid-based thermal management systems for information and communications technology infrastructure, including server and switch liquid cooling, where coolant chemistry and quality control are critical operational factors.
• IO Chiplet — Standardized interchangeable chiplets defined by the UALink Consortium that reside within GPUs and accelerators to provide open-standard scale-up I/O without each accelerator vendor building custom transport silicon.
• IOPS-Optimized SSDs — A new SSD class designed for fine-grained IOPS-per-dollar/TCO rather than the legacy capacity-per-dollar optimization.
• IT/OT Convergence — The integration of information technology (IT) systems with operational technology (OT) equipment in biopharma manufacturing, enabling data flow from instruments to cloud infrastructure as the foundation for digital bioprocessing.
• Immersion Cooling — Immersion Cooling submerges IT equipment entirely in a dielectric fluid that absorbs heat directly from all components, eliminating the need for fans, heat sinks, and individual cold plates.
• Immersion Fluid Lifecycle — Immersion Fluid Lifecycle covers the complete management of dielectric cooling fluids from initial fill through end-of-life disposition, including in-service monitoring, maintenance, cleaning, and responsible disposal or recycling.
• Immersion TCO — Immersion TCO (Total Cost of Ownership) modeling quantifies the full lifecycle costs and benefits of immersion cooling versus air-cooled data center deployments.
• In Vivo CAR-T Therapy — Non-viral approach that programs T-cells inside the patient's body using targeted nucleic acid delivery, avoiding the complexity of ex vivo cell manufacturing.
• In-Network Collective Acceleration — In-Network Collective Acceleration (INC) offloads collective communication operations (AllReduce, AllToAll, AllGather, ReduceScatter) from GPU endpoints to network switch silicon, reducing exposed GPU idle time during AI training.
• InfiniBand — High-performance networking interconnect standard; InfiniBand volume one spans 2,000+ pages; its transport specification, standardized in 2000, was the basis for RoCEv2 and the legacy that UEC is modernizing.
• Intel Arc GPU — Intel's GPU family originating in the client space, now attracting data center interest; Intel is developing Xe3P-based data-center-optimized successors targeting inference workloads.
• Intel Gaudi 3 — Intel's current-generation AI accelerator offered in three form factors (OAM card, Universal Baseboard, PCIe CEM), with a rack-scale system supporting 64 OAMs at 76.8 TB/s HBD bandwidth and 8.2 TB HBM2e.
• Intel Xeon — Intel's server CPU with Advanced Matrix Extensions (AMX), positioned as the optimal host node CPU for agentic AI systems capable of handling lightweight AI inference workloads directly on the CPU.
• Interconnect Wall — The interconnect wall is the physical scaling bottleneck where interconnect bandwidth cannot keep pace with compute scaling, making it the primary constraint in AI architecture — more critical than the memory wall.
• Intrepid NEXUS Backplane — Luxshare-Tech's 224G PAM4 high-density backplane connector system with 2.4x2.4mm pitch, 80 differential pairs per square inch, supporting up to 75 GHz and future 448G data rates.
• Jagged Intelligence — Jagged Intelligence describes the uneven capability profile of frontier LLMs — models that exhibit world-class performance in some domains while failing at seemingly trivial tasks in others.
• KOOLIO CPC — Luxshare-Tech's chip-to-PCB connector (CPC) product compatible with the Intrepid NEXUS backplane system, used at the GPU/switch chip side of the 224G+ channel budget path.
• KV Cache — Key-Value cache storing intermediate attention activations during LLM inference decode, a primary memory capacity and bandwidth bottleneck that grows with conversation context length.
• Key Management Block — Caliptra extension that is lock's sole MEK manager, guarding keys with data protection keys, multi-party protection keys, and an epoch protection key, and combining them when loading an MEK into the AES engine.
• LBA Status Log — A standardized HDD feature that returns a categorized list of Logical Block Addresses (LBAs) — accessible, unhealthy/degrading, and inaccessible — enabling hosts to proactively protect data before head failure.
• LLM Inference Acceleration — The targeted use case for the Rebellions/Samsung/Arm/ADTechnology AI CPU chiplet platform — combining high-performance low-power AI accelerators with best-in-class CPU chiplets to serve large language model inference workloads at ~3x power and performance efficiency versus existing solutions.
• LPDDR6 — LPDDR6 is Samsung's next-generation low-power DRAM delivering 10.7 Gbps/pin and 114.1 GB/s aggregate bandwidth with 20% better power efficiency than LPDDR5X, targeting both AI servers and mobile/edge devices.
• Lab-in-the-Loop — Lab-in-the-loop describes the iterative cycle of algorithm prediction → wet-lab experiment → training data contribution → improved algorithm.
• Latency Monitoring — OCP latency monitoring functionality present since spec 2.0, critical for debugging complex latency at scale; 2.7 makes its logs human-readable.
• Lead-212 — Short-lived alpha-emitting isotope used in targeted alpha therapy; sourced primarily from a single aging DOE reactor at Oak Ridge, creating supply concentration risk that Bicycle Therapeutics addressed by leasing reprocessed uranium.
• Leak Detection and Management — Leak detection and management for liquid-cooled AI racks uses water leak detection cables, water trays, and RMC software to trigger automated IPMI-based server shutdowns when coolant leakage thresholds are exceeded.
• Lenovo Neptune — Lenovo Neptune is Lenovo's proprietary direct water cooling platform for AI servers, now in its 6th generation, achieving 80% GPU+CPU cooling efficiency using 45°C distilled water with no chillers or chemical additives.
• Lentiviral Vector Production — Lentiviral vector production is a major manufacturing bottleneck in gene therapy.
• Life Cycle Assessment — ISO-standard methodology calculating product environmental footprint from cradle to end of life across multiple impact categories; increasingly applied in biopharma with AI-enhanced tools and new PAS 2090 standard for drug products.
• Linear Pluggable Optics — Pluggable optical transceiver variant that removes the DSP retimer to reduce power and latency, relying on the host ASIC's SerDes for signal conditioning.
• Linear Receive Optics — Hybrid optical transceiver with DSP on the transmit side only, combining power savings of LPO receive with signal robustness of DSP transmit.
• Linear-Drive Pluggable Optics — Pluggable optical transceiver standard that eliminates the DSP retimer, reducing power versus traditional pluggables while preserving supply chain flexibility; an intermediate step before CPO.
• Link Flap — An unexpected state transition on an optical link defined by 802.3 loss of lock (three uncorrected codewords); the primary reliability threat to large-scale RDMA AI fabrics.
• Lipid Nanoparticles — Targetable lipid-based delivery vehicles for nucleic acid payloads; CBTX deploys LNPs formulated for single-stranded DNA and decorated with two targeting ligands for organ-specific in vivo T-cell programming.
• Liquid Cooling Filtration — Liquid cooling filtration encompasses the standards and practices for filtering coolant in thermal cooling systems (TCS) to protect cold plate microchannels from particulate fouling.
• Liquid Cooling Loop Design — The engineering discipline of selecting coolant inlet temperatures, CDU configurations, and hose materials for AI rack liquid cooling loops; involves three heat exchange stages and L10/L11 thermal simulation to match CDU specifications to customer data center water conditions.
• Liquid Cooling Production Line Automation — Automated test and process tooling for liquid-cooled server manufacturing lines, consolidating leak test, flushing, filling, draining, burn-in, drying, and nitrogen injection into a single integrated cabinet.
• Liquid Cooling TCS — Liquid Cooling TCS (Thermal Cooling System) refers to the closed-loop liquid distribution infrastructure that delivers coolant from Coolant Distribution Units (CDUs) through piping networks to server racks equipped with cold plates or other liquid-cooled heat exchangers.
• Liquid Cooling for AI Data Centers — Direct Liquid Cooling (DLC) technology for AI GPU racks that reduces cooling power overhead from ~37% to ~2% of total facility power vs air cooling, enabling double the GPU density in the same facility footprint.
• Liquid Cooling for AI Infrastructure — Direct liquid cooling solutions for AI servers and racks including cold plate (D2C), CDU design, and rack-scale integration, replacing air-cooled solutions to address growing thermal density from GPU workloads.
• Liquid Cooling for Data Centers — Direct liquid cooling technologies—cold plates, CDUs, and in-row systems—enabling removal of heat from high-density AI server racks with TDPs up to 3.6 kW per chip.
• Liquid Cooling for High-Density AI Compute — Liquid cooling (CDUs, direct liquid cooling) is mandatory for AI rack densities at or above 1 MW; air cooling cannot support these power levels, and the 2027 consolidated architecture requires liquid-cooled data halls fed by 800 V DC busway.
• Liquid Cooling for Networking — Liquid cooling applied to high-power network switch ASICs is becoming mandatory as Tomahawk 6-class silicon approaches GPU-level power dissipation (~2kW per chiplet), marking the OV3 generation (2026) as the industry inflection point for widespread adoption.
• Liquid Cooling for SSDs — Emerging thermal management approach for solid-state storage, addressing hot-swap serviceability, dual-sided PCB cooling, and the need for new industry specifications.
• Liquid-Cooled Bus Bar — Liquid-cooled bus bars integrate coolant channels into rack power distribution bars, delivering 5x the current capacity of air-cooled equivalents in the same form factor, enabling rack power beyond the ~200kW air-cooling ceiling toward 750kW and eventually 1.2MW.
• Liquid-Cooled I/O — Liquid-cooled I/O connectors integrate coolant channels into optical transceiver cage assemblies, managing the 40W per 1.6T port thermal load that prevents high-density optical I/O from operating within air-cooled rack thermal budgets.
• Live Migration — Moving a running VM between physical hosts without the VM observing the interruption, central to hyperscale SSD reliability and fleet load balancing.
• MEK-MPA — TCG multi-party authorization feature set for media encryption keys — a cross-compatible extension of TCG Opal letting mutually distrustful entities each gate access independently without sharing secrets.
• MGX Rack Form Factor — NVIDIA's MGX modular rack architecture defining B-bay and C-bay slot layouts for IO, power supplies, storage, and rack management controllers; QCT shipped the first MGX rack-scale solution in the world in 2023.
• Machine Learning Failure Prediction — Microsoft's ML-based SSD failure prediction layered over expert-rule thresholds, achieving >80% precision and recall to drive fleet self-healing.
• Manufacturability Assessment — A systematic pre-launch evaluation of biologics manufacturing risks performed by a product technical leader with an operations mindset during the put-to-phase, functioning as an independent check on CMC development to identify and correct issues before commercialization.
• Manufacturing Onshoring — The geopolitical trend driving pharmaceutical and biologic manufacturers to establish or relocate production within the US or regional markets, creating unprecedented capital investment particularly in North Carolina and Indiana and intensifying competition for biomanufacturing talent.
• Mass Balance Approach — Supply chain certification methodology substituting bio-based, bio-circular, or circular feedstocks at the cracker step to produce physically identical end materials with chain-of-custody credit — accepted by regulators and adopted in consumer products for over a decade.
• Material Compatibility Testing — Material Compatibility Testing encompasses the protocols and methodologies used to assess whether server components and infrastructure equipment — plastics, elastomers, metals, PCBs, connectors, and mechanical parts — can survive prolonged exposure to dielectric immersion fluids without degradation of physical, electrical, or mechanical properties.
• Media Encryption Key — The per-range key used by a storage device's AES engine to encrypt data; in lock it is solely managed by the key management block and can be random-and-encrypted-at-rest or derived from secrets.
• Medium Purification Module — Replaceable ion-exchange resin module that is the core quality-control component of pure water coolant systems, maintaining conductivity and microbial levels through continuous ionic filtration without generating wastewater.
• Metachromatic Leukodystrophy — Fatal progressive neurodegenerative disorder treated with hematopoietic stem cell gene therapy; clinical data showed treated patients halted disease progression and gained motor or cognitive milestones versus untreated controls who progressively lost function.
• Microgrid and BESS — Microgrid and BESS (Battery Energy Storage System) refers to the campus-level power management architecture that integrates multiple distributed energy resources — utility grid connections, diesel or gas generators, photovoltaic (PV) arrays, and battery storage — under a unified control system optimized for data center operations.
• Mim8 — Novo Nordisk bispecific antibody bridging factor IX and factor X — analogous to emicizumab but far more potent — projected at 20-50 mg once-monthly pen-injector dosing for hemophilia A; manufactured from two CHO cell lines using controlled Fc-hinge disulfide exchange; BLA submitted within 5 years of Phase 1 start in 2020.
• Model Distillation — Knowledge distillation technique popularized by DeepSeek's open-source release that lets a smaller model learn to replicate the outputs of a larger model, enabling AI entry without a full training data lake.
• Modular Data Center Prefabrication — Off-site factory assembly of integrated power, cooling, and IT modules delivered pre-commissioned to site, cutting deployment timelines and removing the need for building modifications when power capacity scales.
• Modular Hardware System — OCP standard (DC-MHS) defining modular server architecture with interchangeable HPM (motherboard), chassis, and front-end I/O components to enable workload-specific customization while maintaining standard interface points.
• Modular Server Architecture — Modular server architecture decomposes monolithic server motherboards and chassis into standardized, interchangeable modules with shared interfaces, enabling parallel engineering teams and faster integration of diverse silicon architectures without multiplying design teams.
• Module Type Package — An industry standard software element (MTP) that standardizes how bioprocessing equipment connects to manufacturing infrastructure, reducing integration time from 9 months to 2 days.
• Moore's Law — Moore's Law is the empirical observation that transistor density roughly doubles every two years; cited at OCP 2025 as fundamentally insufficient to meet AI training compute demand growth of ~10x per year, creating a 5x annual gap that drives demand for advanced packaging and custom silicon.
• Multi-Core Fiber — Multi-core fiber (MCF) is an optical fiber technology that embeds multiple independent glass cores within a single 125µm cladding — the same outer diameter as conventional single-mode fiber (SMF).
• Multi-Plane Networking — Network fabric design that splits GPU NIC bandwidth across multiple independent leaf-spine planes, reducing switch count and capex while improving fault isolation compared to single-plane topologies.
• Multi-Tenant SSD Sharing — Architecture in which a single physical SSD is shared by multiple VMs or tenants, requiring performance isolation, fair resource allocation, and QoS guarantees to meet customer SLAs.
• Multiple Reclaim Group FDP — Solidigm-proposed FDP configuration exposing many reclaim groups (e.g. one per die) to multiply placement IDs for finer data separation and QoS in hyperscale storage.
• Myotonic Dystrophy Type 1 — Autosomal dominant neuromuscular disease caused by CTG repeat expansion in the DMPK gene with nuclear RNA pathobiology; an entirely underserved indication targeted by Dyne Therapeutics' Dyne-101 using ASO-based DMPK knockdown.
• NVIDIA Holoscan — NVIDIA's high-speed sensor bridge and streaming platform that connects edge sensors (cameras, lidar) to GPUs via GPU-Direct, enabling real-time sensor data ingestion for edge AI inference and simulation.
• NVIDIA Omniverse — NVIDIA's 3D simulation and digital twin platform used as the simulation tier in the three-computer architecture, enabling robots and factory systems to be trained and validated in virtual environments before physical deployment.
• NVMe CLI — Open-source OCP NVMe CLI tooling repeatedly cited as adding industry value; Microsoft proposes extending it for standardized fleet self-healing.
• NVMe Controller Migration — Prior NVMe technical proposals (TP4165 dirty tracking, TP4159 controller migration) that the exported-subsystem work builds upon for live VM migration.
• NVMe Controller SoC Design — NVMe Controller SoC Design is the system-on-chip engineering discipline balancing competing NVMe performance, capacity, security, and power constraints across PCIe generations.
• NVMe Data Areas — NVMe telemetry data organization structure with Data Areas 1-4 partitioning SSD debug information by human-readability, size, and urgency within the OCP Data Center NVMe SSD specification.
• NVMe TP4176 — NVMe Technical Proposal for Quality of Service: PCIe bandwidth and IOPS rate limiting per SSD controller, establishing the foundation for tenant-level performance isolation in shared storage.
• Neocloud — New category of AI-focused cloud infrastructure providers — including GPU cloud companies and sovereign operators — emerging alongside traditional hyperscalers to serve AI model training and inference demand.
• Non-Scheduled Fabric — Meta's production AI scale-out network architecture based on shallow-buffer Ethernet switches with adaptive routing, replacing the Disaggregated Scheduled Fabric for gigawatt-scale clusters.
• OCI MSA — The OCI MSA (Optical Compute Interface Multi-Source Agreement) is an open specification for optical scale-up interconnects, co-founded on March 12, 2026 by AMD, Microsoft, Broadcom, NVIDIA, Meta, and OpenAI.
• OCP Building Block Strategy — Design methodology using pre-validated, modular hardware components (motherboards, chassis, front-end I/O) aligned to OCP MHS standard, allowing workload-specific customization without full new-product development cycles.
• OCP Compliance Test Suite — The OCP NVMe SSD compliance test suite, complementary to NVMe, used to validate drives early in development and qualify them faster.
• OCP Diag AutoVal Framework — Open-source SSD test automation framework released by Meta and OCP on GitHub, providing a shared qualification test suite with setup/execute/cleanup lifecycle and OCP Diag-compliant JSON output.
• OCP Lock — OCP Lock (Layered Open-source Cryptographic Key Management) is an OCP standard providing key management to both drives and hosts, applicable to any storage device, integrated into Caliptra 2.1 with a complementary Trusted Computing Group host interface specification.
• OCP Marketplace — The OCP Marketplace is OCP's online catalog of compliant hardware, where storage listings and SSDs are growing rapidly.
• OCP NVMe Latency Monitoring — OCP-specified NVMe feature that classifies SSD I/O operations into configurable latency buckets, persists stall events across resets, and provides timestamped debug logs enabling fleet-wide tail latency detection.
• OCP NVMe SSD Telemetry — Standardized telemetry framework in the OCP Data Center NVMe SSD specification enabling human-readable SSD debug logs via open-source nvme-cli tooling organized across Data Areas 1-4.
• OCP OAM Module — OCP-standard Open Accelerator Module form factor enabling high-performance AI accelerator integration in rack servers, targeting very large model inference and training in the 10B–1T parameter range.
• OCP Safe — OCP Safe is an OCP security audit and reporting framework that makes hardware security attestation and audits routine and inexpensive; Google now regularly requests OCP Safe reports from its hardware vendors.
• OCP Storage Tech Talk — The OCP Storage Tech Talk is OCP's annual storage technical talk; the fourth edition is a recorded three-hour session.
• OT Cybersecurity — OT (Operational Technology) Cybersecurity addresses the protection of physical infrastructure control systems — PLCs, IPCs, SCADAs, building management systems, and robotic platforms — in data center environments.
• Oligonucleotides — Oligonucleotides are a nucleic-acid drug class whose commercial proof point was Spinraza.
• Ontologies and Knowledge Graphs — Standardized vocabularies and structured data models giving bioprocessing data shared meaning across organizations, enabling AI to generate insights and identify gaps in upstream, downstream, and manufacturing contexts.
• Open All-Photonic Network — IOWN Global Forum's open architecture for disaggregated optical transport, extending photonic infrastructure to small and medium sites via capillarization and softwarization, with DWDM transceivers shifted to customer premises.
• Open BMC — Open-source baseboard management controller firmware that enables unified server management via Redfish APIs, used for diagnosis automation, device telemetry, and firmware updates in complex GPU server systems.
• Open Chiplet Firmware — Open Chiplet Firmware — emerging area of firmware for chiplet-based processors where AMI is positioning itself as a thought leader, expecting a major industry technology shift.
• Open Networking For Telco — Open Networking for Telco refers to the adoption of whitebox switches, open-source NOS (primarily SONiC), and disaggregated software/hardware stacks in telecommunications environments.
• Open Platform Firmware — Open Platform Firmware (OPF) is an OCP work group co-led by HPE, focused on refactoring server firmware for modularity to match modular hardware, enabling auto-discovery and auto-configuration of attached hardware rather than relying on hardcoded system knowledge.
• Open Rack v3 — OCP's 21-inch open rack standard with a 48V DC busbar; the deployment platform for Meta's first QLC flash storage server.
• Open Source Firmware — Open-Source Firmware refers to the ecosystem of open-source firmware stacks for data center server management, boot, and security, centered around OpenBMC (baseboard management controller), LinuxBoot (UEFI alternative), and Cerberus (hardware root of trust).
• Optamax Bulk Cable — Luxshare-Tech's 224G twinax bulk cable product available in 26–31 AWG gauges with 92-ohm impedance and dual ground pin design, used in cable cartridge and cable backplane assemblies.
• Optical Circuit Switch — Switching technology central to IOWN Global Forum's AI accelerator interconnect work item, argued to be essential for fully unlocking co-packaged optics potential in scale-up and scale-out AI networks; the basis for alignment with OCP's OCS subgroup.
• Optical Circuit Switching — Optical Circuit Switching (OCS) establishes physical light paths between fiber ports without electro-optical-electronic (O-E-O) conversion, routing photons directly through the switching fabric.
• Optical Module Telemetry — DSP-level real-time instrumentation of optical transceivers providing signal quality metrics (SNR, FEC histograms, I-height) and health indicators (ESD, fiber multipath) used for predictive link management.
• Optimized Provisioning — A drive-managed HDD regeneration approach (Solution 3) where the drive autonomously virtualizes physical blocks under a failing head, redirecting LBA accesses to healthy heads without requiring host intervention.
• Organizational Decision-Making — A core leadership competency in biotech encompassing who decides, how decisions are communicated, and preventing post-meeting 'decision shopping'; John Cox and Ann Lee both advocate for single-decision-maker models with mandatory debate followed by fully unified execution.
• Overprovisioning — Reserving spare SSD capacity to reduce garbage collection; in FDP, ScaleFlux proposes managing it per reclaim unit rather than as a shared pool.
• P4 Programmable Networking — Domain-specific language and architecture for programmable NIC and switch data planes, enabling custom transport protocols, load balancing, and in-service updates without hardware changes.
• PAM4/PAM6/PAM8 Modulation — PAM4 (4-level pulse amplitude modulation) is the current standard for 112G and 224G copper and optical interconnects in AI data centers; PAM6 and PAM8 are under evaluation to extend per-lane speeds to 336G and 448G+ on copper, reducing the pressure to transition to optics.
• PCIe 6 — Sixth-generation PCI Express specification at 64 GT/s enabling high-bandwidth fabric switches and scale-up networks; Astera Labs has production-deployed PCIe 6 switches following over a year of cloud-scale interop testing.
• PCIe 7.0 — The next PCIe generation whose 32 GHz Nyquist signaling poses signal-integrity challenges driving changes across EDSFF connector, device and cable specs.
• PCIe Exported NVM Subsystem — NVMe technical proposal extending the exported NVM subsystem from NVMe-oF to PCIe to virtualize an SSD with host-defined identifiers and templates for seamless migration.
• PCIe Gen 6 — Next-generation PCIe rapidly gaining traction in storage, bringing higher clock speeds and an I2C-to-I3C management interface shift.
• PCIe Gen Roadmap for SSDs — Evolution of PCIe interface generations (4, 5, 6, 7) for SSDs, with each generation increasing storage bandwidth but also power requirements, forcing the transition from air to liquid cooling.
• PFAS — PFAS (Per- and Polyfluoroalkyl Substances), colloquially called "forever chemicals," are a broad class of synthetic fluorinated compounds with carbon-fluorine bonds that resist degradation in biological and environmental systems.
• PMOs — Phosphorodiamidate morpholino oligomers — uncharged synthetic oligonucleotides used for exon skipping by targeting pre-mRNA in the nucleus; the payload class used in Sarepta's eteplirsen and in Dyne's FORCE platform for DMD.
• Packet Delivery Context — UEC's term for its ephemeral, software-invisible reliability contexts that replace traditional connections; created on demand, torn down when idle, with zero-round-trip pipelined startup and one-RTT close.
• Paravirtualization — Legacy virtual-SSD approach where the hypervisor fully emulates the NVMe device, incurring CPU and IO-path overhead that the exported-subsystem work aims to eliminate.
• Perfusion Bioprocessing for Viral Vectors — Continuous perfusion-based bioreactor strategies for retroviral and lentiviral vector manufacturing, developed at IBET from the early 1990s to address the fundamental biological constraint that retroviruses have an 8-hour half-life at 37°C, making batch culture highly inefficient.
• Personalized CRISPR Therapy — Personalized CRISPR therapy is a bespoke gene-editing treatment manufactured per patient.
• Personalized Cancer Vaccine — mRNA-based individualized cancer vaccines that sequence a patient's tumor to identify unique neoantigens and manufacture a custom construct within weeks, representing a 'bench to one' manufacturing challenge and the frontier of scale-out bioprocessing.
• Physical AI — Physical AI refers to AI systems that perceive and act in the physical world — spanning autonomous vehicles, humanoid robots, autonomous mobile robots (AMRs), drones, and industrial systems.
• Physical AI Scientist — Physical AI Scientist is Medra's term for a robotic system that executes diverse biological experiments, logs process data, and continuously learns from the results.
• Platform Biomanufacturing — The established bioprocessing industry model where standardized development and manufacturing processes are applied to new molecules without customization; increasingly obsolete due to molecular format diversity that requires bespoke approaches for each modality
• Post-Quantum Cryptography — Cryptographic methods designed to be secure against quantum computers, with NIST actively reviewing and recommending standards to replace quantum-vulnerable algorithms like RSA.
• Power Distribution Board — The Power Distribution Board (PDB) in AI rack compute trays converts busbar voltage to component-level supply voltages; QCT's GB200 PDB evolved from a 1.3 kW dual-rail GH200 board to a single 8 kW centralized board delivering 12V to all components while isolating the compute motherboard from power subsystem failures.
• Power Usage Effectiveness — Power Usage Effectiveness (PUE) is a data center efficiency metric measuring total facility power relative to IT load; conventional air cooling achieves PUE 1.6–1.7 while direct liquid cooling targets below 1.1.
• Power and Compute Convergence — Architectural and organizational paradigm requiring IT hardware, power delivery, and cooling to be co-designed as a single integrated stack, replacing the historically siloed data center discipline structure that cannot support AI-scale rack densities.
• Precision Immersion Cooling — Cooling approach using a small amount of dielectric mineral oil circulated within individual server/device enclosures to achieve 100% thermal efficiency without full-tank immersion.
• Predictive Maintenance — AI-driven equipment monitoring that identifies potential failures before they occur in pharmaceutical manufacturing, enabling proactive scheduling and reducing unplanned downtime.
• Prime Editing — Third-generation gene editing platform that makes precise edits without double-strand DNA breaks; capable of transitions, transversions, insertions, and deletions; first published in 2019 and commercialized through Prime Medicine.
• Process Analytical Technology — FDA-recognized framework for real-time in-process measurement and control of biopharmaceutical manufacturing quality, formally established in 2004 guidance; adoption remains slow despite available technology due to business case gaps and organizational risk aversion
• Process Mass Intensity — Quantitative resource efficiency metric for pharmaceutical processes showing mass of all inputs per unit product; for biotherapeutics it highlights water as the dominant problem, diverging from carbon footprint analysis which highlights electricity and consumables.
• Process Modeling and Mechanistic Models — Computational representations of bioprocessing unit operations using first-principles chemical engineering or empirical fitting to predict process behavior, reduce experimental burden, support regulatory filings, and enable model-based lifecycle management
• Process Performance Qualification — Regulatory requirement for biologics manufacturing demonstrating a process consistently produces product meeting its specifications; the FDA mandates prospective PPQ for gene therapy manufacturing components, and Prime Medicine pioneered 'PPQ-as-you-go' to align validation with clinical timelines.
• Process-as-Product Paradigm — The cell and gene therapy industry axiom 'the process is the product,' critiqued by Matthew Dy as historically encoding ignorance rather than discipline, and reframed as a mandate for deep process characterization that enables continuous platform improvement and regulatory knowledge reuse.
• Processing-in-Memory — Processing-in-Memory (PIM) embeds arithmetic logic units directly in DRAM banks, executing instructions in parallel across all banks simultaneously to reduce data movement and improve bandwidth and energy efficiency for LLM workloads.
• Project Deschutes — Project Deschutes is Google's OCP-contributed CDU (Coolant Distribution Unit) specification with open design collateral, encoding 10 years of Google liquid cooling experience for broad industry adoption.
• Pure Water Coolant — Water-based ICT coolant that uses a medium purification module (MPM) for quality control via conductivity, pH, and dissolved oxygen monitoring, offering interoperability and cost advantages over formulated coolants.
• QLC Flash — Quad-level cell NAND flash storing 4 bits per cell; a capacity- and power-efficient middle ground between HDD and TLC SSD, selected by Meta for read-heavy AI and storage workloads.
• QLC NAND — Quad-level-cell NAND positioned between TLC and HDD in the storage pyramid, driving high-capacity SSD demand through density and terabytes-per-watt gains.
• Quantum Circuits — The programming model for quantum computers: ordered sequences of quantum gates applied to qubits, with an infinite theoretical instruction set that gives quantum devices their exponential algorithmic advantage over classical processors.
• Quantum Emulation — Software running on classical silicon that reproduces quantum device behavior error-free for workloads up to ~35–40 qubits, enabling quantum algorithm development and testing without consuming scarce QPU resources.
• Quantum Processing Unit — QPU — a compute accelerator using qubits rather than bits, positioned as a GPU analog that augments classical data centers rather than replacing them, with mass commercial deployment targeted for around 2030.
• Quantum Superposition — Quantum mechanical property enabling qubits to exist in a continuous superposition of 0 and 1 states simultaneously, giving quantum circuits exponentially more computational states than classical bits.
• RDMA — Remote Direct Memory Access; a 25-year-old high-performance networking paradigm enabling kernel-bypass, zero-copy, hardware-offloaded data transfer; the foundation that UEC is modernizing.
• RDMA Fabric — Remote Direct Memory Access network interconnect used in large-scale AI training clusters; highly sensitive to L1 optical link instability which causes checkpoint thrashing and job failures.
• RISC-V — RISC-V is an open hardware instruction set architecture (ISA) competing with x86 and ARM, experiencing rapid growth in open source and open hardware communities, and increasingly requested by enterprise customers seeking energy efficiency and workload-specific performance advantages.
• ROCm — ROCm is AMD's open-source GPU software stack — libraries, compilers, tools, and runtimes — powering the Instinct GPU line and providing portability across PyTorch, Triton, and over a million Hugging Face models.
• Rack Liquid Shipping (Dry vs. Wet) — The two logistics approaches for shipping liquid-cooled AI racks: dry-shipped (coolant drained, nitrogen-purged, lighter but requires on-site refill and de-bubbling before power-on) vs. wet-shipped (pre-filled, heavier, temperature-controlled transport, but ready to power on).
• Rack Scale AI Computing — Rack-Scale AI Computing represents the paradigm shift from server-level (4-8 GPUs per server) to rack-level (32-128+ GPUs per rack) computing.
• Rack-Level Energy Storage — Future-oriented mitigation for AI training data center power swings: battery/capacitor systems at rack, corridor, or data center level that charge during low-power communication phases and discharge during high-power compute phases.
• Rack-Scale AI Infrastructure — The industry shift from individual server (L10) procurement to full rack (L11) and multi-rack cluster (L12) deployments as the primary unit of AI infrastructure, requiring integrated power, networking, cooling, and validation.
• Radiopharmaceuticals — Cancer drugs that deliver radioactive isotopes directly to tumor cells via targeting molecules; extremely short half-lives impose a manufacture-to-dose window of hours, requiring proximity manufacturing and compressed supply chains.
• Rare Disease Manufacturing Platforms — A strategy for achieving manufacturing scale in rare disease therapeutics by identifying common platform technologies (lentiviral vectors, lipid nanoparticles) shared across individually small patient populations, enabling shared economics and regulatory prior-knowledge reuse across programs.
• Rebalancing Hemostasis — Therapeutic strategy for hemophilia that restores hemostatic balance by reducing anticoagulant activity rather than adding procoagulants; two main approaches are siRNA knockdown of antithrombin (fitusiran) and antibody inhibition of TFPI; both work across hemophilia A, B, and inhibitor patients.
• Reclaim Unit Handle — The FDP construct a host writes through to steer data into separate physical reclaim units; modern SSDs typically expose about eight.
• Redfish — DMTF standard REST API for server management that unifies GPU server diagnosis by providing a single interface for state register reads, telemetry access, health status, and firmware updates.
• Redfish Liquid Cooling Management — Redfish is DMTF's (Distributed Management Task Force) open standard for hardware management APIs.
• Reference Architecture — A reference architecture is a standardized, repeatable deployment blueprint for AI data center infrastructure at pod or server-unit (SU) level, promoted by NVIDIA and adopted by vendors to eliminate customization risk and accelerate time-to-deployment.
• Regulatory Science — The government-funded discipline of developing scientific methods and expertise to validate and regulate novel bioprocessing technologies, ensuring regulatory bodies keep pace with industry innovation.
• Regulatory Science for ATMPs — Evolving regulatory frameworks for Advanced Therapy Medicinal Products (ATMPs) including gene and cell therapies, covering registration-ready clinical trial design, risk-based approaches to process changes, and international regulatory convergence.
• Regulatory Strategy for Novel Modalities — Approach to IND and approval pathway for unprecedented therapeutics; emphasizes risk-based evolving data packages, proactive regulator education, and orthogonal analytical methods rather than standard data sets.
• Reinforcement Learning from AI Feedback — Post-training technique (RLAIF / RL Self-Evolution) used in FoxBrain to teach reasoning without human supervisors, referenced in both Nemotron and DeepSeek pipeline examples.
• Retrieval-Augmented Generation — Hybrid Document-RAG and Graph-RAG pipeline powering Foxconn's FoxBrain office workload assistant, combining vector retrieval, full-text search, and graph-based relational reasoning.
• RocksDB — Log-structured merge key-value store repeatedly used as the canonical benchmark for demonstrating FDP write-amplification benefits.
• SAI Switch Abstraction Interface — SAI (Switch Abstraction Interface) is OCP's hardware abstraction layer enabling multi-vendor switch silicon programming through a standardized API.
• SCADA — Scaled Accelerated Data Access — a GPU-initiated storage interface/API giving GPUs unbounded, error-tolerant, fine-grained access to NVMe.
• SFF-TA-1002 Connector — The EDSFF connector spec (also used by OCP NIC 3.0) whose six power pins support up to 79.2W, now the basis for EDSFF power definitions.
• SONiC Ecosystem — SONiC (Software for Open Networking in the Cloud) is an open-source network operating system hosted by the Linux Foundation that has become the dominant NOS for AI data center fabrics.
• SRv6 For AI Networks — SRv6 (Segment Routing over IPv6) applied to AI backend and DCI networks provides explicit source-routed paths for low-entropy elephant flows that are poorly served by ECMP-based routing.
• SSD Firmware Left-Shifting — Practice of moving SSD firmware qualification testing earlier into the vendor's own lab before release, enabling defect detection before the hyperscaler dedicates regression resources.
• SSD Liquid Cooling — An EDSFF standardization effort to add a single cold-plate cooling solution per form factor that is backwards compatible across air and liquid cooling.
• SSD Preconditioning — The process of writing an SSD to a representative steady state before performance testing.
• SSD Thermal Throttling — Mechanism by which SSDs reduce performance when operating temperatures exceed safe thresholds; at 77°C performance drops to ~58%, at 79°C to under 1%.
• Scale-Across Network — Emerging network tier beyond scale-out that connects AI training clusters spanning multiple data centers; requires approximately 16,000 ports and 13 petabits/sec bandwidth, exceeding modular chassis capacity and requiring fixed pizza-box leaf-spine topologies.
• Scale-Across Networking — WAN or multi-site networking that connects GPU clusters across multiple data centers or geographic locations, needed when power, cooling, space, or availability constraints prevent a single-site million-GPU cluster.
• Scale-Out Networking — Backend AI networking fabric connecting multiple racks within a single data center, requiring intelligent load balancing and congestion management to eliminate hash-collision-induced link underutilization.
• Scale-Up Networking — Load-store interconnect network within a rack or node that connects GPUs with high bandwidth; approximately 80% of GPU bandwidth flows over scale-up links.
• Scheduled Ethernet Fabric — Scheduled Ethernet Fabric is an OCP-compliant networking architecture that uses receiver-based scheduling to guarantee congestion-free, collision-free AI training fabric operation.
• Science-Based Targets — Corporate GHG emissions reduction commitments aligned with Paris Agreement climate science, verified through SBTi; increasingly a prerequisite for pharma supplier preferred status, with AstraZeneca among the most demanding purchasers.
• Scope 3 Emissions — Indirect supply chain greenhouse gas emissions; in biopharma framed as mutually interdependent ('my scope 3 is your scope 2'), driving cross-industry collaboration on supplier decarbonization and embedded as a standard business metric.
• SerDes — Serializer/Deserializer (SerDes) is the high-speed electrical interface IP used for chip-to-chip and rack-level interconnects; MediaTek has 224G SerDes in production and is developing 448G, while noting that ~400G may represent the physical copper speed limit, prompting the transition to Co-Packaged Optics.
• Shingled Magnetic Recording — An HDD recording technology that writes tracks in overlapping shingled zones, enabling data-safe drive regeneration by removing only the zones associated with a failed head while leaving all other data intact.
• Signal Integrity in Immersion — Signal Integrity (SI) in immersion cooling addresses how submersion of IT equipment in dielectric fluid affects high-speed electrical signal transmission.
• Silicon Photonics — Silicon photonics uses silicon as a photonic material platform for optical interconnects, switches, and computing elements — leveraging existing CMOS fabrication infrastructure for cost-effective, high-density photonic integrated circuits.
• Single-Use Bioprocessing — Use of disposable polymer and plastic components in biopharmaceutical manufacturing; a major sustainability focus due to plastic waste, with circularity and extended product lifetimes identified as key paths forward.
• Single-Use Technology — Disposable bioprocessing equipment pioneered in China by Henlius for antibody production, now used by the majority of Chinese biologics manufacturers and a critical single-source vulnerability exposed by COVID-19.
• Smart Manufacturing — Robotic-arm assembly systems synchronized with digital simulation that self-adjust during production to reduce human error and improve yield rates, deployed by Pegatron for AI server manufacturing.
• Soft vs Hard Rate Limits — Two SSD QoS rate-limiting modes: hard limits strictly cap a tenant at its configured bandwidth/IOPS; soft limits allow a tenant to consume idle capacity from other tenants to maximize SSD utilization.
• Software 3.0 — Software 3.0 is a framework introduced by Andrej Karpathy describing a new computing paradigm where LLMs serve as programmable interpreters and the context window replaces source code as the primary programming surface.
• Software-Defined Networking — Network control plane approach decoupling software from hardware; used in AI data centers with SONiC, FBOSS, and proprietary NOS options, evolving toward AI-based predictive orchestration starting ~2026.
• Solid-State Transformer — A Solid-State Transformer (SST) is a power electronics device that converts medium-voltage AC (typically 13.8-34.5 kV from the utility grid) directly to low-voltage DC (400V, 800V, or 1500V) suitable for data center rack distribution.
• Sprandom — A fast, pseudo-random SSD preconditioning method that short-circuits to the target overprovisioning distribution in a single pass.
• Standards Harmonization — Standards Harmonization in the data center cooling context refers to the effort to align OCP specifications with those of established standards bodies — ASHRAE, ASTM, UL, IEC, and IEEE — to prevent specification conflicts, reduce duplicate testing, and enable global interoperability.
• Storage Disaggregation — Moving storage out of the compute node into specialized, shareable pools to meet GPU IOPS demands that cannot fit locally.
• Storage Element Depopulation — An HDD firmware feature that removes a failed or failing read/write head from the drive's addressable LBA space, allowing the drive to continue operating at reduced capacity (typically 95% for a 20-head drive).
• Storage Next — A collaborative OCP effort with direct stakeholders to define IOPS/TCO-optimized SSD requirements and bring them to standards.
• Storage Telemetry — Device telemetry in OCP specs, expanded from debug toward periodic uploads for at-scale failure diagnosis and prediction, now human-readable and per-context.
• Supply Chain Resilience in Biomanufacturing — Strategic diversification of raw material and consumable sourcing across geopolitically stable regions, a capability exposed as a critical gap during COVID-19 and intensified by US-China trade tensions.
• Sustainability by Design — Approach to embedding environmental sustainability into product and process design from the outset of development — enabling lower carbon footprints at launch and aligning sustainability with core business metrics for commercial advantage.
• Switch ASIC Scaling — Roadmap of switching ASIC aggregate bandwidth capacity: 51.2T (2025) → 102.4T (2026-2028) → 204.8T (2028-2030), with corresponding SerDes evolution from 100G to 200G per lane driving tighter signal integrity requirements.
• Synthetic Data for LLM Training — AI-generated training data — produced via model self-play, agentic pipelines, and data augmentation — increasingly dominant over real data for post-training and pre-training of large language models.
• Synthetic Ester Cooling Fluids — PFAS-free, biodegradable dielectric fluids based on synthetic esters, designed for data center liquid cooling in both direct-to-chip and immersion configurations; offer high flash points, strong dielectric properties, and eliminate electrical damage risk from leakage.
• T-shaped Skills — A workforce capability model where bioprocessing professionals have deep expertise in one domain but sufficient breadth to understand adjacent upstream and downstream processes.
• TCG Opal — TCG storage security class that models everything as tables and objects, with AES-256 key objects gated by the locking table's read-locked/write-locked states; finalized as version 1.0.
• TFPI Inhibition — Therapeutic strategy for hemophilia that inhibits tissue factor pathway inhibitor (TFPI), a three-Kunitz-domain anticoagulant, at the K2 domain to rebalance coagulation; mastacimab demonstrated 92% bleeding rate reduction in Phase 3; Novo Nordisk's higher-affinity anti-TFPI candidate requires daily dosing.
• TFR1 Targeting — Delivery strategy exploiting transferrin receptor 1's ubiquitous expression on muscle, cardiac tissue, BBB endothelium, and CNS neurons to shuttle oligonucleotide conjugates into cells via receptor-mediated endocytosis.
• Taiwan AI Supply Chain — The ecosystem of Taiwanese companies providing semiconductor, system integration, and hardware solutions that form the midstream layer of the global AI infrastructure value chain alongside US AI chip and software companies.
• Target-Mediated Drug Disposition — Pharmacokinetic phenomenon where antibody binding to its target accelerates systemic clearance, creating a paradox for anti-TFPI antibodies where improving binding affinity worsens PK; resolved for mastacimab via mechanistic PK/PD modeling that identified a moderate nanomolar affinity at 100 mg weekly as sufficient.
• Three-Computer Architecture — NVIDIA's framework dividing AI system development into three compute tiers: training (DGX), simulation/digital twin (OVX/Omniverse), and edge runtime inference (IGX/Jetson), enabling a closed-loop model refinement cycle for physical AI.
• Tightly Coupled Memory — Tightly Coupled Memory is a low-latency, single-cycle-access memory architecture in the ARM Cortex-R82 that streamlines storage firmware command fetch and data structures.
• Time to First Token — Time to first token (TTFT) is the latency from an inference request submission to the generation of the first output token; it has emerged as the defining performance metric driving AI data center design, cooling architecture, and deployment speed priorities.
• Token Bucket Rate Limiting — Algorithm mechanism used to implement SSD QoS rate limiting: a bucket of tokens refills at a constant rate, excess tokens are discarded, and commands consume tokens proportional to IOPS or data size.
• Tomahawk 6 — Tomahawk 6 (TH6) is Broadcom's sixth-generation Ethernet switch ASIC, the world's first 100 terabit non-blocking switch at 102.4Tbps, featuring 2,355 billion transistors, 200G SerDes, and 64x 1.6TbE OSFP ports; the only silicon enabling a two-tier topology for 128K-GPU clusters.
• Total Cost of Ownership — Holistic cost framework for evaluating networking technology choices, encompassing power, hardware cost, reliability, and opportunity cost of GPU deployment density.
• Transformer Architecture — Deep learning model architecture introduced around 2017–2018 that enabled billion-parameter AI models and drove exponential growth in required computing power, sparking the current AI infrastructure buildout.
• Two-Phase Cold Plate — Two-Phase Cold Plate technology uses dielectric fluids that undergo phase change (boiling) within the cold plate to transfer heat from processor dies.
• Two-Phase Immersion Cooling — Two-Phase Immersion Cooling is a data center cooling approach in which IT equipment is fully submerged in a dielectric fluid that undergoes phase change — boiling at chip-level heat flux, condensing at the tank lid or a separate condenser, and returning to liquid continuously.
• Two-Phase Passive Cooling — Two-phase passive cooling uses the latent heat of a refrigerant phase change to move heat from a cold plate to a remote condenser without any pump.
• U-Frog — Kioxia's open-source RocksDB plugin that uses io_uring to bypass the file system and send FDP placement directives directly to the device.
• U.2 Long Form Factor — Pure Storage's interim extended-U.2 drive form factor reaching 300TB while the industry transitioned to the standardized EDSFF E2.
• UALink — Open standard for GPU scale-up interconnect developed by the UALink Consortium; delivers 800 Gbps full-duplex bandwidth with memory-semantic semantics and scales to 1,024 GPUs.
• UCIe — Universal Chiplet Interconnect Express — an open industry standard for die-to-die interconnects in chiplet packages; used in Rebellions' REBEL-Q at 4 TB/s total bandwidth (3 TB/s per chiplet channel).
• Ultra Accelerator Link — Ultra Accelerator Link (UALink) is an emerging open standard for scale-up accelerator interconnects; OCP is developing manageability and security solutions for UALink-based heterogeneous AI clusters.
• Ultra Ethernet — Industry standard extension to Ethernet for AI backend networking, adding intelligent load balancing, lossy-network congestion management, selective acknowledgement, and fast failover for GPU cluster fabrics.
• User Space FTL — A Flash Translation Layer running in host user space rather than drive firmware, enabling dynamic tuning of IU sizes, write buffering, and coalescing without hardware changes.
• Vector Database Search — AI inference component performing similarity search over embedding vectors, requiring ultra-high IOPS fine-grain random access storage at 512-byte granularity.
• Verifiability and Automation — Verifiability and Automation is a predictive framework articulated by Andrej Karpathy for understanding which domains LLMs will automate fastest.
• Virtual Cell Model — A virtual cell model is a computational system that simulates what would occur inside a cell if specific genes, proteins, or other molecular actors were perturbed.
• Wiskott-Aldrich Syndrome — Rare immunodeficiency causing recurrent infections and severe bleedings, treated with hematopoietic stem cell gene therapy; approved in both the US and EU after Fondazione Telethon completed the full regulatory path as the first charity to do so.
• Workforce Readiness in Bioprocessing — The growing mismatch between bioprocessing talent supply and the increasingly complex demands of AI, novel modalities, and advanced manufacturing, identified as a critical bottleneck.
• Write Amplification Factor — Ratio of media writes to host writes on an SSD; the key endurance, energy, and performance metric FDP aims to reduce toward 1.0.
• ZeroFlap Optics — Credo's optical module architecture combining hardened optics, comprehensive DSP-level telemetry, in-band messaging, and autonomous link control to eliminate link flaps in AI cluster networks.
• libfabric — Open-source network API designed for connectionless, unordered operations; the API layer Ultra Ethernet builds upon, proven in HPC at Exascale and across multiple cloud network vendors.
• nvme-cli OCP Plugin — Open-source Linux command-line tool plugin enabling standardized NVMe SSD telemetry collection and human-readable decode per the OCP Data Center NVMe SSD specification.
• openSFI — Open Silicon Firmware Interface — OCP OPF specification defining a common abstraction layer between silicon initialization and host firmware (coreboot/UEFI) to eliminate platform fragmentation.
• siRNA — Double-stranded RNA molecules that silence gene expression by degrading target mRNA in the cytoplasm; the dominant modality for liver-targeting drugs delivered via GalNAc conjugates, with growing application in CNS via TFR1 targeting.

Synthesis (11)

• 2025 OCP Global Summit — Storage: Conference Intelligence — Cross-cutting intelligence from the 2025 OCP Global Summit Storage track: QLC flash reaches hyperscale, FDP drives write amplification toward 1.0, post-quantum security and OCP L.O.C.K. dominate the spec roadmap, GPU-initiated storage opens an IOPS-per-TCO SSD class, and telemetry, latency monitoring, and open test frameworks standardize fleet debuggability.
• Computex 2026 Conference Summary — Computex 2026 delivered the clearest signal the semiconductor industry has produced in a decade: agentic AI has crossed from research into commercial production, and every major chipmaker is reorganizing around it.
• Computex 2026 Net Impact: The 'So What?' Questions by AI Hardware Angle — What is the net impact of all announcements at Computex 2026? For each angle of AI hardware (chips, networking, power, cooling, optical interconnect, edge/physical AI, PCs, and system architecture), what are the "so what?" questions that infrastructure planners, executives, and investors should be asking right now?
• Contrarian Perspectives on the Computex 2026 Consensus: Where the Bull Case Is Weakest — Given the Computex 2026 consensus that agentic AI has arrived and is reshaping semiconductor economics, which contrarian perspectives hold the most water? Where is the consensus most likely wrong? What are the strongest counter-arguments to the bullish infrastructure buildout thesis?
• Danaher Summit 2026: The Next Era of Bioprocessing — Conference Intelligence — Cross-cutting intelligence from 14 Danaher Summit 2026 presentations covering cell & gene therapy manufacturing, nucleic acid drug targeting, digital bioprocessing, China's rise as an innovation hub, sustainability as a procurement gate, and the workforce crisis threatening bioprocessing's next era.
• GB300 NVL72 & HGX B300 NVL16: Hardware Implications, Sellability, and Ecosystem Impact — Jensen announced at Computex 2026 a platform with two key configurations: the GB300 NVL72 rack-scale solution with 72 Blackwell Ultra GPUs and 36 Grace CPUs delivering 1.1 exaflops of FP4 computation, and the HGX B300 NVL16 system offering 11x faster inference on large language models compared to Hopper-based systems.
• Going Into 2027: Biggest Challenges and Best-Positioned Companies — Going into 2027, what are the biggest challenges facing the data center industry and which companies are best positioned to reap the biggest rewards?
• OCP Global Summit 2025 — Executive Sessions: Conference Intelligence — Cross-cutting intelligence from 43 OCP Global Summit 2025 executive sessions — power as the binding constraint, the 800 VDC and liquid-cooling transition, three-tier open-Ethernet networking, co-packaged optics, and open firmware as the competitive moat.
• Orbital Data Centers: Timeline, Technology Obsolescence, and the Future of Terrestrial Infrastructure — What is the realistic timeline for orbital data centers? Which existing terrestrial data center technologies will they obsolete? Will terrestrial data centers even remain relevant?
• Publicly Traded Companies Positioned for AI Infrastructure Growth (2027-2029) — Which publicly traded companies stand to benefit most over the next three years (2027-2029) based on AI data center infrastructure trends documented in the wiki?
• Technical Challenges at Every Layer of AI Data Centers for 2027 — What are the top technical challenges at all layers of AI data centers heading into 2027?

Documents (305)

• 51.2T CPO Switch Practice And Challenge — MICAS Networks presents their next-generation 51.2 Tbps Co-Packaged Optics (CPO) switch system built on Broadcom's Bailly CPO Tomahawk 5 switch ASIC.
• 800G: Empowering Open Platforms for Unprecedented Speed and Agility — Celestica presents the DS5000, a 64x800G (51.2 Tbps) Ethernet switch being contributed to OCP as a design specification, seeking OCP Inspired status.
• A Look Inside Prometheus: Designing Meta's Large AI Cluster Deployments — Meta's Prometheus is a gigawatt-scale AI accelerator cluster that integrates GPU platforms, storage, compute, and networking into a unified open infrastructure.
• A Peek Into the Future: Results from PCIe Gen 6 SSD Simulations — FADU, an SSD controller company, presents simulation results showing PCIe Gen 6 SSDs can achieve 6.9 million IOPS per drive at 4KB random read, closely approaching the theoretical maximum for the interface.
• A SONiC-based Open Packet Broker for 5G Monitoring — NVIDIA and Aviz Networks present a SONiC-based Open Packet Broker (OPB) for 5G telco monitoring and enterprise security use cases.
• ACS Door HX sub-project update — Juan Carlos (Retal) and Jabari George present the annual update for the OCP Door Heat Exchanger (Door HX) sub-project, part of the Advanced Cooling Facilities (ACS/ACF) work stream.
• AI Across The Sciences — This wide-ranging panel examines how AI is transforming three non-biological sciences — materials discovery, mathematics, and robotics — and considers what each field's progress reveals about the broader trajectory of AI-for-science.
• AI Packet Trimming — Packet trimming is a switch hardware feature designed to accelerate loss notification and recovery in AI fabric networks where every packet drop carries a significant cost.
• AI Requirements Driving Liquid Cooling Design — CoolIT Systems presents the technical requirements driving liquid cooling design evolution, from cold plate thermal management to CDU scaling.
• AI at The Bench: Getting More From Each Experiment — Jeremy Elser, founding CEO of Shipthesis (a biologics biotech) and forward deployed engineer (FDE) at Palantir, demonstrates how his lab implemented AI-powered data capture, experiment management, and protocol generation using Palantir Foundry.
• AI-Native Networking For AI Factories — Juniper Networks, now part of HPE, presents its AI-native networking strategy designed from the ground up to address the unique demands of AI workloads -- massive elephant flows, extreme sensitivity to packet loss (which causes large tail latencies and GPU performance degradation), and order-of-magnitude higher bandwidth requirements versus traditional workloads.
• ALPINE SONiC Switchstack Simulation — Google presents ALPINE, a virtual SONiC switch with a software data plane designed to left-shift SONiC switch stack development by catching issues earlier in the development cycle and reducing dependence on expensive hardware test beds.
• AVS for Vendor Sim and Topologies with Broadcom, Arista — This talk presents the Alpine Virtual Switch (AVS), Google's next-generation SONiC virtual switch framework that supports pluggable high-fidelity data plane implementations.
• Accelerated AI With P4 DPUs: Next-Level Flexibility, Performance & Scale on AMD AI NICs — AMD presents its Pensando Pollara 400 AI NIC as a unified scale-out networking platform delivering 10% better RoCEv2 performance than competitors and 25% better performance with UEC-RDMA versus RoCEv2, while enabling up to 50% lower fabric cost through multi-plane networking that eliminates the core switch tier.
• Accelerating AI Hardware NPI Cluster-Less Validation of GPUs and Networking — This talk presents "one-arm" or "cluster-less" validation methodology for AI accelerators and NICs that avoids the need for a full cluster test bed during New Product Introduction (NPI).
• Accelerating AI Infrastructure Through Modular Building Blocks presented by Supermicro — Johnson Jia (Supermicro) presents the company's modular AI infrastructure strategy at OCP Global Summit 2025, covering the AI hype/reality gap, a distillation-based entry framework, and the DCBBS portfolio spanning training superclusters to PCIe inference servers.
• Accelerating AI Innovation With Intel Xeon — This joint Lenovo-Intel presentation covers how Intel Xeon 6 CPUs serve as critical host processors in GPU-based AI servers, alongside Lenovo's vertically integrated manufacturing and supply chain capabilities for rapid deployment.
• Accurate Cable Length Measurement for Optimized Data Center Networking — This presentation argues that cable length measurement has become a first-class operational concern in AI data center networking, where it was historically ignored.
• Advanced Datacenter AC/DC Distribution and Conversion Power Architecture — This panel session features five presentations addressing the fundamental challenge of delivering power to AI data center racks as densities scale from hundreds of kilowatts to megawatts per rack.
• Advanced Photonic Technologies for Future Systems: Optical Computing or Optics for Computing? — Bert Jan Offrein of IBM Research Europe frames the core question facing the photonics industry: is optics best deployed as a communications medium within computing systems ("optics for computing") or as the computational substrate itself ("optical computing")?
• Advanced-Modalities-And-The-Decisions-That-Define-Them — This panel discussion from the Danaher Summit 2026 brings together four industry leaders — Campbell Bunce (Abzena), Hendrik Dietz (CBTX), Mike Henney (Bicycle Therapeutics), and Dan McGrath (Sarepta) — to examine the manufacturing, regulatory, and supply chain challenges of advanced modalities including in vivo CAR-T therapy, radiopharmaceuticals, and novel nucleic acid delivery systems.
• Advancements and Community Growth in OCP Immersion Cooling — Rolf Brink, serving as OCP (Open Compute Project) immersion project lead and founder of Promersion, presents a dual-track keynote: first analyzing liquid cooling market dynamics and then unveiling a new governance model for the OCP Immersion Cooling project.
• Advancing AI With Smarter Cooling And Sustainable Innovation — Dell and NVIDIA jointly present the evolution of GPU rack power and performance from Kepler through Blackwell Ultra, showing 2.6x more power per rack (from ~40-60 kW to ~120 kW) but 41x the performance, achieving an exaFLOP in a single rack.
• Advancing Immersion Cooling: Project Updates, Industry Trends and Consensus Glossary — This session opened the immersion cooling track at OCP Global Summit 2023 with the largest-ever immersion content program at OCP — approximately 50 topics across a full day.
• Advancing OCP Telemetry Standards – Insights from a Drive Vendor — Kevin Brandt, Systems Architect at Micron Technology, details Micron's real-world implementation of OCP Data Center NVMe SSD telemetry standards, tracing the evolution from proprietary binary logs to the open-source nvme-cli OCP Plugin.
• Alert Fatigue in Hyperscale Environments: A Metrics-Based Approach to Signal Tuning in Open Networking — Alert fatigue in hyperscale networks occurs when a single physical failure — such as one ToR port going down — cascades into thousands of correlated alerts across server, ToR, fabric, spine, and resource hub layers simultaneously.
• Alibaba HPN: A Data Center Network for Large Language Model Training — Alibaba Cloud presents HPN 7.0, their seventh-generation High Performance Network architecture purpose-built for large language model training clusters.
• Alibaba's Ethernet Based DC Deployment for AI/ML Workloads Using Merchant Silicon — Alibaba and Broadcom jointly present Alibaba's approach to building large-scale AI/ML training cluster networks using Ethernet-based merchant silicon and open networking stacks.
• Aluminum Micro-Channel Coils for Door HX Solutions — This presentation evaluates aluminum microchannel (ALMC) heat exchangers as a challenger to the incumbent copper tube-and-fin (T&F) construction for rear door heat exchangers (RDHx) in data centers.
• Analysis of Network Traffic Patterns for Collective Communication — Meta engineers present a detailed empirical analysis of how all-to-all collective communication creates unexpected and problematic network traffic patterns in AI training clusters.
• Andrej Karpathy: From Vibe Coding to Agentic Engineering — Andrej Karpathy, in conversation with Sequoia partner Stephanie Zhan at AI Ascent 2026, lays out a framework for understanding the shift from "vibe coding" (his own coinage from 2025) to what he now calls "agentic engineering" — a serious engineering discipline for coordinating fallible, stochastic, but extremely powerful AI agents without sacrificing professional software quality.
• Architecting the Datacenter Requirements for Scaling AI Clusters — This panel discussion at OCP EMEA Summit 2025 (Dublin, 29–30 April) brought together five industry leaders to debate the practical engineering challenges of scaling AI clusters with liquid cooling.
• Are-You-Ready-For-Digital-Bioprocessing — A panel discussion at the Danaher Summit 2026 on The Next Era of Bioprocessing, featuring Andrew Heron (CEO of Portal Biotech) and Michelangelo Canzaneri (Merck KGaA) exploring why digital bioprocessing has underdelivered despite a decade of promises.
• Astera Labs: Scaling AI With PCIe, Ethernet, And UALink Retimers — Astera Labs presents its vision for AI Infrastructure 2.0 where the rack replaces the server as the unit of compute, with clusters of 32, 64, 128+ GPUs operating as a single high-performance system.
• AstraZeneca's AI Playbook: Jorge Reis-Filho With Michelle Lee — Jorge Reis-Filho, Chief of AI for Science Innovation at AstraZeneca, describes the company's comprehensive strategy for AI deployment across drug discovery and clinical development in a fireside chat with Medra CEO Michelle Lee at UNLOCK 2026.
• Axiado's Hardware-Rooted AI Agents Reimagine Security And System Management in The Datacenter — Axiado, founded in 2019 and headquartered in San Jose (~160 employees), presents the AX3080 Trusted Control/Compute Unit (TCU) -- a system-in-package that integrates BMC, Root of Trust, TPM, AI accelerators, NIC, LPDDR, BIOS Flash, BMC Flash, and eMMC into a single module that replaces the entire discrete DC-SCM card.
• Back to the Future: Leveraging Classical Compute to Make the Quantum Leap (presented by Dell) — A Dell Technologies presenter at the 2025 OCP Global Summit provides a comprehensive introduction to quantum computing for data center professionals, arguing that QPUs should be treated as accelerators augmenting classical infrastructure.
• Balancing Innovation, Performance and Safety: The Evolution of Immersion Requirements — This session presents the OCP Immersion Requirements workstream as the consolidating hub for immersion cooling guidance, occupying a unique position in the OCP community pipeline: it synthesizes outputs from early-pipeline groups (e.g., Immersion Challenges) and mid-pipeline technical workstreams (e.g., Fluids & Materials, Signal Integrity) into a holistic, systems-level best practices document.
• Be Ready for CPO: Integrating and Enhancing CPO Switches with SONiC — NTT presents the software challenges of supporting Co-Packaged Optics (CPO) switches in SONiC, drawing from their experience building an OIF IA-compliant CPO switch prototype.
• Best Practices for Liquid & Air Cooling of a 51.2Tbps Switch for High-Density AI Clusters — Broadcom and Alibaba present a joint perspective on cooling 51.2 Tbps switches for high-density AI clusters, covering both silicon-level power optimization and system-level thermal design for air and liquid cooling.
• Beyond SSD: SK Hynix AIN Family — Redefining Storage as the Core Enabler of AI at Scale — CS Kim, SVP of eSSD PD at SK hynix, presented the AIN Family at OCP Global Summit 2025, arguing that AI infrastructure has entered an era where token throughput — not compute FLOPs — is the primary bottleneck.
• Beyond the Rack: Sovereign AI via European Pluggables — Magnus Olson, CTO of Estel — a Belgian-headquartered startup founded roughly two years prior to the talk — presents the case for a European-sovereign optical transceiver manufacturing ecosystem.
• Bio-based Fluids for European AI Data Centres — This presentation argues that traditional multi-loop data center cooling architectures waste 30–40% of energy through unnecessary thermal cycling and intermediate heat transfer steps, and that a single-loop design using a bio-based propylene glycol fluid (Bio PDO25) can eliminate these losses while satisfying European heat reuse legislation.
• Boost your air-cooled server performance with an Industrial Two-Phase Passive Cooling Solution — Calyos and Valeo jointly present a passive two-phase loop heat pipe technology for direct-to-chip cooling in air-cooled servers.
• Breaking The Power Wall With Co-Packaged & Liquid Now! — Micas Networks and Super Micro present a joint solution addressing three critical AI data center challenges -- power, thermal management/liquid cooling, and networking scalability -- through co-packaged optics (CPO) switches with integrated liquid cooling.
• Breaking through the Network Wall with High Radix Switching — Mike Capuano, Chief Business Development and Marketing Officer at Eridu (originally referenced as "Arrcus" in speech but title slide shows "Eridu"), presents the case for a 10x radix increase in AI networking silicon.
• Build Limitless Network with Smart Switch — Microsoft proposes a Smart Switch architecture that integrates a 12.8 terabit switching ASIC with 8 DPUs into a single 2RU chassis, creating a powerful disaggregated network appliance for data center deployments.
• Building the Future of AI Data Centers with Open Systems presented by AMD — AMD presents its open-systems AI data center networking strategy at OCP Global Summit 2025, detailing the Pensando Pollara 400 AI NIC, multi-plane networking TCO, and the next-generation Vulcano 800G NIC for million-GPU-scale clusters.
• Building the Future of AI: Wiwynn Accelerates the Shift to AI-Driven Infrastructure — Wiwynn presents its strategic position in the AI infrastructure buildout at OCP Global Summit 2025, covering the evolution from Transformer-driven compute demand through rack ODM to full AI cluster deployment, with case study of the YTL Malaysia Level 12 deployment.
• Building, Extending, & Customizing Foundation Models: FoxBrain's Journey to Industrial Frontier — Foxconn's AI Research Center presented FoxBrain, their in-house large language model, at the OCP Global Summit 2025 in San Jose.
• Building-Organizations-That-Keep-Pace-With-Change — This panel discussion from the Danaher Summit 2026 features CEO John Cox of Dine Therapeutics and Dr.
• Building-The-Next-Generation-of-Bioprocessing-Capability — This panel discussion from the Danaher Summit 2026 brings together Dan Bracewell (UCL), Steve Bates (UK government life sciences policy), and Alistair (Faculty AI) to examine whether the bioprocessing workforce can keep pace with accelerating scientific and technological change.
• COMPUTEX 2026 Arm CEO Keynote: The Foundation for the Agentic AI Era — Arm CEO Renee Haas delivered a 49-minute keynote at Computex 2026 in Taipei, framing the rise of agentic AI as the defining inflection point for compute architecture.
• COMPUTEX 2026 CEO Keynote: Marvell — Connectivity is the Next Defining Challenge of AI — Marvell CEO Matt Murphy delivered the opening Day 1 keynote at Computex 2026, presenting a comprehensive case that connectivity — not compute or memory — is now the defining bottleneck and growth opportunity in AI infrastructure.
• COMPUTEX 2026 CEO Keynote: NXP — Physical AI and the Neural Axis Architecture — NXP Semiconductors CEO Rafael Sotoayo delivered a 71-minute keynote at Computex 2026 Taipei framing the company's strategic position in physical AI — the deployment of AI into real-world autonomous systems including drones, software-defined vehicles (SDVs), and humanoid robots.
• COMPUTEX 2026 CEO Keynote: Qualcomm — Qualcomm CEO Cristiano Amon delivered the opening keynote at Computex 2026 in Taipei, presenting his thesis that 2026 is "the year of the agents" — a structural inflection where AI evolves from conversational tool to autonomous actor.
• CPO & LPO: The Next Leap in Network Evolution presented by Micas Networks — This OCP Global Summit 2025 executive panel, moderated by a Micas Networks representative and featuring SemiAnalysis, Micas Networks, and Multilane, examines the state of Co-Packaged Optics (CPO) and Linear-Drive Pluggable Optics (LPO) as next-generation networking technologies for AI data centers.
• Capacity,-Scale-And-The-Next-Chapter-For-Established-Biologics — A panel discussion at the Danaher Summit 2026 exploring capacity planning and scale-up decisions for established biologics, featuring Silka Maul (Roche MSAT), Ronda Duffy (Biocon Biologics COO), Tar Samad (Lundbeck Research), and John Matulsky (independent consultant).
• Chiplets at Scale: How Arm And OCP Are Powering The Next Wave of Modular Innovation — Arm presents its strategy for accelerating multi-party chiplet development through the Arm Total Design (ATD) program, now grown to 36 active partners (from 12 at launch two years ago), and announces the contribution of a Foundational Chiplet System Architecture (FCSA) specification to OCP.
• Co-Designing for AI: Open Standards and Ecosystems Driving the Future of Infrastructure — At OCP Global Summit 2025 in San Jose, Rebellions hardware lead Wonguk Kim and Samsung Foundry's Thomas Mermans jointly presented the co-design strategy behind their AI chiplet ecosystem.
• Cold Plate Cooling Loop Requirements Document - Rev 2 — This presentation introduces Cold Plate Cooling Loop Requirements Document Revision 2 — published to the OCP contributions portal on the Friday before OCP Global Summit 2023.
• Coldplate Base Specification — This presentation introduces the initial release of the OCP Coldplate Base Specification — a vendor-neutral, de-facto industry baseline for single-phase direct-to-chip (D2C) cold plates.
• Conceptualizing Next-Generation Memory & Storage Optimized For AI Inference — SK hynix presents a conceptual framework for how AI inference is fundamentally changing memory and storage requirements in two key ways.
• Concurrent Maintainability Considerations for Liquid-Cooled Deployments — This presentation, part of the OCP Advanced Cooling Facilities (ACF) work stream, addresses the practical challenge of achieving concurrent maintainability (CM) in liquid-cooled data center deployments.
• Cooling Environments Update and Cold Plate Subproject Overview — This talk opens the Cooling Environments half-day track at OCP Global Summit 2023, delivered jointly by Steve Mills (Meta, OCP Incubation Committee) standing in for absent project lead Sean Sivapalan, and Jordan Johnson (Intel), cold plate sub-project lead.
• Cooling Environments: Industry Analytics, Project Progress and Cold Plate Developments — This opening keynote for the OCP EMEA Summit 2026 Cooling Environments track (Barcelona, April 29–30) covers two major segments: (1) Promersion's industry analytics on the global data center and liquid cooling market trajectory, and (2) a structured status update from Meta's Steve Mills on the five subprojects within the OCP Cooling Environments project — Door Heat Exchanger, Heat Reuse, Immersion, CDU, and Cold Plate.
• Customer-Driven Advancements in Power And Cabinet Infrastructure — Legrand presents its data center white space product suite spanning Starline busway, power distribution, cabinet PDUs, OCP-compatible cabinets, and OCP power shelves.
• DENT SAI Integration — The DENT project, a Linux Foundation initiative focused on enterprise and edge networking, is integrating SAI (Switch Abstraction Interface) as an alternative hardware abstraction layer alongside its existing SwitchDev approach.
• Data Center Facility Project Highlights — This session provides a comprehensive overview of all active and newly launched work streams within the OCP Data Center Facilities (DCF) group as of the 2025 Global Summit.
• Data Center In-Row Cooling Integrated Solutions — Delta Electronics, founded in 1971, presents its comprehensive data center cooling portfolio spanning in-row cooling (room cooling, chiller water, direct expansion, rear door heat exchangers), liquid-to-air CDUs (300 kW featured unit on display at booth 51), liquid-to-liquid CDUs, and single-phase immersion cooling.
• Data Center One-Stop Liquid Cooling Solution Presented by TCT — TCT (Thermal Control Technology), founded in 2018 and headquartered in Dongguan, China, presents its full-stack liquid cooling solution portfolio spanning component-level products (cold plates, QDs, valves) through module-level products (manifolds, rear-door heat exchangers, rack-based CDUs) to floor-based CDUs and secondary cooling networks.
• Data Center Sustainability & Decommissioning -- The Ugly Truth — This panel session argues that data center sustainability is overwhelmingly treated as a marketing term rather than an operational discipline.
• Data Engine For Autonomous Science: Michelle Lee, Medra — Michelle Lee, founder and CEO of Medra, presents the case for "physical AI scientists" as the missing infrastructure for autonomous science.
• Dealmaking & Partnerships: Who's Actually Writing The Checks in AI x Pharma — This panel brings together five leaders to examine how pharma-AI deals are actually being structured in 2026, covering perspectives from a frontier AI lab (Anthropic), big pharma business development (AbbVie, Johnson & Johnson), and an AI-first biotech (Terra Therapeutics).
• Deep Dive Into X-Cell: Ci Chu And Bo Wang, Xaira Therapeutics — Ci Chu and Bo Wang from Xaira Therapeutics present a detailed technical deep dive into X-Cell, Xaira's virtual cell model designed to predict how cells respond to perturbations.
• Delivering Tailored AI Infrastructure: Pegatron's Rack Innovation and Manufacturing Expertise — At OCP Global Summit 2025, Pegatron SVR presented its end-to-end AI infrastructure capabilities, from board-level design and smart manufacturing to full rack-scale liquid-cooled cluster deployment.
• Delta 800 VDC Power and Cooling Solutions for Data Centers presented by Delta — At the 2025 OCP Global Summit, Delta Electronics presented its 800 VDC power and cooling strategy for next-generation AI data centers, co-developed with NVIDIA, Meta, Microsoft, and Google.
• Densifying Modular Data Centers with Hybrid Cooling — This presentation documents a collaborative OCP effort to dramatically increase the power density of modular data centers (MDCs) by combining direct-to-chip cold plate cooling with rear door heat exchangers (RDHx) — a hybrid approach.
• Deploying Heterogeneous Hardware in the Era of Agentic AI presented by Intel — Intel VP Saurabh Kulkarni presented Intel's heterogeneous hardware strategy for agentic AI at OCP Global Summit 2025, arguing that ~80% of data center AI compute will shift to inference by 2028 and that hardware diversity is the foundational architecture for next-generation AI data centers.
• Deploying QLC at Scale: Balancing Density, Power, Performance, And Reliability — Pure Storage makes the case that QLC flash is the optimal balanced storage technology for emerging AI workloads, offering 150 TB drives today with 600 TB expected within 12-18 months, compared to ~50 TB for HAMR HDDs and 30 TB for TLC.
• Deploying and scaling QLC Flash at Meta — At OCP Global Summit 2025, Meta Hardware Systems Engineers Madhavan Ravi and Jesse Koh announced Meta's first QLC flash storage server: a 2U ORv3 chassis with 24 front-accessible U.2-long drive slots supporting both standard NVMe SSDs and Pure Storage Direct Flash Modules.
• Deterministic ISSU Qualification — ISSU (In-Service Software Upgrade) — also called warm upgrade or warm reboot in SONiC — allows network switches to upgrade their software without dropping a single packet.
• Direct-to-Plug Liquid Cooling For Pluggable Optical Modules — Ciena presents a direct-to-plug (D2P) liquid cooling solution for OSFP pluggable optical modules that integrates a cold plate directly into the module body, eliminating the dry sliding thermal interface that limits both riding heat sink (air) and riding cold plate (liquid) approaches.
• Door Heat Exchanger Sub-Project Updates — Jabari George, sub-project lead for the OCP Door Heat Exchanger (DoE HX) sub-project, presents a comprehensive status update covering four active workstreams, community engagement history, and strategic direction for door heat exchangers as data center cooling technology evolves.
• Drive Regeneration in Action: Enhancing Fault Tolerance in Datacenters — Curtis Stevens of Seagate Technology presents drive regeneration techniques at OCP Global Summit 2025, arguing that single-head failures — the dominant real failure mode in modern 20-head HDDs — can be resolved without replacing the drive.
• Driving AI at Scale with 1.6T Networking: How Open 100T Switches Will Redefine Data Centers in 2026 and Beyond — Celestica VP Tareq Bustami and Broadcom Head of Software Products Hasan Siraj presented at OCP Global Summit 2025 on the evolution of AI networking to the industry's first 100 terabit switch, announcing Tomahawk 6, the Thor Ultra 800G NIC, Celestica DS6000/DS6001, and the OCP ESUN work stream for open AI scale-up networking.
• Driving Innovation in Data Center Computing and ASIC Technologies presented by MediaTek — MediaTek Senior Marketing Director C.K. Peng presented the company's ASIC and data center technology strategy at OCP Global Summit 2025, covering the explosive growth of the custom AI accelerator market, the Moore's Law gap, and MediaTek's roadmap across 2nm/A16 process, 3.5D packaging, Custom HBM, and Co-Packaged Optics.
• Driving Open Innovation at Scale: AMI's Strategic Role in the Open-Source Firmware Ecosystem — AMI provides foundational platform firmware technology for the compute ecosystem, encompassing three major components: UEFI/BIOS boot firmware (Aptio), baseboard management controller firmware (MegaRAC via OpenBMC), and hardware root of trust security (Tektagon via Cerberus).
• Driving the Open Data Center Ecosystem presented by Google — Google VP/Fellow Partha Ranganathan presented Google's strategy for driving the open data center ecosystem at OCP Global Summit 2025, covering 70+ contributions across fungible power delivery, liquid cooling, DC telemetry, sustainability, scalable networking, and open security, framed by explosive AI token demand growth.
• Dual ToR Evolution: Active-Active ToR Deep Dive — Microsoft presents the evolution of dual Top-of-Rack (ToR) technology in SONiC, progressing from single ToR to active-standby dual ToR and now to active-active dual ToR.
• Efficiently Powering The Next Wave of Agentic Reasoning Applications: Making Gen AI Attainable — d-Matrix presents its Digital In-Memory Compute (DIMC) architecture as a purpose-built solution for AI inference, arguing that the GPU-dominated training era is giving way to an inference era where latency, opex, and multimodality matter more than raw capex-driven compute.
• Embracing AI Infrastructure and Advanced Thermal Solutions presented by Auras — Sean Lim, Manager at Auras Technology, presented at OCP Global Summit 2025 on the escalating thermal challenges of AI infrastructure and Auras's complete liquid cooling ecosystem, spanning next-generation cold plates, a 300 kW liquid-cooled PSU, a 100 kW HRU, and a 1.6 MW in-row CDU achieving PUE below 1.1.
• Empowering AI at Scale with Lenovo Neptune® Liquid Cooling presented by Lenovo — A panel of four experts from Lenovo, NVIDIA, and Nscale at OCP Global Summit 2025 explored how Lenovo Neptune liquid cooling enables AI infrastructure at scale, covering Neptune's 45°C distilled water approach, 80% efficiency, and 40% energy savings, and forecasting rack power trending from 130 kW toward 1 MW.
• Empowering Scale-Up and Scale-Out Networking With Samtec CPX Architecture: CPC & CPO — Samtec presents its CPX (Co-Package Interconnect) architecture as a unified form factor that supports both co-packaged copper (CPC) and co-packaged optics (CPO) through the same socket on the same chip package.
• Empowering Software-defined Memory Solutions with Genuine CXL Pooling Technology — This three-company joint presentation showcases a production-ready CXL memory pooling solution combining Samsung's CMM-D (CXL Memory Module with DRAM) memory expanders, H3 Platform's Falcon C5022 CXL switch-based memory appliance, and Montage Technology's CXL Memory eXpander Controller (MXC) silicon.
• Empowering The Future of AI Connectivity With Alpha Networks' 1.6T Innovation — Alpha Networks presents its 1.6 Tbps Ethernet switch designed for AI backend networks, built on the Broadcom Tomahawk 6 ASIC with 64 ports of 1.6T OSFP interfaces and liquid cooling.
• Enabling AI Applications With Efficient And Sustainable Power Distribution — Rittal presents its approach to power distribution for AI data centers, focusing on copper consumption reduction through architectural choices and its RiLineX busbar system.
• Enabling AI at Scale with 1.6T and Beyond: Energy-Efficient DSP, LPO, LRO, and CPO Optics — Tiger Ninomiya of Accelink Technologies presented at OCP Global Summit 2025 on optical transceiver technologies for AI-scale interconnects, covering DSP, LPO, LRO, and Co-Packaged Optics.
• Enabling High-Voltage Distribution in Data Centers Through Next-Generation Connectors — Amphenol presents the interconnect challenges of transitioning data center power architectures from legacy 12/48V DC to high-voltage DC (HVDC) at +/-400V (800V) and beyond.
• Enabling Liquid Cooling on a Global Scale — Microsoft presents its strategy for deploying liquid cooling globally across Azure's 70+ regions by developing a next-generation Heat Exchange Unit (HXU) that can be deployed in existing air-cooled data centers without facility modifications.
• Enabling Next Gen AI Infrastructure with 224G+ Cable Cartridge presented by Luxshare-Tech — Robbie Xu of Luxshare-Tech presented the Intrepid NEXUS 224G+ cable backplane and cable cartridge solution at OCP Global Summit 2025, demonstrating 24.13 dB bump-to-bump at 53 GHz.
• Enabling Scalable Infrastructure: Alibaba's High-Performance 102.4T Ethernet Switch — This talk presents Alibaba's HPN 8.0 (High Performance Network 8.0), the next-generation AI fabric architecture centered on a 102.4T Ethernet switch, and the detailed hardware design of two switch SKUs — "Simba" (128×800G, code name) and "Kimba" (64×1.6T, code name).
• Energy-Efficient Optical Interconnects for Multi-Rack AI Pods — Fotini Karinou from Microsoft's Azure Hardware Systems and Infrastructure organization presented the hyperscaler's perspective on optical interconnects for AI scale-up networking.
• Energy-Efficient VCSEL-based Transceivers for AI Datacenters — Hideyuki Nasu, a Fellow at Furukawa Electric, presents the case for VCSEL (Vertical-Cavity Surface-Emitting Laser) technology as the energy-efficient optical interconnect for AI data center scale-up networks, arguing it occupies a compelling sweet spot between copper cables (short reach, high reliability) and silicon photonics (long reach, higher power).
• Envicool Full Chain Liquid Cooling 2GW+ Deployment Case Study — Ning Liu of Envicool presented the company's full-chain liquid cooling strategy and 2 GW+ deployment track record at OCP Global Summit 2025, covering CDUs from 300 kW to 2.5 MW and the industry shift to mandatory liquid cooling.
• European 1MW Rack - HDINI — The H-DINI project (High Density Infrastructure for Next-generation Infrastructure) is a Vinnova (Sweden's Innovation Agency)-funded research initiative led by RISE (Research Institutes of Sweden) with a broad consortium of Swedish and European industrial partners.
• Evolution of Ethernet-Based Switch Platforms and Fabrics to Meet Meta's AI Training Clusters — This talk describes Meta's Non-Scheduled Fabric (NSF) architecture for AI training networks and introduces the Minipack3N platform — the first Meta switch using Nvidia's Spectrum 4 ASIC alongside the existing Minipack3 (Broadcom Tomahawk 5) and Kodiak3 (Cisco G200) platforms.
• Evolving the Infrastructure Stack: Rack-Scale Designs, Open Firmware, and Intelligent Sustainability — AMI's Chief Product Officer Zach Julie (substituting for CEO Sanjoy Maity) presents AMI's perspective on three intersecting forces reshaping digital infrastructure: rack-scale architecture, open-source firmware, and intelligent sustainability.
• Expanding Horizons: Data Center Market Outlook And MediaTek's Leading AI ASIC Technologies — MediaTek presents its data center ASIC capabilities, highlighting the growing complexity of AI accelerator silicon and interconnects.
• Experiences in Designing a Better NVMe Controller — Tim Amundsen (ScaleFlux) and Matt Bromage (ARM) present their collaborative development of next-generation NVMe SoCs at OCP Global Summit 2025.
• Experimental seismic performance for liquid immersion cooling system — Noemi Bonessio (METAseismic) presented the results of the first seismic certification test of a liquid immersion cooling unit, conducted in collaboration with Hyper-tech, Sub-mere, and Wyoming Hyperscale as OCP community partners.
• Exploring the Potential of Single-Phase Immersion Cooling Through Quantitative Thermal Performance — This presentation delivers the most quantitatively dense fluid benchmarking dataset presented at OCP 2024, covering 8 commercial immersion fluids tested across a comprehensive parametric matrix on a custom-built 2U test stand.
• Expo Session Presented by Auras Technology — Auras Technology presents its comprehensive liquid cooling solutions for AI data center infrastructure, covering cold plates, CDUs, rack manifolds, quick connectors, and complete thermal management systems.
• Extending Copper Interconnects to 448G: Innovations Overcoming Conventional Limitations — Luxshare presents measured results demonstrating that 448 Gbps PAM4 over copper interconnects is achievable through improvements in three key areas: the CPC (Co-Packaged Copper) interface using Coolio elastomeric technology, front-panel IO using a new OSFP-like elastomeric connector, and twinax cable design.
• Fabric Resiliency at Scale — This presentation demonstrates why global adaptive routing (Global AR) is essential for maintaining AI training performance under fabric link failures.
• Flash in Fleet — Meta engineers Vineet Parekh and Minh Hoang presented two complementary advances in Meta's fleet-scale NVMe SSD quality program at OCP Global Summit 2025.
• Forging Ahead In Immersion Cooling: A Hands-On Approach To Validating Optical Interfaces — This presentation introduces the newly formed OCP Immersion Component Validation workstream and describes its scope, governance, and test methodology, with optical cable interfaces as the initial focus.
• Frame Based Deployment of Rear Door Heat Exchanger — This presentation proposes decoupling the rear door heat exchanger (RDHx) from the IT rack via a permanent facility-mounted frame structure, enabling deferred capital expenditure and modular deployment.
• From Complexity to Efficiency: Open Hardware Strategy Through Collaboration — ASRock Rack and Scaleway co-present their collaboration on a 4U 12-node ORV3-compatible server system running OpenBMC.
• From Concept to Capability: Innovating The Future of OCP Reference Designs — Vertiv's CTO office presents its innovation process for developing OCP-compatible reference designs for power and cooling infrastructure.
• From Edge to Trillion-Token AI: ASUS Infrastructure For Every Scale — ASUS presents its comprehensive AI infrastructure portfolio spanning edge to data center scale, with over 20 years of hardware experience evolving from board-level to system-level to rack-scale solutions.
• From Fans to Fluids: The Critical Need For Liquid Cooling in AI Data Centers — Bobby Kintol, a self-described "liquid cooling super fan" with experience dating to 2006 (half-megawatt containerized data centers), presents the case for liquid cooling as the standard for AI data centers.
• From Vision to Execution: The Story of Alibaba's 51.2T Switch — Alibaba Cloud presents Tigatron, its home-built 51.2 terabit fabric switch designed as the cornerstone of its HPN 7.0 (High Performance Network) architecture for AI/ML training clusters.
• Fungible Hardware Tables as an Ordered Chain for Traffic Engineering and AI/ML Applications — Meta and Broadcom present a collaborative solution to traffic engineering challenges in Meta's backend AI/ML training network.
• Future Proofed DLC Piping Infrastructure — Mark Bulmer of Georg Fischer (GF) Flow Solutions presents the case for high-performance polymer piping as a future-proof DLC (direct liquid cooling) infrastructure material, challenging the default choice of stainless steel.
• Future-Ready AI Clusters: Liquid-Cooled, Leak-Secure, and OCP-Optimized presented by MiTAC — MiTAC Computing Technology and AMD jointly presented liquid-cooled, OCP-compliant AI cluster solutions at OCP Global Summit 2025, covering AMD's Instinct roadmap through MI500 and MiTAC's 64-GPU in-rack DLC solution.
• Guidelines and Base Specification for Dielectric Fluid-Based Two-Phase Cold Plate Heat Transfer Fluid — This presentation introduces two OCP documents from the Cooling Environments — Cold Plate workstream: (1) the Guidelines document (already published and available) and (2) the Base Specification (submitted and in review at time of presentation, planned publication within weeks).
• Hardware Management for Liquid Cooling: Enabling Efficient Operations in Immersion Solutions — This presentation reported on the OCP Hardware Management workstream's completion of a unified Redfish data model for liquid cooling equipment, published by the DMTF as Redfish Release 2023.1 in April 2023 and subsequently updated in Release 2023.2.
• Harmonizing Open Compute Infrastructure with Global Standards — This presentation reports on the activities of the OCP Cooling Environments Industry Liaison Team — the group responsible for bridging OCP's liquid cooling technical work with the external standards ecosystem.
• Harness AI Opinions with Sensor Facts - sensorCFD AI — AKCP (est. USA 1981), self-described as the world's oldest and largest company making environmental monitoring sensors, presents "Quicklime" — an AI-enabled DCIM (Data Center Infrastructure Management) platform that combines sensor-backed Computational Fluid Dynamics (sensorCFD) with a suite of AI optimization modules.
• Harnessing Time Coordinated Computing (TCC) to Enable the Next Generation of Applications from the Cloud to the Edge — Intel's Time Coordinated Computing (TCC) is a comprehensive set of platform features ensuring low-latency, real-time compute performance by solving jitter and latency issues on commodity hardware.
• Heat Reuse Outlook — This presentation provided an annual update from the OCP Heat Reuse sub-project, the youngest sub-project under the Cooling Environments umbrella, now approximately two years old.
• High-Voltage Connectivity, Power & Cooling for the AI Data Center — TE Connectivity presents its "whip to chip" power connectivity portfolio for AI data centers, addressing the evolution from HPR v1/v2 (up to 200 kW, power/GPU/CPU in single rack) through HPRv3 Side Power Rack (approximately 300 kW, separate power and compute racks at 50V) to HPRv4 High Voltage Rack (over 1 MW at +/-400V or 800V, with sidecar power).
• How Colos and Hyperscalers Co-Evolve — Google presents a call to action for deeper technical co-design between hyperscalers and colocation providers, moving beyond the current "commercial co-design" of wholesale data centers to address the "Cambrian explosion" of hardware heterogeneity in AI infrastructure.
• How Will Copper Drive High-Speed Interconnect Solutions Fueling Next-Generation AI Data Centers — BizLink Group presents comprehensive R&D results on pushing copper interconnect performance toward 448 Gbps per lane and beyond, demonstrating that copper remains viable for next-generation AI data center interconnects through advances in cable gauge miniaturization (down to 39 AWG), higher bandwidth designs (past 100 GHz), co-packaged copper solutions, PFAS-free materials, and even space-rated cables.
• How to Train an AI For Science — This panel discusses the practical challenges of building AI models for biological science, featuring three different perspectives: academic foundation model development (Yusuf Roohani, Arc Institute), mechanistic interpretability tooling (Daniel Balsam, Goodfire), and applied AI-first antibody and protein design (Peyton Greenside, Big Hat Biosciences).
• How-Sustainability-Has-Become-The-New-Innovation-Currency — A 44-minute panel discussion from the Danaher Summit 2026 on The Next Era of Bioprocessing, exploring how sustainability has become a core innovation driver in biopharma.
• IT Equipment in Immersion Cooling: Design and Validation Guidelines — This presentation announces and scopes the major refresh of the OCP ITE (IT Equipment) Design Guidelines for Immersion Cooling, originally published in December 2020.
• Immersion Cooling: Progress, Trends, and Industry Growth — This keynote-style presentation opened the OCP EMEA 2025 immersion cooling track, delivered by Rolf Brink (Promersion), who serves as OCP CE-Immersion Project Co-chair alongside Jessica Gullbrand of Intel.
• Immersion Fluid Specification — The OCP Base Specification for Immersion Fluids (Revision 1.0, approved December 2022) addresses a fundamental gap between immersion fluid development and IT equipment requirements.
• Immersion IT Equipment (ITE) Workstreams Update — Austin Hipes, OCP ITE Committee Chair and Chief Technologist at Unicom Engineering, presented the status of the three active workstreams under the OCP Immersion IT Equipment (ITE) Committee at OCP EMEA Summit 2025 (Dublin, 29–30 April).
• Immersion Integrated OCP Racks: The ORv3 Ecosystem Into Single-Phase Tanks — This presentation by Submer documents the end-to-end qualification and production deployment of ORv3 (Open Rack Version 3) hardware ecosystem components inside single-phase immersion tanks, covering busbars, power shelves/PSUs, compute nodes, and network switches.
• Immersion Reliability Committee Update — Brian Kinkade (FUCHS) and Betsy Cossette (ExxonMobil) presented the status of the OCP Immersion Reliability Committee's three active workstreams at OCP EMEA Summit 2025 (Dublin, 29–30 April).
• Immersion Server Demonstrated Reliability at Datacenter Scale Deployment — Austin Hipes (Unicom Engineering) presented the most comprehensive publicly disclosed real-world reliability dataset for single-phase immersion cooling at OCP EMEA Summit 2025 (Dublin, 29–30 April).
• Immersion Solutions: A Systems-Level Approach to Cooling at Scale — This session is a coordinated report-out from the five workstream leads of the OCP Immersion Solutions Technical Committee, presented at the 2025 OCP EMEA Summit in Dublin (29–30 April).
• In-Network Collective Acceleration for AI Fabrics — Collective operations (AllReduce, AllToAll, AllGather, ReduceScatter) account for more than 90% of AI fabric bandwidth.
• Industry Liaison Update on Progress to Accelerate Immersion Cooling — Presented at the 2025 OCP EMEA Summit in Dublin (29–30 April), this session provides the first six-month progress report of the OCP Immersion Industry Liaison Team (ILT), a group formed to bridge the gap between the OCP immersion community's technical expertise and the formal standards bodies that govern data center safety, electrical, and fire protection requirements.
• Innovative Solutions for New Generation Liquid-Cooling AI Data Centers Based on NVIDIA GPUs — Aivres presents its end-to-end liquid-cooled AI data center solution portfolio based on NVIDIA B200 and B300 GPUs, spanning servers, racks, CDUs, secondary cooling loops, monitoring systems, and containerized modular deployment.
• Insights from Production: Scheduled Ethernet Fabric in Large AI Training Clusters — This presentation details ByteDance's journey from evaluating scheduled Ethernet fabric technology in 2022 to deploying it in production in June 2024, co-presented with Broadcom who provides the underlying Jericho and Ramon silicon.
• Integrated Power and Precision Cooling for 800 VDC Architecture — Liteon Technology presents its integrated solution for ultra-dense AI clusters powered by 800 VDC (0 to 800V, not +/-400V) architecture, addressing the jump from 120 kW racks at 48V to 1-1.2 MW racks requiring fundamentally different power distribution.
• Intel Computex 2026 Keynote — Intel CEO Lip-Bu Tan anchored a 66-minute Computex 2026 keynote framed as a transformation story: 14 months into his tenure, Tan presented Intel not as an incumbent defending legacy share but as a company rebuilding from engineering fundamentals outward.
• Interconnect Design Optimization & Predictive Reliability Models for Immersion Cooled Interconnects — Molex presents a three-year research program (2022–2024) on interconnect design optimization and predictive reliability modeling for immersion cooled environments, focusing on two primary connector candidates: the OCP IT Gear input connector (ORv3 specification) and the Molex Extreme 1060 board-to-board connector.
• Interconnect Solutions for Next Generation High Density AI Data Centers presented by TE Connectivity — TE Connectivity presented at OCP Global Summit 2025 its interconnect, power, and thermal solutions for next-generation high-density AI data centers, introducing the 'interconnect wall' thesis and showcasing first-to-market liquid-cooled bus bars deployed at Meta.
• Interoperability as an Innovation Catalyst: Accelerating Data Center Transformation Together — HPE's Jean-Marie Verdun (DC-MHS co-creator) presents at OCP Global Summit 2025 the case that DC-MHS cross-vendor interoperability is currently broken despite OCP compliance, and outlines five calls to action for the OCP community including PnP finalization, system-level firmware, and workload benchmarks.
• Inventec Innovations in Liquid Cooling: Advancing Sustainable Data Centers — Inventec presents its liquid cooling journey spanning from 2012 to 2026, positioning itself as the fourth-ranked global company in liquid cooling patents and number one among Taiwanese ODMs (per Digit Times).
• Investigation of Immersion Cooling For Data Center Hardware: Progress, Opportunities And Challenges — This presentation reports Meta's multi-year investigation into single-phase immersion cooling as a path to supporting next-generation AI hardware in their data centers.
• LITEON GPU Power Solutions presented by LITEON — LITEON Technology's GPU power solutions roadmap at OCP Global Summit 2025, tracing the evolution from embedded PSUs to 800 VDC 1.2 MW power racks with bulk capacitor power smoothing innovations.
• Leading GreenIT for Modern AI Data Centers, Powered by DLC 2 and DCBBS — Super Micro presents its transformation from rack system provider to full data center solution provider through two key technologies: DLC2 (Direct Liquid Cooling 2, covering 95%+ of chips with liquid cooling) and DCBBS (Data Center Building Block Solutions, encompassing everything inside and outside the data center).
• Leading the Future of AI Networking — This panel discussion between Micas Networks and Super Micro addresses liquid-cooled networking and co-packaged optics (CPO) as critical enablers for AI factory networking at gigawatt scale.
• Leveraging Open Technologies to Monitor Packet Drops in AI Cluster Fabrics — Aldrin Isaac from eBay presents a critique of traditional network telemetry and proposes sFlow Drop Notification as a practical, vendor-neutral solution for monitoring packet drops in AI cluster fabrics.
• Liquid Coolant Bazaar: Everything you always wanted to know about Immersion but were afraid to ask. — This session was a live Q&A forum at OCP Global Summit 2023, structured around a physical "bazaar" displaying 18 immersion cooling fluids from 13 manufacturers — all submitted as professional samples that passed cursory business-level due diligence.
• Liquid Cooling SSDs Isn't Easy: Challenges, Innovations, and the Need for New Standards — OCP Global Summit 2025 session from Solidigm and Iceotope arguing that liquid cooling for SSDs is no longer optional, covering engineering solutions for hot-swap cold plate integration and precision immersion cooling.
• Liquid-to-Chip Cooling at the 1MW Frontier: Jabil and Mikros Technologies Roadmap — Mikros Technologies (acquired by Jabil approximately one year prior) presents its approach to advanced cold plate design for 1 MW rack cooling, emphasizing that the critical metric is not total thermal design power (TDP) but heat flux density (watts per degree C per square centimeter).
• MangoBoost Full-Stack AI Infrastructure Solutions: MLPerf Inference, Training, Storage Case Studies — MangoBoost presents its full-stack AI infrastructure solution comprising three product lines: LLMBoost (full-stack AI software for inference, training, and fine-tuning with auto-tuning), GPUBoost (FPGA-based hardware-accelerated RoCEv2 RDMA NIC with programmable congestion control and UEC features), and BoostX (NVMe-over-Fabric accelerator for storage with full hardware TCP/RoCEv2/NVMe offload engines).
• Manufacturing-The-Impossible-Cell-&-Gene-Therapies-From-Bench-to-Billions — This panel at Danaher Summit 2026 gathers four cell and gene therapy manufacturing leaders — Amy Walker (Forbase Bio), Matthew Dy (Cell and Gene Therapy Catapult), Edwin Stone (Cellular Origins), and Manuel Corando (IBET) — to address the transition from bench-scale to commercial production.
• Material Compatibility Evaluation of High-Power Interconnects For Single-Phase Immersion Cooling — This presentation reports on material compatibility and reliability testing of high-power interconnects for single-phase immersion cooling, with the goal of identifying what changes are needed to adapt the EIA-364 connector reliability standard for the immersion environment.
• Meeting the Scalability Demands of Rack Scale AI Infrastructure through Open Source Firmware — Sanjoy Maity (AMI CEO) presents AMI's open-source firmware strategy at OCP Global Summit 2025: 54,000 annual engineering hours maintaining an unfragmented OpenBMC LTS codebase and three OCP SAFE-compliant open-source products.
• Megmeet 800V 1MW Sidecar Power System for AI Data Centers — Megmeet, a top-three global power supply company ($1.1 billion revenue, approximately 10,000 employees with 35% in R&D, established in Shenzhen in 2003), presents its 800V DC 1 MW sidecar power system for AI data centers developed in close collaboration with NVIDIA.
• Memory for Every AI Need: Samsung's Next-gen Memory Solutions to Lead the Future of AI — Taeksang Song (Samsung Corporate VP of DRAM Solution Engineering) presents Samsung's full AI-era memory portfolio at OCP Global Summit 2025, spanning HBM4/HBM4E, custom HBM, DDR5 RDIMM/MRDIMM, LPDDR6, LPCAMM2, SOCAMM2, Processing-in-Memory, and CXL CMM-D modules.
• Meta 51.2T Ethernet Switch — Meta introduces its 51.2T Ethernet switch platform in two SKUs: Minipack3 (based on Broadcom Tomahawk5 ASIC) and Cisco 8501 (based on Cisco Silicon One G200 ASIC).
• Methodology and Observation of Congestion Control Impact on MoE Training Job Completion Time — Keysight Technologies presents research demonstrating that Mixture of Experts (MoE) AI training workloads create a unique congestion control challenge not present in dense 3D model-parallel training.
• Modularity, Standards, and Cloud Services: HPE and OVH Cloud Partnership — HPE and OVH Cloud jointly present their partnership approach to modular hardware and cloud service delivery.
• Modularizing the Datacenter: A Layered Approach From HV to Compute and Cooling — JP Bazelle from Eaton presents a comprehensive argument for modularizing data center infrastructure across four blocks -- grid connection, gray space, white space, and cooling -- unified by software.
• Molex Genesis SFF-TA-1040: Enabling the Next Generation of PCIe Internal Cabling Standards — Molex presents Genesis, a direct-attach internal cable assembly designed from the ground up for PCI Express 7.0 at 128 Gbps per lane using PAM-4 signaling at 85 ohm nominal impedance, compliant with the SFF-TA-1040 standard.
• More Than Cooling: Lifecycle Service For AI Factory — Envicool, a Chinese thermal solutions company established in 2005 and publicly listed since 2016, presents its expanded lifecycle service approach for liquid-cooled AI data centers.
• Multi-Core Fiber - A solution for Scale-up, Out and Across — Taylor Bremekamp of Sumitomo Electric Lightwave presents multi-core fiber (MCF) as the next-generation fiber infrastructure solution for AI data center networking.
• NVIDIA Spectrum-X Network Platform Architecture — NVIDIA Spectrum-X is an end-to-end Ethernet networking platform purpose-built for the East-West GPU-to-GPU fabric in generative AI cloud environments.
• NVMe™ Interoperable Migration of PCIe™ virtual SSDs — At the 2025 OCP Global Summit, Microsoft's Lee Prewitt and NVMe technical-proposal author Mike Allison present interoperable live migration of PCIe virtual SSDs.
• Navigating Immersion Challenges: Addressing Risks and Warranty Guidelines in Immersion Cooling — This joint presentation covers two related OCP workstreams: the Immersion Challenges (Hurdles) Workstream led by Raul Alvarez and David T, and the Hardware Warranty Guidelines Workstream led by Rich Lappenbusch and Michael Cordle.
• Navigating Shared Storage: Mitigating Performance Challenges with SSD Quality of Service — Dan Helmick (Samsung) and Yuliya Tarnikova (Google) present the motivation and current status of SSD Quality of Service standardization at OCP Global Summit 2025.
• Navigating the Road to Co-Packaged Optics: Challenges and Considerations for Scalable Deployment — Kishore Atreya (Marvell Senior Director, Cloud Platform) presents the system-level deployment barriers blocking CPO mass adoption at OCP Global Summit 2025: field serviceability, 1U density, and rack-scale fiber management for 36,000+ fibers per rack.
• Networking for AI: Scaling Up, Out, and for the Future presented by Meta — Omar Baldonado (Meta Senior Director, Network Infrastructure) presents Meta's AI networking stack at OCP Global Summit 2025, announcing the Non-Scheduled Fabric, Minipack3N switch with NVIDIA Spectrum-4, and ESUN — the Ethernet for Scale-Up Networking OCP initiative.
• Next Gen AI Data Centers with Industry-proven Direct Current Power Supply — Hartwig Stammberger presents on behalf of the Open Direct Current Alliance (ODCA), an international nonprofit organization with ~90 members.
• Next-Generation Co-Packaged Optics Systems — This presentation describes the first collaborative co-packaged optics (CPO) switch system jointly designed and built by Marvell and Jabil, currently being demonstrated on the OCP 2025 show floor.
• Not All AI Racks Are Built the Same presented by QCT — QCT's presentation at OCP Global Summit 2025 argues that while all GB200 NVL72 racks share NVIDIA's reference design, the engineering choices made by each ODM — in power architecture, hose materials, thermal simulation, and manufacturing scale — determine real-world reliability and TCO. The talk walks through QCT's four-generation development journey from GH200 to GB200 NVL72 production, liquid cooling guidance, and the RM61X rack management controller.
• Novel Approaches to Network Telemetry: Essential for AI Performance — This presentation argues that traditional network telemetry, designed for workloads with diverse traffic patterns and fat-tailed distributions, is fundamentally unsuitable for AI training clusters.
• OCP Immersion Cooling Reliability — This presentation introduces the newly formed OCP Immersion Reliability Committee, explaining its structure, the thermochemical reliability framework it is building, and the four-phase component compatibility testing methodology captured in a whitepaper currently in progress (expected publication 2025).
• OCP Immersion Fluids Guidelines, Specifications and Lifecycle Management — This presentation provides a comprehensive update from the OCP Fluids Committee — one of the four new technical committees in the restructured OCP Immersion project — covering the current landscape of immersion fluid documentation, four active workstreams, and a detailed dive into the newly launched Fluid Cleaning Guidelines workstream.
• OCP L.O.C.K., Caliptra 2.1, and MEK-MPA Use Cases — At the 2025 OCP Global Summit storage session, Google security experts Charles and Jeff presented OCP L.O.C.K. (Layered Open-source Cryptographic Key management) alongside the new TCG MEK-MPA multi-party authorization feature set.
• OCP ORv3 Blind Mate Liquid Cooling Component Update — Nick Goenner (SafeWay, engineering manager) and Chris SC Lee (Wiwynn, mechanical engineer) present the component-level update for the OCP ORv3 Blind Mate Liquid Cooling project — the companion presentation to the system-level overview by Meta.
• OCP Optical Circuit Switching Subproject Update — Optical Circuit Switching (OCS) is an emerging data center technology that routes light between fiber ports without any electrical conversion, making it transparent to protocol, speed, and packet format.
• OCP Oxidation Sub-Group Progress Update — The OCP (Open Compute Project) Oxidation Stability Sub-Group, co-led by Ken Hope and Spencer Kerns of Chevron Phillips Chemical Company, is investigating how to measure and monitor the degradation of single-phase immersion cooling fluids over time.
• OCP SSD Test Panel — An OCP Global Summit panel, moderated by Microsoft's Lee Puit, in which two test-tool vendors explain how OCP NVMe SSD compliance test suites let drive vendors 'shift left' and catch defects early in development.
• OCP Solutions with NVIDIA GPUs for New Gen AI presented by Aivres — James Zo, research fellow at Aivres, presents the company's OCP-conformant AI server portfolio at the OCP Global Summit 2025 in San Jose.
• OCP Telemetry White Paper -- A Summary of the Work Led by Hyperscalers on DC Telemetry Guidelines — Scott Sharp, a mechanical engineer at Google and long-time OCP contributor, presents a telemetry standardization initiative led by Google, Microsoft, Meta, and Nvidia under the OCP Data Center Facility track.
• OCP and IOWN Global Forum for Open Optical Acceleration — Masahisa Kawashima of IOWN Global Forum presents the forum's three-layer optical infrastructure co-evolution framework and proposes a joint OCP-IOWN collaboration on edge-cloud convergence, distributed data centers, and open AI infrastructure with co-packaged optics and optical circuit switches.
• ORv3 Blind Mate Liquid Cooling Interfaces Overview & Update — Glenn Charest and Grace Piette (Meta engineers) present the system-level overview and 2023 progress update for the OCP ORv3 Blind Mate Liquid Cooling project — a closed workstream developing a holistic liquid cooling solution for ORv3 racks that enables self-aligning blind-mate fluid connections when chassis are inserted.
• OpenBMC and FBOSS on Network Switch — Arista Networks and Meta have been co-developing network switches since 2017, beginning with the 128x100G Tomahawk 3 switch which first added a BMC to what was traditionally a BMC-less Arista design.
• Opening Remarks from Rainer Blair — Danaher CEO Rainer Blair opens the company's fourth annual bioprocessing summit, framing it as a candid working forum on how biomanufacturing got here and the manufacturing challenges facing the next wave of modalities, anchored by patient stories like baby KJ.
• Operating Temperatures to Support Performance, Energy Efficiency, and Heat Recovery — NVIDIA engineers Mark Steinke and Wade Vinson present an analysis of how facility water supply (FWS) temperature selection affects cooling energy consumption, cooling power demand, and -- critically -- how much compute capacity can be deployed at gigawatt-scale AI factory sites.
• Optimal Path Utilization for Multi-Plane Fabric Design — This joint ByteDance-Broadcom presentation addresses the limitations of Adaptive Routing and Switching (ARS) in multi-tier AI fabric topologies, where ARS's reliance on local link state leaves it blind to remote downstream congestion and failures.
• Optimizing AI Infrastructure: The Advantages of OCP Busbar Power For GPU Modules And Full Rack — Pegatron presents a high-density GPU rack design using the OCP ORv3 busbar architecture to maximize GPU density within a 51OU rack.
• Optimizing Data Center Cooling: TCO And Design Comparisons Between Single-Phase And Two-Phase — Accelsius, a two-phase direct-to-chip (D2C) liquid cooling company and OCP startup member, presents a TCO analysis comparing single-phase and two-phase D2C cooling for a 10 MW data center.
• Optimizing Immersion Fluids: Insights into Lifecycle Management, Cleaning and Longevity — This session covers the OCP Immersion Fluids Committee's active workstreams and their 2025 roadmap, presented at the OCP EMEA Summit 2025 in Dublin.
• Optimizing Signal Integrity in Immersion Cooled IT Platforms — This presentation from Intel, Foxconn Interconnect, and FIT reports on the OCP Signal Integrity (SI) Working Group's ongoing effort to characterize and mitigate the impact of single-phase immersion cooling fluids on high-speed interconnects.
• PBMC (Pivoting Blind Mate Coupling) Update & Specification — The PBMC (Pivoting Blind Mate Coupling) is a newly published OCP specification for a multi-source, interoperable blind mate valve designed to handle the high flow rates demanded by next-generation high-power AI and compute systems.
• PICs for Quantum and AI — Valerio Pruneri, working from ICFO (Institut de Ciències Fotòniques) approximately 15 miles from the OCP EMEA Summit venue in Barcelona, presents PIX Europe — a European pilot line for photonic integrated circuits (PICs) funded under the Chips Joint Undertaking — and illustrates its scope through three technology clusters: quantum random number generation, quantum key distribution, and a novel phase-imaging technique for PIC metrology and failure analysis.
• Panel -- Data Center Automation with Robotics & Cybersecurity — This panel, moderated by Meta's Eric Xu, is the only dedicated data center automation forum in the industry and covers the full spectrum of physical robotics deployment in hyperscale data centers alongside the cybersecurity requirements that must accompany connected machines on the data hall floor.
• Panel -- Insights for Cleaner Liquid Cooling: Cleaning & Commissioning the TCS — This panel session, closing the Data Center Facilities and Cooling Environments tracks at OCP Global Summit 2025, addresses the commissioning and cleanliness requirements for Thermal Cooling Systems (TCS) in liquid-cooled data centers.
• Panel: Coolant Temperatures for Durable Data Center Designs — This panel presented and debated an industry-wide recommendation to standardize liquid coolant supply temperature at 30°C for direct-to-chip (cold plate) liquid cooling in AI data centers.
• Panel: EDSFF New and Upcoming Updates — A panel at the 2025 OCP Global Summit storage track on new and upcoming EDSFF (Enterprise and Datacenter SSD Form Factor) developments.
• Panel: Flexible Data Placement (FDP) — An OCP Global Summit storage-track panel on NVMe Flexible Data Placement (FDP), the host-directed data-placement feature that reduces SSD write amplification (WAF) toward 1.0.
• Panel: Future of 2-Phase Immersion Cooling: Addressing the PFAS Challenge — This panel at OCP Global Summit 2023 convened five industry experts to address what moderator Rolf Brink described as "the biggest hurdle" facing two-phase immersion cooling: the PFAS (Per- and Polyfluoroalkyl Substances) regulatory challenge.
• Pioneering with OCP: Jabil's Building Blocks for the Modern Data Center — Brad Rampelberg, Chief Sales Officer for Jabil's Intelligent Infrastructure segment, traces the OCP origin story from Facebook's Project Freedom (c. 2008) through to Jabil's current building-block strategy at the 2025 OCP Global Summit.
• Power Distribution in Immersion Environments: Challenges and Opportunities — This presentation reports on the OCP Power Distribution in Immersion workstream, which is drafting a base specification for integrating ORV3 (Open Rack Version 3) power infrastructure into single-phase immersion tanks.
• Power Distribution in Immersion: Enabling High Power Delivery in Immersion Cooling — This presentation covers the OCP Power Distribution in Immersion workstream, which addresses how to deliver high-density power to Open Rack Version 3 (ORv3) hardware inside immersion cooling tanks.
• Power Stabilization for AI Training DCs presented by Microsoft — Microsoft, NVIDIA, and OpenAI presented their ~2.5-year cross-company research on power stabilization for AI training data centers at OCP Global Summit 2025, addressing the problem of GPU clusters creating massive, grid-destabilizing power oscillations at gigawatt scale.
• Powering Next Generation Switch Architecture for AI Hyperscale Infrastructure through Open Standards — Michael Lane, VP of Networking at Hyve Solutions, presents a framework for next-generation switch architecture requirements for AI hyperscale data centers at OCP Global Summit 2025.
• Powering Progress: Murata's Role in Advancing OCP Innovation — Murata, a multi-billion-dollar Japanese components company with over 40 product groups, presents its end-to-end power delivery portfolio for AI data centers spanning AC input through to processor-level power.
• Powering The AI Era: High-Density Racks, Cutting-Edge Systems, And Advanced Cooling From Wiwynn — Wiwynn, a Taiwanese AI server and data center solution provider, presents a comprehensive overview of its rack-level, system-level, and component-level technologies for AI data centers.
• Predictions to Proof: Translating AI Into Real Biology — This panel takes stock of what AI has actually delivered in drug discovery as of Q1 2026, diagnoses the remaining gaps, and debates whether a "ChatGPT moment" for clinical success improvement is plausible.
• Pumping Innovation - Long-term Testing with an Immersion Cooling Fluid — This session presents the results of a two-year cross-company collaboration between Wilo SE (pump manufacturer) and Castrol (lubricants/fluids) to develop and validate a pump-fluid compatibility testing methodology for immersion cooling applications.
• QUIC Kernel and Device Offload — Broadcom engineers presented a detailed architecture for offloading QUIC protocol cryptographic operations to NIC hardware, building on the successful KTLS offload model.
• Quantum-Resilient Security For Datacenters: A QRNG-RoT FPGA Framework With Live Demonstration — Lattice Semiconductor and Quside (Barcelona-based quantum technology company) jointly present a quantum-resilient trust infrastructure platform combining a quantum random number generator (QRNG) with an FPGA-based root of trust.
• Rack Emulator Design for Multigenerational IT Racks — This presentation by NVIDIA and Boyd introduces a purpose-built rack emulator designed to replace traditional load banks and TTV (Thermal Test Vehicle) emulators for data center commissioning.
• Rack Manifold Requirements and Qualification — This presentation covers the OCP Rack Manifold Requirement and Qualification Guideline, a white paper in final community review at the time of OCP Global Summit 2023.
• Rearchitecting the Data Center for the Convergence of Power and Compute presented by Flex — At OCP Global Summit 2025, Flex executives Rob Campbell and Chris Butler presented a roadmap for rearchitecting data centers by breaking down the organizational and physical walls between power delivery, cooling, and IT hardware.
• Redefine Networking, Awaken Cognitive AI with Cisco — Rakesh Chopra (SVP and Fellow, Hardware Architecture, Cisco) presented Cisco's case for rethinking AI networking at OCP Global Summit 2025, arguing that power — not bandwidth — is the fundamental constraint on AI cluster scaling.
• Redefining Storage for the AI Era — Maarten Guijt, Seagate's VP for Europe, presented a data-centric argument that AI has fundamentally elevated storage from an afterthought to a first-class infrastructure building block.
• Redfish Practices and Learnings in Immersion Cooling Management — Mohan Kumar (Intel Fellow) presented Intel's practical experience building a Redfish-based management interface for an actual immersion cooling tank deployment.
• Reducing Compliance Complexity: A Unified OCP Safe and SESIP Path to CRA — This presentation from Brightsight (an SGS company) maps two established security evaluation frameworks — OCP S.A.F.E.
• Removing the Bottleneck: Building Scalable Infrastructure for the Generative AI Era — Matt Roman from Celestica presented the company's 100-terabit switching portfolio and its role in enabling the next generation of GPU cluster networking.
• Revolutionizing AI Deployment: Workload-Centric Edge Clusters from Edge to Cloud — Dr. Jen-Yao Chung of Inventec presents the Physical AI Three-Computer DevKit — a Kubernetes-orchestrated rack cluster implementing NVIDIA's three-computer architecture for on-premises edge AI in factories, hospitals, and moving vehicles.
• Revolutionizing Data Center Cooling with Synthetic Esters presented by Perstorp — Perstorp's Shubha Mitra Basu presents a case for replacing water-glycol direct-to-chip coolants with PFAS-free, biodegradable synthetic ester dielectric fluids, addressing heat as the central bottleneck limiting AI infrastructure performance and sustainability.
• Revolutionizing Networking for Gigawatt AI Factories — Yuval Barenshtein, Spectrum X Product Manager at NVIDIA, presents the full networking architecture for gigawatt-scale AI factories, structured as a "five-layer cake": energy → infrastructure → models → applications → AI agents.
• Revolutionizing Networking for the Era of Agentic AI — NVIDIA's Spectrum X platform is purpose-built for AI scale-out networks, and this talk articulates why off-the-shelf (OTS) Ethernet switches — while interoperable — fail to deliver the bandwidth efficiency, latency determinism, and lossless behavior required for large-scale GPU training and inference.
• SAI APM: Advanced Performance Monitoring — Microsoft and Broadcom jointly presented a proposal for SAI Advanced Performance Monitoring (APM), a new framework for measuring and monitoring the performance of the SAI (Switch Abstraction Interface) adapter layer in production network deployments.
• SAI Specifications for Newer Technologies Introduced by Ultra Ethernet Consortium — The Ultra Ethernet Consortium (UEC) is standardizing a set of new link-layer and telemetry technologies specifically for AI backend networks, where packet loss from buffer overflow or frame corruption causes hundreds of GPUs to stall waiting for end-to-end retransmissions.
• SAI for Tunnel Security with IPsec — Arista Networks presented SAI APIs for enabling hardware-accelerated IPsec tunnel security at line rate on switches where tunnels are terminated.
• SONiC - NOS of Choice for AI Data Centre — Presented jointly by two executives from Upscale AI, this session argues that SONiC (Software for Open Networking in the Cloud) is becoming the de facto Network Operating System for AI data center fabrics — for both scale-up (intra-rack GPU-to-GPU) and scale-out (inter-rack) networking.
• SONiC SRv6 VPN Deployment for Predictable WAN — Alibaba and Cisco presented their joint effort to build a predictable global WAN using SRv6 VPN on SONiC-based white box routers.
• SRv6 for AI Backend — Microsoft is deploying Segment Routing over IPv6 (SRv6) on its AI backend networks using SONiC as the network operating system, addressing two core challenges: the low-entropy nature of AI training traffic (small numbers of large elephant flows that don't distribute well across equal-cost paths) and the need to connect geographically distributed GPU clusters across multiple physical data centers.
• Safe Handling Guidelines for Immersion Fluids — This presentation provides a progress update on the OCP Safe Handling Guidelines for Immersion Fluids workstream.
• Sam Stanton, Coefficient Bio With Nan Li, Dimension Capital — This fireside chat at UNLOCK 2026 covers the founding, product development, and Anthropic acquisition of Coefficient Bio, a startup that built AI tools for scientific reasoning and clinical development decision-making.
• Scale or Stall: The Case for Purpose-Built Connectivity in Open Ecosystems presented by Astera Labs — Astera Labs presents its vision for purpose-built, open-standards connectivity as the path to scaling AI infrastructure beyond single-vendor bottlenecks, arguing that open UALink — backed by major industry promoters — enables multi-vendor, purpose-built scale-up of up to 1,000-GPU clusters.
• Scaling AI Infrastructure With Optical Interconnects — Marvell presents a comprehensive overview of optical interconnect technologies required for exascale AI factories, covering three distinct application domains: scale-up (XPU-to-XPU within and beyond racks via CPO), scale-out (rack-to-rack via PAM4 pluggable transceivers), and scale-across (site-to-site via coherent and ZR/ZR+ pluggables).
• Scaling AI Networks: Incremental Innovation or Photonic Disruption? — This 33-minute panel discussion features four companies across the AI networking connectivity spectrum — two established scale-up connectivity vendors (Astera Labs, Credo Semiconductor) and two startups proposing photonic disruption of the switching layer (Astra Networks, Oriole Networks).
• Scaling AI: The Case For 100T Switching — Celestica, a Toronto-headquartered technology company with approximately $10 billion in annual revenue, 27,000+ employees across 40+ locations in 16 countries, presents the transition from 51.2T to 102.4T switching silicon and its new DS6000/DS6001 switch platforms.
• Scaling Sustainable Datacenters with Open Firmware: A Successful Ecosystem Collaboration across MiTAC, OSFF, ISVs, and Open Hardware Standards — 2025 OCP Global Summit session by MiTAC, AMD, and 9elements making the case for open-source firmware as the foundation for scalable and sustainable datacenters, covering OpenBMC/OPF, the openSFI specification, and the Open Source Firmware Foundation.
• Shell Immersion Coolant Solutions for Data Centers — Shell presents its single-phase immersion coolant portfolio for data center cooling, centered on the S3X grade -- a 3 centistokes synthetic hydrocarbon produced through Shell's gas-to-liquid (GTL) catalytic process.
• Signal Integrity in Immersion: Overcoming Challenges for High-Performance IT Platforms — This presentation addresses signal integrity (SI) challenges introduced by immersing air-designed IT hardware in dielectric immersion fluids — a topic that emerged in importance alongside the increasing data rates of modern interconnects.
• Silicon Photonics Switching: From Technology to Deployment — iPronics founder Daniel Pérez-López argues that silicon photonics-based optical circuit switches (OCS) are the "missing link" for super scale-up networking, and presents his company's deployed product as evidence that the technology is ready now.
• Simplifying The Deployment of Agentic AI at Scale — Accton/Edgecore presents their in-house experience deploying agentic AI at enterprise scale, addressing the operational challenges that make enterprise AI adoption difficult despite exciting technology breakthroughs.
• Single-Phase Immersion Heat Sinks: Thermal Evaluation and Targeted Flow Technology — Intel's data center group presents two linked innovations for single-phase immersion cooling: a standardized Immersion Flow Tunnel metrology test bed for reproducible heat sink thermal characterization, and Targeted Flow Technology — a method of directing fluid flow directly to CPU heat sinks via a topside manifold rather than relying on bulk tank convection.
• Solving-The-Hardest-Problems-Through-Partnership-And-Innovation-Part-1 — A researcher from the San Raffaele Telethon Institute for Gene Therapy presents the institute's 30-year journey from foundational science to three approved hematopoietic stem cell gene therapies for ADA-SCID, metachromatic leukodystrophy, and Wiskott-Aldrich syndrome.
• Solving-The-Hardest-Problems-Through-Partnership-And-Innovation-Part-2 — This talk is a deep-dive interview with a gene therapy pioneer at Ospedale San Raffaele in Milan, covering the personal, scientific, manufacturing, regulatory, and systemic dimensions of bringing rare disease gene therapies from academic discovery to commercial approval.
• Source Routing for AI Fabrics — Marvell proposes Flow Path Routing (FPR), a source-routing approach for managing traffic flows in AI training fabrics as an alternative to existing congestion-control methods like packet spraying, virtual output queues (VoQs), and telemetry-assisted adaptive congestion control.
• Spec Updates & What Customers Care About — An OCP Global Summit 2025 storage panel updating the OCP Data Center SAS/SATA (1.5) and NVMe SSD (2.7) specifications and surfacing what hyperscaler and OEM customers care about.
• Sprandom: A Fast, Pseudo-Random Preconditioning Method for High-Capacity Solid State Drives — At the 2025 OCP Global Summit, SanDisk engineer Steven Sprouse introduced Sprandom, a fast pseudo-random preconditioning method for high-capacity SSDs.
• Standards-Based Computing For The Modern Datacenter — MSI Enterprise Platform Solution presents its full adoption of the OCP DC-MHS (Data Center Modular Hardware System) architecture for all current and next-generation server motherboards.
• Storage Project Close — Closing remarks for the Storage track at the 2025 OCP Global Summit.
• Strategies to Boost Switch Programming — Google presents a systematic analysis of route programming performance bottlenecks in their PINS (P4 Integrated Network Switch) stack and the strategies they employed to achieve an 8x improvement for link-down convergence and 16x improvement for link-up convergence.
• Switch Cable Diagnostic with TDR Using SAI and PAI — Broadcom presents a Time-Domain Reflectometer (TDR) cable diagnostic capability implemented in SONiC using SAI and PAI (PHY Abstraction Interface).
• Synthetic Ester Coolants: Powering Safer, More Sustainable And Efficient Data Centers — Perstorp, a Swedish engineered fluids producer with 140 years of experience (acquired by Petronas in 2022), presents the case for using dielectric synthetic ester coolants as an alternative to PG25 (propylene glycol 25%) water-based coolants in direct-to-chip (cold plate) liquid cooling systems.
• Systems-Level Integration of Immersion Cooling: An Overview of the OCP Solutions Technical Committee — This presentation is the opening of the OCP Immersion Solutions Technical Committee segment at OCP Global Summit 2024.
• TCO Modeling for Immersion: Unlocking Cost Savings and Efficiency in Data Centers — This presentation introduces the OCP Immersion TCO Workstream and its in-development Excel-based TCO modeling tool, targeting Rev 1.0 release by end of Q4 2023.
• Technical Paths to the New Era of GPU-Initiated Storage — At the 2025 OCP Global Summit, CJ Newburn and Vikram Mailthody present technical paths to a new era of GPU-initiated storage: reinventing the GPU so it directly initiates fine-grained, massively concurrent access to NVMe storage independent of the CPU.
• Telemetry Based Load Balancing of AI/ML Workloads — Broadcom presents a production deployment case study from Tencent's hyperscale AI network demonstrating how inband telemetry (IFA -- In-band Flow Analyzer) and telemetry-based load balancing deliver up to 40% bandwidth utilization improvement over baseline DCQCN (Data Center Quantized Congestion Notification).
• The Blueprint For Brilliance: QCT's Modular Approach to AI Solutions — QCT (Quanta Cloud Technology), the cloud and data center subsidiary of Quanta Computer, presents its modular design philosophy for coping with the accelerating pace of AI chip releases -- from approximately 5 new chips per year before 2022 to 10+ per year since.
• The Challenges and Practices of Network Stability in Alibaba's Large-Scale Computing Clusters — This joint Alibaba-Broadcom presentation addresses two fundamental challenges in maintaining network stability for distributed training at Alibaba's scale: (1) rapidly detecting, localizing, and recovering from communication faults, and (2) observing real network communication behavior to identify performance optimization opportunities.
• The Future of Single-Phase Direct Liquid Cooling — Kamal Mostafavi, CTO of CoolIT Systems (celebrating 25 years), presented a comprehensive data-driven forecast for liquid cooling adoption in AI data centers, arguing that single-phase Direct Liquid Cooling (DLC) will remain the mainstream thermal management solution for the foreseeable future, even as rack densities scale toward 250 kW and pod-level deployments reach 9–10 MW.
• The Immediate Impact of Open Networking — The Rakuten Use Case — Rakuten Mobile — the Telco division of the Japanese Rakuten Group — executed a three-phase proof-of-concept of open networking for its data center infrastructure in 2025, driven by power constraints, space limitations, subscriber growth, and a desire to reduce vendor lock-in.
• The Impact of Flexible Data Placement in a Software Development Environment — Kioxia presents experimental results demonstrating that NVMe Flexible Data Placement (FDP), a new SSD technology coming to market in 2025, can dramatically reduce write amplification factor (WAF) in real workloads without any application code changes.
• The Journey to OCP Product Recognition for a Single-Phase Immersion Cooling Fluid — FUCHS SE became the first immersion cooling fluid vendor to achieve OCP Product Recognition listing on the OCP Marketplace with their RENOLIN FECC 5 SYNTH fluid, completing the process between May and August 2024.
• The Path to Zero Flap: Reinventing Optical Reliability for Scalable AI Clusters — OCP Global Summit 2025 session by Credo and Oracle making the case that optical link flaps are the primary reliability threat to large-scale AI clusters, and presenting ZeroFlap optics as the solution.
• The Power Solution Has Changed - Efficiency Must Follow the Load — Philip Zuk from OmniOn Power argued that the traditional data center power efficiency metric — peak efficiency at a single operating point — is insufficient for AI workloads, which are fundamentally dynamic.
• The-Changing-Map-For-Global-Biomanufacturing — This panel discussion from the Danaher Summit 2026, moderated by Cytiva China leader Alan Lee, brings together four senior leaders from Asia-Pacific biopharmaceutical companies to discuss the shifting global landscape of biomanufacturing.
• The-Inflection-Point-For-Nucleic-Acid-Drugs-is-a-Targeting-Problem — The CEO of Dyne Therapeutics presents a 40-year history of oligonucleotide therapeutics at the 2026 Danaher Summit, arguing that targeted delivery — not chemistry — is now the critical remaining bottleneck.
• The-Unfinished-Work-of-Traditional-Bioprocessing — This Danaher Summit 2026 panel features three senior bioprocessing leaders — Ranga Godavarti (Pfizer), John Kaufman (AstraZeneca), and Thomas Saur (Sanofi) — representing a combined 90+ years of biologics development experience, examining the unfinished challenges of traditional bioprocessing.
• Thermochemical Reliability of Components in Immersion Solutions — Intel's OCP Immersion Reliability Committee lead, Samantha Yates, PhD, presents a five-step thermochemical reliability methodology developed and applied to Intel Ice Lake (3rd-generation Xeon) CPUs submerged in PAO 6 (polyalphaolefin grade 6) dielectric fluid.
• Total Cost of Ownership (TCO) Model for Liquid Cooled Data Centers — This presentation introduced OCP TCO tool v1.3, an open-source, macro-free Excel spreadsheet for modeling the total cost of ownership of data centers across any combination of air cooling, liquid cooling, and hybrid configurations.
• Towards Circular Economy: Managing the Lifecycle of Immersion Cooling Fluids — This presentation reports on a newly formed OCP workstream dedicated to fluid lifecycle management — what to do with immersion cooling fluids when they reach end of usable life or when equipment is retired.
• Towards Wafer-scale Optical Interconnect Relying on Silicon Photonics and Advanced 3D Assembly — This presentation by imec's Peter Ossieur makes the case that copper-based scale-up interconnect is approaching a fundamental wall, and that optical interconnect — combined with silicon photonics and advanced 3D assembly — is the only credible path to scaling AI accelerator fabrics beyond single-rack boundaries.
• Transforming Deployment Challenges Into Opportunities With Cisco AI Networking — Cisco presents its approach to mainstreaming AI networking infrastructure beyond hyperscalers into neocloud, sovereign cloud, and enterprise environments.
• Two Phase Immersion Cooling Study For OCP Accelerator Module OAM — Meta and Wiwynn jointly conducted a systematic two-phase (2P) immersion cooling study targeting the OCP Accelerator Module (OAM) form factor, motivated by growing power density in AI/ML data center infrastructure.
• UNLOCK 2026 Biology For AI The Next Breakthroughs Are Methods, Not Models Henry Lee, Cultivarium — Henry Lee, representing Cultivarium (a nonprofit research institution in Boston, Massachusetts), argues that the dominant bottleneck for AI in biology is not model architecture but biological data generation methods.
• UNLOCK 2026 How to Build Agents For Scientific Discovery in 2026 Andrew White, Edison Scientific — Andrew White, co-founder of Edison Scientific (spun out of Future House), delivers a technically dense 18-minute talk covering the state of AI agents for scientific discovery as of 2026, drawing on three years of personal experience building these systems.
• UNLOCK 2026 Recap AI x Life Sciences Conference in San Francisco — This is the official highlight reel recap video for the inaugural UNLOCK 2026 conference, an AI x Life Sciences event held in San Francisco on April 22, 2026.
• UNLOCK 2026 Unlocking Clinical Response Prediction With Multimodal World Models Ron Alfa, NOETIK — Ron Alfa, representing NOETIK, presents the company's approach to building foundation models of human biology with the explicit goal of predicting clinical drug response.
• UQDv2 - Interoperability & Future Roadmap — This presentation by Elizabeth Langer of CPC reports on the post-design-freeze progress of the OCP Universal Quick Disconnect version 2 (UQDv2) specification, covering interoperability testing results from Wave 1, the methodology developed for bench-independent pressure drop characterization, and the newly introduced seal washout test protocol.
• Ultra Ethernet: Escaping the Chaos of Order presented by UEC — OCP Global Summit 2025 session by UEC's Keith D Underwood (HPE) arguing that ordered network transports are architecturally self-defeating and presenting Ultra Ethernet's clean-slate full-stack architecture with out-of-order packet delivery and ECMP packet spraying.
• Unifying Immersion Insights Into Minimum Requirements And Fluids & Materials in Immersion Cooling — This presentation provides a comprehensive status update on the OCP Fluids & Materials (F&M) workstream as of October 2023, reporting progress across seven active focus groups tasked with qualifying immersion cooling fluids for large-scale data center deployment.
• Watch Nvidia's Computex 2026 Keynote Live — Jensen Huang delivered NVIDIA's 134-minute GTC Taiwan keynote at Computex 2026, organized around a single thesis: "useful AI has arrived." Agentic AI — AI systems that observe, reason, plan, use tools, and act autonomously — has crossed from research into production, validated by GitHub data showing 90 million pull requests merged, 1.4 billion commits, and 20 million new repositories per month in early 2026, with commits nearly tripling year-over-year (Image 205, 22:55).
• Water-based Coolant for ICT Liquid Cooling System presented by TCT — OCP Global Summit 2025 session by TCT engineer Royal presenting a comparative analysis of pure water coolant versus formulated coolant for ICT liquid cooling, introducing the medium purification module (MPM) and presenting a five-year China verification project.
• Welcome & Kickoff Address — This kickoff address by Richard Pitwon (Seagate / Resolute Photonics), serving as session host, opens the OCP EMEA Summit 2026 Photonics track in Barcelona on 29–30 April 2026.
• Welcome & Kickoff to Networking Track — The OCP Networking track opens with project co-leads Rindra Sunar and Barak Gafni introducing the state of the OCP Networking project, welcoming new steering committee representative Shu Wang from Meta, and providing a comprehensive recap of the year's activities and contributions.
• Welcome & Storage Project Introduction — The opening welcome to the storage track at the 2025 OCP Global Summit, hosted by Meta's Ross Stenfort with co-lead Lee Prudhomme and steering committee representative Mike Allison.
• Welcome to UNLOCK 2026 Michelle Lee, CEO & Founder, Medra — Michelle Lee, CEO and Founder of Medra, delivers the opening welcome address at the inaugural UNLOCK 2026 conference in San Francisco on April 22, 2026.
• What is Hype And What is Real? — This panel at the UNLOCK 2026 conference — "AI Science: Cutting Through the Hype" — brings together practitioners from Eli Lilly, Arc Institute, NewLimit, and Ginkgo Bioworks to distinguish genuine AI progress in drug discovery from overhyped claims.
• When Storage, DTC, And Immersion Meet, Data Stays Cool — Solidigm and DUG Technology jointly present three approaches to cooling storage in AI data centers: air cooling, direct-to-chip (DTC) liquid cooling with cold plates, and full immersion cooling.
• When-The-Patient-Becomes-The-Scientist — A hemophilia A patient-turned-scientist presents the personal and scientific journey of hemophilia treatment from daily hospital IV infusions to subcutaneous pen-injector biologics, tracing three paradigm shifts: factor replacement, gene therapy, and hemostasis rebalancing.
• Why Open Compute Project innovations heat the city swimming pool? — David Garon (Impon) presented a concrete case study from Midon, a suburb of Paris, where an IT boiler (a server-based heat generation system) was evaluated to replace gas boilers heating a municipal Olympic-size swimming pool.
• Wired For Success: Enhancing AI Cluster Reliability With Copper Cables — Credo presents the case for active electrical copper cables (AECs) as a reliability-critical alternative to optics for the first link between GPUs and leaf switches in AI clusters.
• eCore Architecture and Whitebox Evolution of Alibaba's Service-Oriented DCI Network — This talk describes Alibaba's eCore (ecore network), a next-generation service-oriented Data Center Interconnect (DCI) network that replaced their legacy global backbone (Aen) after a five-year development program.
• iDLC Implications on Balance of Immersion Cooling System — John Bean (co-lead of the OCP Immersion project and Principal at GRC — Green Revolution Cooling) presents a systems-level analysis of how iDLC (immersion Direct Liquid Cooling) — specifically the addition of targeted/directed flow heat sinks to immersion tanks — increases total dielectric fluid flow rates through the CDU (Coolant Distribution Unit), and how intelligent heat exchanger bypass valve strategies can recover pumping efficiency without sacrificing thermal performance.