Evolving the Infrastructure Stack: Rack-Scale Designs, Open Firmware, and Intelligent Sustainability

Author: Zach Julie, Chief Product Officer, AMI (presenting on behalf of Sanjoy Maity, CEO, AMI) Collection: 2026-OCP-EMEA-Summit-Executive-Sessions

Summary

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. The talk frames the convergence of AI training/inference workloads with the architectural imperative to move from individual servers to pools of disaggregated compute, memory, storage, power, and cooling managed as a unified system.

On rack-scale architecture, the presentation explains why the industry has moved from server-centric to rack-centric management: disaggregated components can be upgraded on independent cycles, workloads can be placed across pooled resources with higher utilization, and the control plane requires open APIs (Redfish/CXL) operating at the rack level. The rack architecture stack is built bottom-up: power and cooling (now moved to external power shelf with DC supply), memory/storage pools (CXL-attached), compute/accelerators (GPU trays), networking fabric (NVLink scale-up, InfiniBand/Ethernet scale-out, UALink/EUSON alternatives), and a management and control plane. AMI positions its firmware stack — Aptio (boot), MegaRAC (management), Tektagon (security) — as the firmware foundation for this stack, all offered as community editions under OCP.

On sustainability, the talk argues that AI power draws have temporarily suppressed sustainability focus, but power and cooling limits now force a return to efficiency. AMI's MegaRack software can manage CDUs and power shelves via Redfish, enabling intelligent power and cooling decisions at the rack, row, and room level. The AMI rack manager has a policy engine that orchestrates composability, power budgeting, and power sequencing across all pooled components — addressing a gap in the neo-cloud and enterprise market that hyperscalers have solved internally but others have not.