Intel Computex 2026 Keynote

Author: Lip-Bu Tan (CEO, Intel), with Alex (Client Compute & Physical AI), Kavoke (Data Center), Sini (Purpose-Built Silicon), and multiple partner executives Collection: Computex 2026

Summary

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. The session organized Intel's portfolio across four compute ecosystems — Personal Computers, Edge & Physical AI, Data Centers, and Intelligence Centers — and demonstrated that the same Intel 18A process technology and x86 IP base now spans all four. Key product news included the launch of Intel Core Ultra Series 3 (the first 18A process product, shipping with 300+ designs), the Intel Core Series 3 mainstream processor (already at 70+ designs just weeks after April introduction), the Arc G3 GPU for handheld gaming (40% faster than competition, half the power at parity performance), and the Intel Xeon 6 Plus with 288 E-cores (efficiency cores) and 576 MB of L3 cache on 18A — positioned explicitly for agentic AI workloads where CPU-to-GPU ratios are moving from 1:7 toward near parity.

The data center segment introduced the concept of the "Agentic Compute Shift," backed by a live demo comparing traditional inference (7:1 GPU-to-CPU ratio) against an agentic pipeline (1:1.3 CPU-heavy ratio) running on Xeon 6 Plus. Intel announced rack-scale blueprints built on open standards with three pillars: Open Standards, Turn-Key Solutions, and Heterogeneous Infrastructure. A partnership with SambaNova Systems demonstrated disaggregated heterogeneous inference — Xeon CPUs handling tooling and orchestration, SambaNova RDUs (Reconfigurable Dataflow Units) handling token decode, and Nvidia GPUs performing prompt caching and prefill — achieving 2–3x faster end-to-end latency versus GPU-only stacks, with Vista Equity Partners committing as a launch customer via the Vector Core Compute VC2 platform.

The closing segment unveiled Intel's entry into purpose-built silicon as a foundry capability for external customers. Announced customer designs include an Infrastructure Processing Unit (IPU) for Google at hyperscale and next-generation wireless infrastructure silicon for Ericsson — both already in deployment. Partners for emerging applications include Echo Neuro Technologies (brain-trained algorithms for speech), Greenstone Biosciences (world's largest iPSC bio bank for drug discovery), Hitachi (industrial AI and quantum computing), and Siemens (EDA automation and manufacturing). Intel also revealed its partnership with McLaren Racing as official compute partner.