Arm‑Meta AGI CPU leak
A new Arm‑Meta AGI CPU demo shows a 136‑core efficiency design aimed at reasoning and learning rather than chat — explicitly positioned to challenge x86 for agentic workloads. Early clips and commentary highlight the chip’s efficiency and its focus on non‑chat AGI tasks. (x.com)
Arm revealed the Arm AGI CPU at its Arm Everywhere event in San Francisco on March 24, 2026. (techcrunch.com) The flagship device uses up to 136 Arm Neoverse V3 cores spread across two dies and is fabricated on TSMC’s 3nm process. (siliconangle.com) Arm’s product brief specifies about 6 GB/s of memory bandwidth per core and sub-100 ns memory latency targets for the part. (convergedigest.com) The company lists the top SKUs at roughly a 300 W TDP and says the design assigns a dedicated core per program thread to preserve deterministic throughput under sustained load. (videocardz.com) Arm projects rack-scale densities of up to 8,160 cores in air-cooled configurations and more than 45,000 cores with liquid cooling, and claims the platform delivers over 2x performance per rack compared with x86 in its materials. ( ) Industry summaries and early coverage have cited approximate ~50% power-efficiency improvements versus x86 in preliminary comparisons and market writeups. (tech-insider.org) Meta is named as the lead partner and first deployer for the AGI CPU, with OpenAI and other cloud and ODM partners listed among launch collaborators committed to production. ( ) Arm framed the announcement as its first production silicon in the company’s 35-year history and a strategic move from licensing IP toward selling finished CPUs, a shift emphasized by CEO Rene Haas at the reveal and reflected in immediate positive market reaction. ( ) Early event footage and social clips — including an X post highlighting a hands-on demo and Arm’s own introductory video — circulated alongside the technical brief, with commentators focusing on the chip’s sustained-density and orchestration-oriented design rather than conversational-chat benchmarks. ( )