Particle.news

Download on the App Store

OpenAI Strikes Broadcom Deal to Build 10 GW of Custom AI Accelerators

The co-design pact gives OpenAI tighter control over hardware to curb costs, reducing single‑supplier dependence.

Overview

  • OpenAI and Broadcom signed a term sheet for a multi‑year collaboration to co‑develop and deploy racks of OpenAI‑designed accelerators integrated with Broadcom’s Ethernet networking.
  • Deployments are scheduled to start in the second half of 2026 and roll out through 2029 across OpenAI facilities and partner data centers.
  • Broadcom’s shares jumped about 10%, adding roughly $150 billion in market value after the announcement, while financial terms were not disclosed.
  • The 10‑gigawatt target underscores the scale of next‑gen AI infrastructure, with reports equating the power draw to a major city’s peak demand.
  • The agreement extends OpenAI’s rapid 2025 compute buildout, complementing a ~10 GW Nvidia framework and a ~6 GW AMD deal with warrants, even as financing, supply‑chain capacity, and energy availability remain key risks.