Particle.news
Download on the App Store
66 ARTICLES
·
3h ago

OpenAI Taps Broadcom to Co-Develop 10 GW of Custom AI Accelerators

The pact gives OpenAI more control over its compute stack, with Broadcom responsible for manufacture, Ethernet networking integration and deployment starting in 2026.

AI (Artificial Intelligence) letters and robot hand are placed on computer motherboard in this illustration created on June 23, 2023. REUTERS/Dado Ruvic/Illustration/File Photo
Image
Image
A smartphone with a displayed AMD logo is placed on a computer motherboard in this illustration taken March 6, 2023. REUTERS/Dado Ruvic/Illustration

Overview

  • OpenAI and Broadcom confirmed a multiyear collaboration to design and deploy custom accelerator racks totaling 10 gigawatts of capacity across OpenAI and partner data centers.
  • Rollouts are slated to begin in the second half of 2026 and continue through 2029 under a signed term sheet that formalizes earlier co-development work.
  • OpenAI will design the chips and systems, while Broadcom will build and deploy full racks that use its Ethernet-based networking for scaling within and across clusters.
  • Broadcom shares jumped roughly 10%, adding about $150 billion in market value after the announcement, reflecting expectations for a larger role in AI infrastructure.
  • The deal expands OpenAI’s multi-vendor buildout alongside recent Nvidia, AMD and Oracle agreements, drawing scrutiny over funding, circular financing risks and supply-chain constraints in foundry capacity, advanced packaging, memory and power.