Particle.news
Download on the App Store
114 ARTICLES
·
41m ago

OpenAI Taps Broadcom for 10 GW of Custom AI Chips as Oracle Sets 50,000‑GPU AMD Rollout

With rollouts not due until 2026, the scale of the commitments shifts attention to execution, power supply, financing.

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
Larry Fink AI
Image
Image

Overview

  • OpenAI will design custom accelerators and systems while Broadcom develops, builds and deploys racks targeting 10 gigawatts of capacity from 2026 through 2029.
  • Broadcom shares jumped sharply after the announcement, adding roughly $150 billion in market value according to market reports.
  • Oracle Cloud Infrastructure said it will begin deploying 50,000 AMD Instinct MI450 GPUs in the second half of 2026, signaling expanded options beyond Nvidia.
  • The Broadcom pact joins recent agreements that include a multiyear AMD deal for up to 6 gigawatts and a September letter of intent tied to 10 gigawatts of Nvidia systems.
  • Investors and analysts highlight risks including circular financing links, massive capex and power constraints, with some reports estimating the Broadcom effort could cost OpenAI $350 billion to $500 billion, which the companies have not disclosed.