Particle.news

Download on the App Store

OpenAI Seals Broadcom Pact to Co‑Develop 10 GW of Custom AI Accelerators Starting in 2026

The tie-up advances OpenAI’s shift to bespoke silicon to secure efficiency gains at vast scale.

Overview

  • OpenAI and Broadcom signed a multi‑year agreement to build and deploy 10 gigawatts of OpenAI‑designed accelerator racks and networking, with rollouts scheduled from the second half of 2026 through 2029.
  • Under the deal, OpenAI designs the chips and systems while Broadcom develops, manufactures, and installs full racks using its Ethernet‑based networking for scale‑up and scale‑out clusters.
  • Broadcom’s shares jumped roughly 10%–13% on the news, adding more than $150 billion in market value, as investors bet on the company’s deeper role in AI infrastructure.
  • The move extends a recent spree of capacity deals that includes OpenAI’s multi‑GW agreement with AMD featuring performance‑linked warrants, Nvidia’s letter of intent for at least 10 GW with up to $100 billion of staged investment, and a reported $300 billion Oracle cloud arrangement.
  • The scale underscores resource and oversight challenges, with 10 GW comparable to the power needs of over 8 million U.S. households and analysts warning about financing complexity, market concentration, and circular deal structures.