Overview

• OpenAI and Broadcom will co‑develop and deploy racks of OpenAI‑designed accelerators with Broadcom networking and memory built on its Ethernet stack, targeting roughly 10 gigawatts of compute.
• Initial deployment is slated to begin late next year, formalizing about 18 months of prior joint development work.
• Broadcom shares rallied following the announcement, with market reports saying the move added roughly US$150 billion to its valuation.
• OpenAI says it used its own AI models to optimize chip layouts, achieving what President Greg Brockman called massive area reductions.
• The deal slots into OpenAI’s broader multi‑vendor buildout, lifting announced compute commitments to about 33 gigawatts beyond existing Nvidia and AMD arrangements.