AI Radar

Noteworthy AI conversations and writing, summarized from publisher notes.

OpenAI x Broadcom: Custom Silicon Partnership for 10 GW of Compute

· OpenAI Podcast

OpenAI announces a major partnership with Broadcom to co-design custom AI accelerators (XPUs) and full-stack systems, with deployment of 10 gigawatts of new compute capacity beginning late next year and ramping over the following three years. The collaboration, which began roughly 18 months ago, started with chip IP and expanded to encompass the entire system—chips, racks, networking, and algorithms—enabling OpenAI to vertically integrate from transistor to output token. Greg Brockman highlighted that OpenAI used its own models to optimize chip designs, achieving substantial area reductions and accelerating timelines, while Sam Altman framed the 10 GW figure as merely 'a drop in the bucket' relative to the compute needed for OpenAI's AGI vision.

AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.