Overview
- OpenAI will design the chips and systems, while Broadcom will develop, manufacture, and integrate full accelerator racks built on its Ethernet stack.
- Initial racks are slated to ship in the second half of 2026, with large-scale deployment extending toward 2029.
- Financial terms were not disclosed, and Broadcom shares rose roughly 9% following the announcement.
- The deal complements OpenAI’s multi-vendor strategy, including a 6 GW AMD supply agreement with a contingent equity option and large-scale arrangements with NVIDIA and cloud partners such as Oracle and CoreWeave.
- OpenAI operates a little over 2 GW of compute today and has stated far larger ambitions, with reports citing a long-term target on the order of 250 GW.