Hypernovelty Institute · Global Chess Map
Broadcom–OpenAI: 10 GW of custom accelerators
Recorded event: 2025-10-13
OpenAI and Broadcom signed a term sheet to deploy 10 GW of OpenAI-designed accelerators and Broadcom networking, with rack deployments targeted to start in H2 2026 and finish by the end of 2029. Broadcom unveiled Jalapeño in June 2026. In August, OpenAI reported chip and system tests, with production qualification and preparation for scaled operation continuing and deployment planned by year-end.
Record over time
- 2026-06-24 · Report: 2026-06-24 · OpenAI and Broadcom unveiled Jalapeño, OpenAI's first custom inference accelerator, reporting lab operation of engineering samples.
- Event date not established · Report: 2026-08-25 · OpenAI reported Jalapeño chip and system benchmark tests and said production qualification and preparation for scaled operation were continuing, with deployment planned by year-end 2026.
Tracked promise
Begin deploying OpenAI–Broadcom racks in H2 2026; complete by end of 2029.
Deadline: not established · unclear
Limitations
- Foundry and packaging allocation for the chips is not disclosed.
No new connection is implied by this narrative record. Only separately reviewed connections are displayed.