- OpenAI says Jalapeño delivered 1.5 to 1.9 times more AI work per watt and 1.7 to 3.6 times lower end-to-end latency than tested Nvidia GB200 and GB300 systems.
- Broadcom handled Jalapeño’s physical implementation and supplied Tomahawk networking; TSMC manufactured the chip and Celestica is handling board, rack and system integration.
- OpenAI and Broadcom plan to deploy 10 gigawatts of OpenAI-designed accelerators through 2029, extending Broadcom’s custom-chip opportunity beyond its existing AI semiconductor growth.
- Jalapeño targets inference rather than training, so it does not replace Nvidia’s CUDA ecosystem; production economics remain unproven ahead of expected year-end deployment.
Quotes
“OpenAI says its first homegrown inference chip can do up to 1.9 times more AI work per watt than the comparison systems, and it went from initial design to tape-out in nine months.”
“Jalapeño is not yet an Nvidia killer.”