OpenAI says its Jalapeño chip can power faster AI responses than the competition

JUST IN: OpenAI unveils its own AI chip 'Jalapeño', claiming it delivers lower latency and higher throughput than Nvidia's GB200/GB300 superchips, completing tasks more efficiently and returning responses faster.

OpenAI named its first chip Jalapeño, and says it beats Nvidia's best.

The numbers come from the company's own blog post today. On a benchmark called InferenceX, Jalapeño delivered one point five to one point nine times more AI work per watt than Nvidia's GB200 and GB300 superchips, and up to three point six times lower latency. Hardware boss Richard Ho calls it the best of both worlds, faster responses and more of them, which is usually a trade-off you don't get to skip.

The chip's an ASIC built with Broadcom, designed purely for inference, actually running models rather than training them. And inference is exactly where Nvidia prints most of its money. Sam Altman's biggest supplier is now also his benchmark punching bag, which makes for an awkward relationship.

One caveat I'd hold onto. OpenAI graded its own homework here. Until someone independent runs those tests, Jalapeño is a claim, not a crown.

This is Liu Yang, reporting for HIT.