Why OpenAI’s Jalapeño Might Sicken Nvidia
Exactly who at OpenAI decided to name the company’s first internally designed chip Jalapeño? That might have seemed like a good idea a few months ago—perhaps OpenAI imagined that name would serve as a subtle indicator that the chip is a hot property. But it’s not ideal right now, when the Food and Drug Administration is warning that products containing jalapeños shouldn’t be eaten, served or sold thanks to a salmonella outbreak tied to the peppers.
OpenAI must be hoping that’s not an omen, and so far it doesn’t seem to be. The AI firm on Tuesday unveiled the first test results for Jalapeño, claiming the chip showed “a significant performance advance: Jalapeño can serve more AI work per unit of power while also returning responses more quickly.” OpenAI CEO Sam Altman tweeted that “we made a chip and it is fast.” AI research firm SemiAnalysis put out a review claiming that Jalapeño is “better than Nvidia Blackwell.” Ouch.
Nvidia chips are of course the ones to beat in AI. It must particularly gall Nvidia CEO Jensen Huang that he’s a major financial backer of OpenAI. Nvidia committed $30 billion to OpenAI just a few months ago. One of the reasons Nvidia invests money in companies is to underwrite demand for its products!
OpenAI has a different agenda. As we reported in this story earlier this year, developing its own chip—which it did in conjunction with Broadcom—is a way for it to reduce hardware costs and improve its margins. Nvidia’s AI chips are pricey, after all. And it’s not just about the cash outlays for chips. OpenAI also hopes a chip specially designed for inference—running AI models—will operate more efficiently than Nvidia’s more general-purpose AI chips.
OpenAI noted today that it would continue to use Nvidia chips for both training and inference once it deploys Jalapeño, of course. But that’s a bit like the statements we often hear from Amazon and Google, both of which talk up their own AI chips while also acknowledging that they make Nvidia chips available in their cloud units for customers. The more easily Nvidia’s customers can replace its chips with alternatives, the worse things are for Jensen Huang. Perhaps he should remind people of the health dangers of eating jalapeños.