OpenAI has developed a custom chip, codenamed Jalapeño, that reportedly matches Nvidia's Blackwell architecture in performance while delivering a 50% cost advantage, according to Broadcom's CEO. The chip could disrupt the AI hardware market and cut into Nvidia's grip on GPU supply.
The Jalapeño Chip
Jalapeño is a custom processor designed by OpenAI, built specifically for AI workloads. The claim of performance parity with Nvidia's Blackwell line is a strong one. Blackwell is Nvidia's latest GPU architecture powering many of today's large-scale AI training and inference deployments. If Jalapeño holds up to that standard, it would give OpenAI a serious alternative to buying Nvidia hardware at a fraction of the price.
A 50% Cost Edge
Broadcom's CEO made the cost claim public, saying the chip delivers a 50% advantage over Nvidia Blackwell. That's a huge margin in a market where GPU costs dominate AI infrastructure budgets. A 50% reduction would make a real difference in how much OpenAI spends on compute, and it could put pressure on Nvidia's pricing model.
The specifics of how Jalapeño achieves this cost gap haven't been detailed. It could come from a smaller die, different memory architecture, or simply a better manufacturing deal. Whatever the cause, the claim puts Nvidia on notice: the AI hardware market isn't a one-horse race anymore.
If Jalapeño really matches Blackwell, OpenAI could deploy its own chips across data centers and reduce the number of Nvidia GPUs it needs. That would be a direct hit to Nvidia's dominance in AI accelerators. OpenAI is one of the biggest buyers of Nvidia hardware, so even a partial shift would ripple through the supply chain.
Other AI companies might follow suit if the chip proves itself. That would chip away at Nvidia's margins and force it to answer with better prices or faster innovation. For now, Nvidia has not responded to the claim.
What Happens Next
The chip is in development, but no release date has been announced. It's unclear if Jalapeño will ever ship in volume, or if it's just a internal experiment. Broadcom's CEO, who's not a neutral observer, made the claim — so there's room for skepticism. But if the cost advantage holds up in production, OpenAI could start switching over within a couple of years.
Right now, the big question is whether OpenAI will actually build Jalapeño into its data centers, and how Nvidia responds if it does. Neither company has given a timeline.
