OpenAI is closing in on the finish line for its second-generation AI chip, according to Businesskorea, which reports that the design has entered its final stage.
The chip carries the name Jalapeno. In a report carried by MSN, OpenAI claims Jalapeno outperformed Nvidia's GB300 systems in key tests, and says the chip promises AI that is both faster and cheaper to run.
Those are OpenAI's own claims, not independently verified results, and the source items don't specify which tests were run or how the comparison was structured. That caveat matters, because benchmark claims in the chip industry are notoriously sensitive to how a test is set up.
Still, the direction of travel is significant. Nvidia currently supplies the overwhelming majority of the specialized processors that train and run large AI models, and its newest systems — the GB300 line that OpenAI says it beat — are the hardware most AI labs are competing to buy. Designing your own silicon is expensive and slow, which is why only the largest technology companies attempt it. Reaching the final stage of a second-generation design suggests OpenAI has moved past the experimental phase and is building a real hardware program.
The economics are the point. Running AI models is enormously costly, and the bill is dominated by chips. A processor OpenAI controls could lower what it pays per unit of computing power and reduce its dependence on a single outside supplier that sets prices and controls allocation.
It matters because if OpenAI's claims hold up under outside scrutiny, the company that defined the modern AI boom would no longer be entirely at the mercy of the company that has been selling it the shovels.