OpenAI has pulled back the curtain on its first piece of custom silicon, a chip called Jalapeño designed in partnership with Broadcom — a step that pushes the company beyond software and deeper into the hardware that powers artificial intelligence.
Built for inference
The chip was unveiled on June 24 and remains in testing, TechCrunch reported. Jalapeño is an inference processor: it is built to run already-trained models in response to user queries — the work that happens every time someone types into ChatGPT — rather than to train new models from scratch.
OpenAI says the processor "was designed specifically for the unique needs of OpenAI's inference systems," and that early results show "significantly better performance-per-watt than current state-of-the-art alternatives." The company also said its own AI models helped design the chip. Those performance claims are OpenAI's own and have not been independently verified.
Jalapeño is not for sale to outside customers; it is an internal tool to run OpenAI's own products.
The Broadcom partnership
The reveal is the first product to emerge from a partnership the two companies announced in October 2025, under which they set a target of deploying 10 gigawatts of OpenAI-designed AI accelerators — a measure of computing capacity that points to data-center build-outs on an enormous scale. CNBC reported the chip's name and the partnership as OpenAI's first foray into bespoke processors.
Why build your own chip
The strategic logic is straightforward. Nvidia's graphics processors remain the dominant engine for AI workloads, and intense demand has kept them expensive and in short supply. By designing its own inference silicon, OpenAI aims to diversify its supply chain, gain more control over costs, and tailor the hardware to its own models in ways general-purpose chips cannot fully match.
OpenAI is far from alone. Google has run its own Tensor Processing Units since 2016, Amazon offers custom Trainium and Inferentia chips through its cloud, and Meta has developed its own AI accelerators. What sets OpenAI apart is that it is primarily a model developer rather than a cloud provider — making its move into custom silicon a notable bet that running its own hardware at scale will ultimately pay off.
For now, Nvidia's chips remain the backbone of OpenAI's model training, and Jalapeño does not change that. But if the company's efficiency claims hold up in production, OpenAI will have begun to reshape the economics of serving AI to hundreds of millions of users.
