OpenAI and Broadcom have unveiled Jalapeño, their first jointly developed custom chip, eight months after announcing their silicon partnership. The disclosure is OpenAI's first physical hardware milestone in its push to, as the company frames it, "build the full stack."
From Paper Deal to Named Silicon
The eight-month interval between the partnership announcement and Jalapeño's reveal is a meaningful signal about where the program stands. Custom chip development runs on long cycles: architecture work, design iterations, and manufacturing handoffs happen well before a chip surfaces in public. Reaching the point of a named, disclosed product means the OpenAI-Broadcom collaboration has cleared early milestones and is far enough along to commit to an identity. A name is also a form of accountability — Jalapeño is now a product that can be tracked and compared against what comes next.
What "Build the Full Stack" Means for Supply Chain
OpenAI's stated ambition is to own the full stack, a phrase that in hardware terms means controlling the compute substrate — not just the models that run on it. The practical difference between procuring compute from an existing silicon vendor and defining your own chip is structural: a company that designs its own silicon can tune it to specific workloads, manage its own capacity planning, and shift its position in the supply chain from buyer to principal. Broadcom is the development partner in that arrangement, handling the engineering collaboration that translates OpenAI's workload requirements into a manufacturable design.
What the Announcement Leaves Open
Jalapeño is a first joint project, not a deployed product. The announcement names the chip and confirms the collaboration without disclosing production volumes, target workloads, or timelines for when the silicon will run in data centers. That gap between a chip reveal and chips provisioned at meaningful scale is where custom silicon programs face the harder variables — yield, packaging, and allocation decisions that get made well outside the press cycle. OpenAI's full-stack thesis will be tested at that stage, not at this one.