OpenAI and Broadcom have a new chip, and it has a name that sounds more fun than its job description.
The two companies announced Jalapeño, a custom silicon design built specifically for running large language model inference in data centers. This is not a general-purpose GPU — it is purpose-built for the kind of workload that powers ChatGPT and Codex at scale. Both companies framed it as the first generation of a longer collaboration, meaning the hardware will be iterated on over time rather than treated as a one-off product.
Why it matters: designing your own inference chip is how you stop renting compute from Nvidia at whatever price Nvidia decides to charge. OpenAI has been one of the largest consumers of GPU infrastructure on the planet, so a proprietary silicon path — even a first-gen one — gives it leverage it did not have before. Broadcom brings the manufacturing and supply chain experience that a software company building its first chip badly needs.
The announcement joins a crowded field: Google has been running its own TPUs for years, Amazon has Inferentia, and Microsoft has been investing in custom silicon too. OpenAI arriving here in 2026 is not early — but a chip named Jalapeño is at least a memorable entry.