OpenAI unveils Jalapeño, its first custom AI chip with Broadcom — a bid to own the full stack
OpenAI and Broadcom revealed Jalapeño, a custom LLM-optimized inference accelerator, targeting initial deployment by the end of 2026 and 10 gigawatts of capacity by 2029. A frontier lab designing its own silicon is the clearest sign yet that owning the compute stack — not just renting it — is now a strategic necessity at the top of the market.
Designing its own chip changes what OpenAI is. Renting GPUs makes you a customer; taping out a custom inference accelerator makes you an infrastructure company that happens to make models. Jalapeño — optimized specifically for serving large language models rather than general training — is OpenAI declaring that the economics of inference at its scale justify owning the silicon, not just the model that runs on it.
The Broadcom partnership is the how. Broadcom's custom-ASIC expertise, already behind Google's TPUs, lets OpenAI reach its own silicon without becoming a chip fab — a route that trades some control for speed to market. The 10-gigawatt target by 2029 is the scale that makes the multi-year design investment pay off; below that, buying merchant GPUs stays cheaper.
The strategic signal is a break from single-supplier dependence. A lab whose entire business runs on one vendor's accelerators is exposed to that vendor's pricing and roadmap; a custom chip is leverage and insurance at once. Jalapeño is OpenAI buying its way out of being purely a tenant on someone else's compute — the same vertical-integration move that reshaped every prior computing era.
OpenAI — OpenAI and Broadcom unveil LLM-optimized inference chip → · CNBC — OpenAI unveils first chip as part of Broadcom deal in effort to build the full stack →