Enterprise

OpenAI unveils 'Jalapeño,' its first custom inference chip, built with Broadcom

The ChatGPT maker says the in-house ASIC was designed in nine months and is showing roughly 50% cost savings versus standard AI GPUs, per Broadcom CEO Hock Tan.

By Asher Glick · Enterprise software desk · 2026-06-30 · 3 min read

Photo: Unsplash / Igor Omilaev — Close-up render of an AI processor chip on a circuit board

OpenAI and Broadcom unveiled Jalapeño on June 24, an inference-only ASIC co-developed with contract manufacturer Celestica that Broadcom CEO Hock Tan says delivers roughly 50% cost savings versus typical AI GPUs. It’s OpenAI’s first custom Intelligence Processor, and its arrival completes the migration of every frontier AI lab into the silicon business.

The headline engineering claim is the cycle time. OpenAI and Broadcom say Jalapeño went from initial design to manufacturing tape-out in nine months, possibly the fastest ever for a high-performance ASIC. OpenAI’s own models did the compression. “the degree to which our models have been able to accelerate it was very surprising to us,” President Greg Brockman told CNBC, a remark that doubles as the cleanest available evidence that recursive self-improvement on chip design is already a production workflow rather than a thought experiment.

Jalapeño is purpose-built for inference, the workload of running existing models against user prompts. The more performance-intensive job of pre-training is expected to remain on Nvidia hardware. This is the same bifurcation Google quietly normalized with TPUs and that Amazon has pursued with Trainium and Inferentia: custom silicon for the bulk economics of serving, merchant GPUs for the experimental edge.

The timeline is staged. Tan described “small prototype development” by the end of 2026, a ramp through 2027, and deployment going “full tilt” in the first half of 2028. Broadcom separately said the platform is intended to support gigawatt-scale data centers with Microsoft and other partners starting in 2026.

The strategic logic is scarcity. Brockman said OpenAI “cannot get compute fast enough.” Tan, speaking about Broadcom’s six custom-silicon customers, called demand “simply insatiable,” adding, “not just ‘26, not ‘27, we’re seeing that same and even elevated demand in ‘28 as well.” Broadcom shares are up about 10% in 2026 and have risen almost sevenfold since the end of 2022, a re-rating driven almost entirely by the hyperscaler custom-chip business.

Engineering samples are already running GPT-5.3-Codex-Spark in OpenAI’s lab at production target frequency and power. A full technical report is still months away. The 50% number is the one investors are pricing.

Sources