OpenAI and Broadcom unveiled Jalapeño on June 24, 2026 — OpenAI's first custom silicon, engineered specifically for large language model inference rather than adapted from a general-purpose design. The chip was physically handed to OpenAI CEO Sam Altman and President Greg Brockman by Broadcom's top executives, marking what the companies called a key milestone in OpenAI's strategy to own more of its own infrastructure stack.
Developed from initial design to manufacturing tape-out in nine months, Jalapeño is believed to represent one of the fastest ASIC development cycles ever achieved in high-performance advanced semiconductors. OpenAI's own AI models assisted in the design and optimization process. Early lab results show performance-per-watt substantially better than current state-of-the-art accelerators, and Broadcom CEO Hock Tan cited cost savings of roughly 50% compared with typical AI GPUs. The chip is already running ML workloads in the lab — including GPT-5.3-Codex-Spark — at production target frequency and power.
The announcement places OpenAI alongside Google, which has long operated its custom TPU line, and Amazon, which produces its Trainium chips for AWS. Until now, OpenAI had relied almost entirely on Nvidia hardware. Jalapeño is explicitly designed for inference — the process of serving model outputs to users — while training workloads are expected to continue running on Nvidia hardware for the foreseeable future. Broadcom's Tomahawk networking silicon and manufacturing partner Celestica round out the hardware platform.
Initial deployment is targeted for the end of 2026, with volume scaling through 2027 and 2028 at gigawatt-scale data centers operated by Microsoft and other partners. "By designing more of the stack ourselves, we can serve more intelligence with greater efficiency," said Greg Brockman. The chip's economics could prove significant: as agentic AI products multiply interactions per user, inference costs increasingly determine which companies can sustain growth profitably.