OpenAI unveils Jalapeño custom AI chip with Broadcom, targeting late-2026 deployment
OpenAI and Broadcom on Wednesday unveiled Jalapeño, OpenAI's first custom AI inference chip, as part of a partnership to deploy 10 gigawatts of AI accelerators starting late 2026. The ASIC, designed in nine months, represents OpenAI's entry into vertical integration for AI infrastructure.
OpenAI and Broadcom on Wednesday unveiled Jalapeño, OpenAI's first custom artificial intelligence chip, designed for inference workloads that power ChatGPT and other applications [1]. The application-specific integrated circuit was developed in nine months and represents OpenAI's entry into vertical integration, enabling the company to "build the full stack behind its models and products," according to the press release [1]. A physical sample will be delivered to OpenAI on Wednesday, with initial deployment targeted by the end of 2026 [1].
The chip is part of a broader partnership announced in October 2025, under which OpenAI and Broadcom plan to deploy racks of custom silicon requiring 10 gigawatts of power starting late this year [1][2]. Greg Brockman, OpenAI's president, said the move allows the company to "serve more intelligence with greater efficiency and keep pushing advanced AI toward broader access" [1]. OpenAI has also signed chip supply agreements with Amazon Web Services for Trainium accelerators, Advanced Micro Devices, and Cerebras, which completed its initial public offering in May [1].
Continue reading
3 more paragraphs for subscribers.
Free readers get a daily preview and the weekly long-form. Premium unlocks every brief in full, AM & PM editions, alerts, and the archive.