What is Jalapeño? OpenAI unveils first custom AI chip with Broadcom for ChatGPT, Codex

OpenAI has introduced its first custom-built artificial intelligence chip, Jalapeño, marking the company's expansion beyond AI applications into core computing infrastructure.Developed in partnership with Broadcom, the processor is designed to run inference workloads powering ChatGPT, Codex, OpenAI's API and future AI agents. According to the company, the chip has been built specifically for large language model workloads rather than as a general-purpose processor.OpenAI said engineering samples are already running machine learning workloads in its labs at the intended production frequency and power levels, including GPT-5.3-Codex-Spark. While final performance evaluations are still underway, the company said early tests indicate Jalapeño delivers substantially better performance per watt than current leading alternatives.Broadcom Chief Executive Officer Hock Tan said the chip is currently being tested by OpenAI after the first samples were delivered. He added that initial results indicate roughly 50 percent cost savings compared with conventional AI graphics processing units.The companies first announced their custom chip partnership last year with plans to build processors capable of supporting 10 gigawatts of computing. Jalapeño is the first product from that collaboration and is expected to be deployed by the end of 2026, with additional generations planned in the coming years.OpenAI said Jalapeño was designed from scratch using its understanding of large language models, model roadmaps, serving systems and product requirements. Broadcom collaborated on the chip's development, while Celestica supported industrialising the platform.In a post on X, OpenAI said, "We've designed and built our first AI chip: Jalapeño." The company added that developing its own chips expands its platform from products and models into infrastructure and will help it scale AI services while broadening access.OpenAI also said the chip moved from initial design to manufacturing tape-out in nine months, describing it as the fastest ASIC development cycle for high-performance advanced semiconductors. The company said it used its own AI models to accelerate parts of the chip's design and optimisation process.The announcement reflects a broader industry trend in which major technology companies are designing custom processors tailored for their own AI systems. Such chips are intended to improve efficiency, reduce operating costs and lessen dependence on third-party hardware.The launch comes as OpenAI prepares for an initial public offering that could value the company at around $1 trillion, increasing the focus on expanding revenue and strengthening its AI infrastructure.Greg Brockman, OpenAI's co-founder and president, said designing more of the technology stack internally would allow the company to deliver AI more efficiently and make advanced capabilities accessible to more users.