OpenAI has officially unveiled Jalapeño, its first custom AI accelerator developed in partnership with Broadcom. The chip represents the company’s first major step into designing its own silicon, as it seeks greater control over the hardware infrastructure that powers services such as ChatGPT and its growing portfolio of AI products.
According to OpenAI, Jalapeño is designed specifically for inference workloads, which involve running trained AI models to generate responses and perform tasks. The company described it as its first “Intelligence Processor”, architected around its vision for the future of large language model (LLM) inference.
Late 2026 Deployment
By controlling more of the hardware and software stack, the company believes it can improve operational efficiency, reduce infrastructure costs, and deliver faster and more reliable AI experiences. The move also reflects a trend within the technology industry, where companies such as Google, Amazon and Meta have increasingly invested in custom silicon to reduce reliance on third-party suppliers while tailoring hardware for their own AI workloads.
OpenAI said Jalapeño is slated for initial deployment in data centres beginning in late 2026. The company and Broadcom also characterised the project as the first product in what is intended to become a multi-generation compute platform.
Built For AI Inference
Unlike general-purpose AI accelerators that cater to a broad range of machine learning applications, Jalapeño was developed with a singular focus on LLM inference. OpenAI said the chip is designed to maximise efficiency when serving AI models at scale, which has become increasingly important as usage of ChatGPT and other AI-powered tools continues to grow.
While full performance figures have yet to be disclosed, OpenAI claims that early testing shows Jalapeño delivers substantially better performance-per-watt compared to current state-of-the-art solutions. The company added that final validation is still underway, with a more detailed technical report expected in the coming months.
Developed In Just Nine Months With AI Help
One of the more notable aspects of the project is its development timeline. OpenAI said Jalapeño progressed from initial design to manufacturing tape-out in just nine months, an unusually rapid pace for a high-performance semiconductor project.
The company attributed this to close collaboration between OpenAI’s engineering teams and Broadcom’s silicon implementation experts. It also revealed that the firm’s own AI models were used to accelerate parts of the chip’s design and optimisation process.
Broadcom contributed its expertise in chip design and networking technologies, while Canadian hardware manufacturer Celestica assisted with board, rack and system integration. OpenAI noted that engineering samples are already operating in its laboratories, running workloads including GPT-5.3-Codex-Spark at target production frequencies and power levels.
