Launch of Jalapeño
On June 24, 2026, OpenAI and Broadcom introduced Jalapeño, OpenAI’s first Intelligence Processor specifically designed for large language model (LLM) inference. This chip is a key component of a planned multi-generation compute platform aimed at enhancing the performance and accessibility of advanced AI.
Design and Development
Jalapeño was developed from the ground up, leveraging OpenAI’s deep understanding of LLM fundamentals, including kernels, memory movement, and networking patterns. Broadcom contributed its expertise in silicon implementation and networking technologies, such as Tomahawk networking silicon, while Celestica provided board, rack, and system integration support. OpenAI claims that the chip progressed from initial design to manufacturing tape-out in just nine months, which they describe as potentially the fastest ASIC development cycle achieved in high-performance advanced semiconductors. The design process was accelerated by utilizing OpenAI’s models to optimize various aspects of the chip.
Testing and Performance
Engineering samples of the Jalapeño chip are currently running machine-learning workloads in laboratory settings at target production frequency and power, including tests with GPT‑5.3‑Codex‑Spark. While OpenAI continues to evaluate the final performance metrics, early tests indicate that Jalapeño will deliver performance per watt significantly better than current state-of-the-art solutions. A comprehensive technical report detailing performance metrics is expected to be released in the coming months.
Future Deployment Plans
Broadcom’s CEO, Hock Tan, emphasized that this collaboration marks the beginning of a multi-generation roadmap, with plans to enable the deployment of gigawatt-scale data centers in partnership with Microsoft and others starting in 2026. OpenAI anticipates initial deployment of Jalapeño by the end of 2026, with further expansions planned in subsequent years.
Strategic Importance
OpenAI positions Jalapeño as a critical element of its full-stack strategy, which encompasses chip architecture, kernels, memory systems, networking, scheduling, and deployment systems. This strategy is designed to support interactive LLM products at scale. Unlike general-purpose accelerators, Jalapeño is specifically tailored for inference workloads, aiming to combine high throughput with low latency, making it suitable for large-scale interactive applications.
OpenAI and Broadcom are committed to evolving this platform over multiple generations, with a public technical report and additional deployment details to be shared within the established timelines.
Original source: OpenAI News