OpenAI Unveils Jalapeño, Its First Inference Chip

OpenAI and Broadcom announced Jalapeño on Wednesday. It is OpenAI’s first custom AI inference chip and was developed specifically to run large language models. Early versions of the chip are in lab testing with models including GPT-5.3-Codex-Spark.

The chip is built to handle inference workloads, the process that generates responses for chatbots and other language-model applications. OpenAI said the accelerator was designed to increase compute capacity while reducing the energy required for those inference tasks.

Jalapeño is focused on inference rather than on training or general-purpose data-center computing. The company claims the chip can deliver greater computing power while using less energy than current leading AI accelerators, but it has not released benchmark data to support those claims.

OpenAI described Jalapeño as the first product in a multi-generation compute platform. The company expects the platform to begin appearing in data centers later this year, with later generations aimed at supporting gigawatt-class facilities in cooperation with cloud partners.

OpenAI President Greg Brockman said, “Jalapeño is part of our long-term, full-stack infrastructure strategy to make compute more abundant, resulting in AI which is faster, more reliable, more affordable for people and businesses, and can be used to solve more important problems.”

Broadcom is the project’s co-developer. Broadcom President and CEO Hock Tan called the collaboration “a fundamental commitment to scaling the physical infrastructure required for the next decade of AI” and added the partners are preparing for gigawatt-scale deployments beginning in 2026.

The project follows months of internal planning aimed at reducing reliance on third-party GPUs and diversifying the hardware that supports OpenAI’s services. Industry observers expect published benchmarks and real-world deployment data to be released before claims about performance and energy efficiency can be verified.