OpenAI Unveils In-House Jalapeño AI Chip

OpenAI has debuted Jalapeño, its first custom AI chip developed in-house with Broadcom's backend support for training and inference tasks.

OpenAI has officially introduced Jalapeño, its first custom-designed AI chip developed entirely in-house. While the design was created by OpenAI, the company utilized Broadcom for backend support to transfer the architecture to manufacturing, a strategy also employed by Google and Tensordyne. The Jalapeño chip is engineered to power large language model workloads, including ChatGPT, Codex, the OpenAI API, and upcoming AI agents. Preliminary observations of the design indicate it features one compute chiplet, one I/O chiplet, and three HBM3E stacks. Although OpenAI has not yet disclosed specific technical specifications, the hardware is intended for both the training and inferencing of AI models. This move marks a significant step in OpenAI's infrastructure development for current and future AI applications.