In a significant development for the field of artificial intelligence, OpenAI and Broadcom have announced the design of Jalapeño, their first jointly-developed chip. This custom AI accelerator is specifically designed to run OpenAI's large language models, with a focus on optimizing performance per watt. According to OpenAI, Jalapeño is already showing promising results, with performance per watt substantially better than current state-of-the-art chip technology. While final performance testing is still underway, the company plans to release a more detailed technical report in the coming months. The Jalapeño chip is slated for initial deployment in data centers in late 2026, marking OpenAI's first foray into chip production. By controlling more components of the process, OpenAI aims to increase efficiency and expand the usability of its ChatGPT tools. The development of Jalapeño was a collaborative effort, with OpenAI and Broadcom working together from initial design to manufacturing tape-out in just nine months. This rapid development was made possible by deep software-hardware co-development and the use of OpenAI models to accelerate parts of the design and optimization process. As the first AI accelerator from the Broadcom and OpenAI partnership, Jalapeño represents a significant step towards creating a multi-generation compute platform that can make advanced AI faster, more reliable, and more accessible to a wider audience.

Source: Engadget