OpenAI and Broadcom unveil first custom AI chip for model inference
This story involves financial_misconduct. Any claims of wrongdoing described here are allegations, not established facts, unless a court or official body has ruled otherwise. This disclosure does not remove our responsibility for what's published below.
See sources Request a correction Report a serious problem / request takedown Corrections & takedown policy
Generated , updated since first publication on as new sources were added. Not reviewed by a human editor before publication.
OpenAI has launched a new silicon chip called Jalapeño in partnership with Broadcom to power large language models. The move aims to improve operational efficiency and reduce the company's reliance on third-party hardware providers.
Amalgamated from Slashdot (opens in new tab), Engadget (opens in new tab), The Verge (opens in new tab), Times of India (opens in new tab), TechCrunch (opens in new tab), Android Authority (opens in new tab), CNBC (opens in new tab)
OpenAI has unveiled its first custom artificial intelligence chip, named Jalapeño, developed in collaboration with Broadcom. The hardware is an Application-Specific Integrated Circuit (ASIC) designed specifically to handle inference for large language models, including those powering ChatGPT and Codex. According to reports from Slashdot and The Verge, the project is part of a strategic effort by OpenAI to build the full stack behind its products and services.
The design of Jalapeño focuses on inference rather than model training. While general-purpose processors offer significant flexibility across various applications, ASICs are engineered for specific functions. Reports from the Times of India indicate that these custom chips are intended to provide higher performance per watt compared to off-the-shelf options. OpenAI officials stated that by developing more of the stack internally, the company intends to serve intelligence with greater efficiency and expand access to advanced AI capabilities.
This shift also marks a significant change in OpenAI's hardware infrastructure. According to the Times of India, the move is intended to reduce the company's dependence on third-party manufacturers such as Nvidia. By producing its own silicon, OpenAI joins other major technology firms, including Google, Amazon, Microsoft, and Meta, which have all developed custom chips to power their respective AI infrastructures.
Initial deployments of the Jalapeño chip are expected by the end of 2026, with plans for a larger rollout continuing through 2027. The transition toward specialized hardware is aimed at optimizing the computational costs and energy requirements associated with running large-scale models in production environments.