OpenAI and Broadcom announce a custom artificial intelligence chip named Jalapeño, described as OpenAI’s first in-house processor for powering large language model workloads. Multiple outlets report that the chip is built as an ASIC, meaning it is designed for a specific purpose rather than general computing. Sources characterize Jalapeño as focused on AI inference—running models for results—rather than training. OpenAI positions the chip around efficiency and cost reduction, with claims of better performance per watt and, in one report, the potential to cut costs by about 50% compared with existing options.
According to the reporting, OpenAI is developing and testing chip samples after the reveal, and the company’s broader strategy is to expand a “full stack” around its models and products. The outlets also frame the effort as part of a larger industry move toward custom silicon among major cloud and AI players. Timelines vary by outlet, but at least one source says early deployments are expected by year’s end and scaling could continue through 2027.
