OpenAI said its self-developed Jalapeno AI chip, developed with Broadcom, outperformed Nvidia's GB300 in tests on two measures: processing capability per unit of power consumption and response speed. The chip consumes about 700 watts and is designed for AI inference rather than model training. OpenAI plans to begin using Jalapeno later this year. In public testing, the company used one of its smaller open-source models, as well as third-party models from DeepSeek and Moonshot AI, with the strongest advantage seen on the Kimi model. OpenAI said the second-generation chip has entered late development and is expected to complete tape-out in the coming months, while stressing that Nvidia remains an important chip supplier.