OpenAI's Jalapeño Chip Outperforms Competitors in AI Inference Benchmarks
OpenAI has announced that its new AI chip, Jalapeño, outperformed Nvidia’s GB200 and GB300 superchips on an inference benchmark test. The ASIC was developed in partnership with Broadcom for use in AI inference tasks and offers improved efficiency and faster response times compared to existing systems. According to OpenAI hardware vice president Richard Ho, Jalapeño provides 1.5 to 1.9 times more work per watt while delivering up to 3.6 times lower latency than the tested Nvidia chips across multiple models including GPT-OSS 120B and Kimi K2.5 1T. The chip is expected to enable faster responses, better agent responsiveness, and improved scalability as AI demand grows. OpenAI plans limited deployment of Jalapeño by year-end with a gradual increase in production through 2027. Despite the performance gains, OpenAI will continue using other partners like Nvidia for its broader compute strategy.
Topics: AI chips benchmarking Jalapeño chip Nvidia OpenAI