OpenAI says its Jalapeño chip outperformed Nvidia in AI inference tests

OpenAI said its custom Jalapeño chip delivered faster and more efficient AI inference in tests than Nvidia’s current high-end systems, marking an early sign that the company’s in-house hardware efforts may reduce reliance on outside chip suppliers. The results, based on public benchmarking, showed better performance per watt and lower response times on models such as GPT-OSS, DeepSeek R1, and Kimi K2.5. The claims matter because AI companies are racing to cut the cost and energy use of running large models, and Nvidia remains the dominant supplier for that work.