OpenAI Unveils Jalapeño Chip: A Breakthrough in Custom Silicon for AI Efficiency

OpenAI has made significant strides in the development of custom AI chips with its Jalapeño chip, which outperforms commercial systems in AI inference efficiency. The company announced measured results for Jalapeño, demonstrating higher peak throughput per kilowatt and lower token latency on benchmarks like GPT-OSS 120B, DeepSeek R1, and Kimi K2.

The Jalapeño chip strengthens OpenAI's ability to optimize AI workloads across training, inference, and deployment. By integrating software, hardware, and network design, the company claims it can deliver better energy efficiency and lower costs, key factors as AI demand continues to surge.

Custom silicon like Jalapeño is part of OpenAI's broader approach to maintain leadership in AI economics. The company leverages a mix of proprietary technology and partnerships with major players such as Microsoft, NVIDIA, AWS, AMD, and others.