OpenAI has unveiled its first custom AI chip, called Jalapeño, marking a major step towards building more of the hardware needed to power its artificial intelligence models. Developed in collaboration with Broadcom, the chip is designed to make AI systems faster, more efficient and less power-hungry.
OpenAI CEO Sam Altman celebrated the development by saying, “We made a chip and it is fast.” The company said Jalapeño has been built specifically for AI inference, the process through which trained models handle requests and generate responses.
The chip was tested using models including GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T. According to OpenAI, Jalapeño delivered 1.5 to 1.9 times more AI work per watt than comparison systems. It also achieved 1.7 to 3.6 times lower end-to-end latency, with performance for highly interactive workloads improving by up to 4.1 times.
OpenAI developed the chip alongside Broadcom, while Celestica handled boards, rack systems and production hardware. The company said designing the chip, software, memory, networking and serving systems around its models could improve efficiency across the entire AI infrastructure stack.
Jalapeño does not mean OpenAI is abandoning Nvidia. The company plans to deploy the new chip in small volumes by the end of 2026 before expanding its use through 2027. OpenAI will continue working with Nvidia and other hardware partners while developing future generations of its own AI chips.
Leave a Reply