We made a chip and it is fast, Sam Altman says after OpenAI takes on Nvidia with first AI chip

San Francisco (TIP): OpenAI is moving beyond building AI models. The company is now designing and manufacturing the hardware that runs them too, and its latest effort is called Jalapeo, a custom chip built specifically for running AI models in collaboration with Broadcom. In a blog post, OpenAI said the chip is designed to deliver more AI work while using less power and reducing response times. The company tested it with models including GPT-OSS 120B, DeepSeek R1 and Kimi K2.5 1T.
OpenAI said the chip addresses a common challenge in AI hardware: achieving high performance without increasing latency. “Jalapeo is designed to deliver high performance and low latency at the same time,” the company said.
The keyword here is inference. In simple terms, inference is what happens when a trained AI model processes a request and generates an answer. Every time you ask ChatGPT a question or use an AI coding agent, inference is taking place.
Jalapeo is designed specifically for this task. OpenAI developed it with Broadcom, while Celestica worked on the boards, rack systems and production hardware. OpenAI said the chip was designed around the way its AI models operate.
“By co-designing the chip, software, memory, networking and serving systems around our models, we can improve performance and efficiency across the entire stack,” the company said. OpenAI said Jalapeo delivered 1.5 to 1.9 times more AI work per watt than the comparison systems across the three tested models. It also recorded 1.7 to 3.6 times lower end-to-end latency. For highly interactive workloads, the company said performance was up to 4.1 times higher.
In practical terms, OpenAI wants Jalapeo to process more requests without using as much electricity, while also reducing the time users wait for responses. For AI systems handling millions of requests, even modest efficiency improvements can lead to significant savings.
Jalapeo is part of OpenAI’s broader push to control more of the technology stack behind its AI. Instead of designing models and then simply relying on third-party hardware to run them, OpenAI wants to optimise the chip, software, memory, networking and serving systems together.