OpenAI Unveils First Custom LLM Chip Built with Broadcom
In Focus
- OpenAI’s AI chip is called Jalapeño
- The new chip is designed to work efficiently with LLMs that power AI applications
- Engineers designed the AI processor over a nine-month period
- The new chip meets target power levels in OpenAI’s GPT-5.3-Codex-Spark model
OpenAI has unveiled its first custom AI chip designed in partnership with Broadcom. The new AI chip is designed to support AI inference tasks. OpenAI’s LLM chip comes at a time when the AI developer is seeking to fast-track AI infrastructure development.
Why has OpenAI Developed a Custom AI Chip?
Increasingly, AI developers such as OpenAI and Anthropic are experiencing difficulties obtaining the compute capacity they need to run advanced AI systems that power coding apps and chatbots. To address this challenge, OpenAI decided to build an in-house chip.
Working with Broadcom, OpenAI engineers designed a chip called Jalapeño. The chip lowers compute power costs for the company while offering an alternative to NVIDIA’s AI processors.
“The world is moving to a compute-powered economy. Jalapeño is part of our long-term full-stack infrastructure strategy to make compute more abundant, resulting in AI which is faster, more reliable, more affordable for people and businesses,” OpenAI Co-founder and President, Greg Brockman noted.
OpenAI’s partnership with Broadcom to design and produce custom chips was first announced in September 2025. At the time, the AI developer said the initial batch of the chips would start shipping in 2026.
How Did OpenAI Build an AI Chip?
OpenAI’s Broadcom chip is built to perform inference tasks, which involves generating responses to user queries in chatbots like ChatGPT. OpenAI said the new chip is designed to work quickly and efficiently with large language models (LLMs) that power AI applications. It’s capable of improving the speed and performance of AI systems.
“Jalapeño was designed from the ground up for LLM inference using detailed insights from our close collaboration with OpenAI researchers. We optimized the architecture around the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models. Based on early testing, Jalapeño will efficiently execute our most important workloads close to the hardware’s theoretical limits,” OpenAI’s Hardware Program Lead, Richard Ho said.
OpenAI added that the capabilities of OpenAI’s LLM inference chip matches those of NVIDIA’s Blackwell processors or Google’s Tensor Processing Units (TPUs). OpenAI engineers took about nine months to complete the Jalapeño chip design.
AI tools helped the engineers to fast-track parts of the design process. The AI processor has already been sent to TSMC for production. Google has also partnered with Broadcom to design its in-house chips.
How Does the AI Chip Perform in AI Models?
OpenAI said samples of the new chip are already running in its labs. The company added that the chips are meeting target power and performance levels when used with the GPT-5.3-Codex-Spark AI model.
As the company prepares to deploy the new chips, Celestica, the Canadian electronics manufacturer, will build the server systems. Like Jalapeño, these systems will be used exclusively by OpenAI. Meta and Amazon are also developing their own in-house chips.
