OpenAI unveils Jalapeño, its first custom chip, built with Broadcom

OpenAI on Wednesday unveiled its first custom-built inference processor, designed and manufactured in collaboration with Broadcom. The chip is named Jalapeño, and the company says it was built specifically around the needs of OpenAI's inference systems. OpenAI's own AI models assisted in the development work, according to the announcement.

Early results and testing

The processor is still being tested. OpenAI says early results show significantly better performance-per-watt than current state-of-the-art alternatives, a measure that feeds directly into the cost of serving models at scale.

The Broadcom partnership was officially announced in October, but OpenAI's plans to build its own silicon had long been rumoured as a way to reduce its dependence on Nvidia's GPUs. Google and Amazon have both built custom chips for a similar purpose — so-called AI accelerators, designed to speed up machine-learning workloads.

Built for inference, not training

Jalapeño is designed for inference: running pre-built models in response to user requests. In its announcement, OpenAI highlighted the chip's low operating cost when running real-time coding models. More performance-intensive work such as pre-training will likely continue to rely on Nvidia hardware, but even modest reductions in inference costs can improve the economics of the business.

OpenAI president Greg Brockman described the company's approach to chip development on its in-house podcast shortly after the Broadcom partnership was announced. "We have a deep understanding of the workload," Brockman said. "We've really been looking for specific workloads that are underserved, [and asking] how can we build something that will be able to accelerate what's possible?"

Optimising across the stack

OpenAI is already building agentic products such as Codex, the models that power them, and the data centres that run those models. Purpose-built chips extend that work further down the stack. "OpenAI is not only developing frontier models or building products on top of them; it is designing the infrastructure underneath them: chip architecture, kernels, memory systems, networking, scheduling, deployment systems, and product experience," the company wrote. "Because OpenAI operates across the stack, each layer can be optimized around the same goal: making its models faster, more reliable, and more affordable for users."