OpenAI launches custom-built Jalapeño inference chip with Broadcom

OpenAI has unveiled Jalapeño, its first custom-built inference processor, designed and manufactured in partnership with Broadcom. The chip represents a strategic effort to reduce dependence on Nvidia's GPUs while optimizing costs for AI inference—the process of running trained models to generate real-time responses.

Design and Performance#

The chip was designed specifically for OpenAI's inference workloads, with the company's own AI models assisting in development. Early testing shows significantly better performance-per-watt compared to current state-of-the-art alternatives, according to OpenAI.

Strategic Context#

The official partnership was announced in October, though OpenAI's chip ambitions have long been speculated. The move mirrors similar efforts by Google and Amazon, which have built custom "AI accelerators" to optimize machine learning workloads.

OpenAI president Greg Brockman emphasized the company's philosophy: "We have a deep understanding of the workload. 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?"

Implications#

Jalapeño is purpose-built for inference—particularly cost-sensitive workloads like real-time code generation. More computationally intensive tasks like model pre-training will likely continue relying on Nvidia hardware. Even modest reductions in inference costs could meaningfully improve OpenAI's unit economics.

The move underscores a broader strategy: by controlling the entire stack—from chip architecture to deployment systems—OpenAI aims to optimize every layer around speed, reliability, and affordability.