Broadcom unveils custom chip for OpenAI, challenging Nvidia’s dominance

OpenAI now has its own chip. The company revealed “Jalapeño,” its first custom AI accelerator, built in partnership with Broadcom and optimized specifically for inference workloads. The physical sample was delivered on June 24, with initial deployments targeted for late 2026.

From partnership to silicon

OpenAI and Broadcom have been co-developing chips for roughly 18 months, a collaboration that was formally announced back in October 2025. The scope of that original partnership was staggering: 10 gigawatts of custom AI accelerators, with OpenAI handling chip and system design while Broadcom manages development, deployment, and networking.

TSMC, the world’s dominant chip manufacturer, is handling the actual fabrication. That’s the same foundry that produces chips for Apple, AMD, and yes, Nvidia.

The initial rollout follows a phased approach. Late 2026 marks the first deployments, with scaling expected through 2027 and 2028. A full rollout is projected by late 2029.

Why this matters beyond the chip itself

Jalapeño is optimized for inference, not training. Training is the computationally brutal process of building an AI model from scratch. Inference is what happens after, when that trained model actually answers your questions, generates images, or writes code. As AI products scale to hundreds of millions of users, inference becomes the dominant cost center.

Custom application-specific integrated circuits, or ASICs, solve this by doing fewer things but doing them cheaper and more efficiently. Google has been running this playbook for years with its Tensor Processing Units. Meta has its own custom chip efforts. Amazon’s Trainium chips serve a similar purpose for AWS customers.

The Nvidia question

Broadcom’s stock apparently noticed the opportunity. Reports of an approximately $10 billion order from an unnamed customer contributed to a share price increase of roughly 9–11%.

Rising demand for AI computing power has historically correlated with tighter supply for mining-grade hardware, which can influence both equipment costs and hash rate dynamics. The AI hardware market is fragmenting, with Nvidia’s own customers now among its competitors. Whether Jalapeño lives up to its name won’t be clear until late 2026 at the earliest.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.