OpenAI has unveiled its first custom AI chip, Jalapeño, developed with Broadcom to power inference workloads, reduce hardware dependency, and strengthen its AI infrastructure strategy.

OpenAI is taking a major step beyond software and into AI hardware.

In partnership with Broadcom, the company has introduced its first custom artificial intelligence processor, called Jalapeño, marking its official entry into the fast-growing market for specialized AI chips.

According to the announcement, Broadcom will manufacture the new chip, while OpenAI plans to use it primarily for inference workloads — the stage where trained AI models generate responses for users across ChatGPT and other AI-powered services.

This move signals a broader strategy from OpenAI: building a more integrated technology stack that supports its expanding AI ecosystem while lowering dependence on third-party hardware suppliers. As demand for generative AI continues to surge worldwide, controlling more of the underlying infrastructure is becoming a strategic advantage.

Greg Brockman, President of OpenAI, said designing more internal infrastructure components could help the company deliver stronger AI capabilities with better efficiency, reliability, and wider access to advanced tools.

Why OpenAI Is Reducing Dependence on Nvidia

Since the generative AI boom accelerated in 2022, OpenAI has been one of Nvidia’s major customers, relying heavily on its chips for training and running AI models. But rising usage of OpenAI’s products has pushed the company to diversify its computing resources and explore alternative hardware options.

Earlier this year, OpenAI signed an agreement with Amazon Web Services to use Trainium chips for AI workloads. It has also expanded partnerships with AMD and Cerebras, both of which are active in the development of high-performance AI processors.

Jalapeño: A Custom AI Chip for Inference

The new Jalapeño processor belongs to the ASIC category — application-specific integrated circuits. Unlike traditional GPUs, ASIC chips are less flexible, but they are generally more cost-efficient and can be optimized for very specific workloads.

OpenAI says the chip was designed in just nine months, and that the company also contributed to key parts of the computing system that will run alongside it.

OpenAI and Broadcom describe Jalapeño as an “intelligence processor” and the first AI accelerator in a new technology platform designed to make advanced AI systems faster, more dependable, and more scalable.

A Long-Term Infrastructure Play

The Jalapeño project is part of a much larger infrastructure roadmap. Back in October, OpenAI and Broadcom revealed plans to develop servers powered by OpenAI-designed chips starting in late 2026. Their long-term vision involves building a massive AI infrastructure platform that could require nearly 10 gigawatts of electricity.

Commercial production of Jalapeño is expected to begin by the end of 2026, with deployment expanding in the years that follow.

For Broadcom, the partnership further strengthens its position as one of the major beneficiaries of the AI boom, as it continues helping cloud providers and advanced AI labs build custom silicon for next-generation computing.