OpenAI has revealed its first AI chip, called Jalapeño, marking a major step toward hardware independence in the race to power advanced models. The chip was developed with Broadcom and is designed to run the company’s large language models while reducing reliance on Nvidia, which still dominates the global AI chip market.
The company said the project reflects a broader strategic shift: OpenAI wants to become more than a model developer and play a larger role in the hardware stack as well. A notable element of the effort is that OpenAI used its own AI models to help design and develop the chip, meaning the systems that will eventually run on Jalapeño were involved in planning it.
Jalapeño is built specifically for inference, the stage in which an AI system generates answers, images, or code for users. OpenAI said the chip combines high computing power with very low latency, and early tests showed a better performance-to-power ratio than competing solutions. Its architecture was designed to reduce internal data movement and use processing, memory, and communication resources more efficiently.
The move comes as Nvidia has become the near-exclusive supplier of advanced AI chips, driving shortages and extremely high costs across the industry. OpenAI, which operates massive server farms for millions of users, wants greater control over the technology behind its services. The development also brings the company closer to Google’s model, where software and custom hardware are developed together. OpenAI said the first Jalapeño deployments are expected to begin by the end of the year in Microsoft data centers and at other partner facilities.
