OpenAI unveils ‘Jalapeño’ chip to reduce dependence on Nvidia

OpenAI has taken a major step toward reducing its dependence on third-party chipmakers by unveiling **Jalapeño**, the company's first custom-designed artificial intelligence processor. Developed in partnership with semiconductor leader **Broadcom**, the new chip is purpose-built to power AI inference—the stage where models generate responses to user prompts—across ChatGPT, Codex, API services, and OpenAI’s next generation of autonomous AI agents. The announcement marks one of the most significant infrastructure developments in OpenAI’s history as the race to build faster, more efficient AI systems accelerates worldwide.

For years, OpenAI has relied heavily on **Nvidia’s GPUs** to train and deploy its increasingly sophisticated language models. However, explosive global demand for AI computing has made advanced processors scarce and expensive, pushing leading technology companies to design their own specialized silicon. With Jalapeño, OpenAI joins companies such as Google, Amazon, Meta, and Microsoft in developing proprietary AI chips to gain greater control over performance, costs, and long-term scalability.

According to OpenAI, Jalapeño was engineered specifically around the computational needs of modern large language models rather than adapting existing chip architectures. Working closely with Broadcom, the company completed the processor’s design-to-production cycle in just **nine months**, an unusually rapid timeline for semiconductor development. OpenAI also revealed that its own AI models assisted engineers during portions of the design process, demonstrating how artificial intelligence is beginning to accelerate hardware innovation itself.

The processor will initially focus on **AI inference**, which has become one of the fastest-growing workloads in computing as hundreds of millions of users interact with conversational AI every day. Unlike model training, which requires enormous computational resources over extended periods, inference demands extremely fast, energy-efficient processing capable of serving millions of requests simultaneously. OpenAI believes Jalapeño can significantly improve performance while lowering operating costs across its expanding ecosystem of AI products.

Manufacturing of the new chip is being handled by **Taiwan Semiconductor Manufacturing Company (TSMC)**, the world's largest contract chipmaker, while Canadian technology company **Celestica** is assisting with hardware integration and deployment. Broadcom CEO **Hock Tan** described Jalapeño as the foundation of a broader multi-generation infrastructure roadmap designed to support increasingly capable AI models over the coming years. OpenAI plans to begin deploying the processor throughout its production infrastructure later this year.

Although OpenAI has not yet released detailed benchmark figures, the company says early internal testing indicates Jalapeño delivers substantially better **performance per watt** than many existing inference solutions. Improving energy efficiency has become one of the industry's biggest priorities as AI data centers consume rapidly increasing amounts of electricity. Lower power consumption could reduce operational expenses while enabling OpenAI to expand ChatGPT and future AI services more economically.

Industry analysts view the announcement as another milestone in the growing competition between AI software companies and traditional semiconductor manufacturers. While Nvidia continues to dominate AI training hardware, companies increasingly want customized processors optimized for their own workloads rather than relying entirely on general-purpose GPUs. Proprietary chips also provide greater supply-chain stability at a time when global demand for AI hardware continues to outpace manufacturing capacity.

Despite the launch, OpenAI emphasized that Jalapeño is **not intended to replace Nvidia** for every workload. The company expects to continue using Nvidia's advanced GPUs for training its largest frontier models, which remain among the most computationally demanding tasks in technology. Instead, Jalapeño is designed to complement existing infrastructure by making everyday AI interactions faster, more reliable, and more cost-effective.

The unveiling also reflects OpenAI’s broader ambition to build a complete AI infrastructure ecosystem spanning software, hardware, cloud services, and specialized computing platforms. As AI models become increasingly capable and user adoption continues to grow worldwide, controlling more of the underlying technology stack could provide OpenAI with strategic advantages in both performance and economics.

OpenAI described Jalapeño as only the **first member of a larger family of custom AI processors** currently under development. Future generations are expected to support increasingly powerful AI systems, autonomous software agents, and next-generation reasoning models. With demand for artificial intelligence continuing to surge across industries, OpenAI’s entry into custom silicon signals that the future of AI competition will be shaped not only by smarter models, but also by the specialized hardware that powers them.