OpenAI Unveils Its First AI Chip 'Jalapeño', Takes Direct Aim at Nvidia

OpenAI has officially entered the AI chip race. The ChatGPT maker has unveiled Jalapeño, its first in-house AI processor developed in partnership with Broadcom. The chip is designed specifically for AI inference, the process of generating responses to user queries, marking OpenAI's biggest step yet towards reducing its dependence on Nvidia's GPUs.

What Is Jalapeño?

Unlike Nvidia's general-purpose AI GPUs, Jalapeño is an application-specific integrated circuit (ASIC) built solely for inference workloads. That means the chip is designed to run AI models such as ChatGPT and Codex more efficiently after they have already been trained. Inference has become one of the fastest-growing segments of the AI industry as millions of users interact with chatbots and AI agents every day.

OpenAI says engineering samples are already running workloads in its own labs, including the GPT-5.3-Codex-Spark coding model, with broader deployment planned later this year.

Why OpenAI Is Making Its Own Chips

Running ChatGPT isn't cheap. OpenAI has relied heavily on Nvidia's GPUs to train and serve its AI models, but soaring demand has made those chips both expensive and difficult to secure. By designing its own processors, the company hopes to lower operating costs, improve efficiency, and gain greater control over its AI infrastructure.

The project was developed with Broadcom, which provided silicon implementation and networking expertise, while Canadian electronics manufacturer Celestica worked on the underlying hardware systems. Manufacturing will be handled by TSMC.

Can It Challenge Nvidia?

According to Broadcom CEO Hock Tan, Jalapeño delivers performance comparable to Nvidia's Blackwell chips and Google's Tensor Processing Units (TPUs). OpenAI also claims early testing shows the chip offers significantly better performance per watt than today's leading AI processors, although detailed benchmarks have not yet been released.

That does not mean Nvidia is suddenly in trouble. Nvidia continues to dominate AI training, where the largest and most powerful models are created. Jalapeño is currently focused only on inference, an increasingly important market but just one part of the AI computing ecosystem.

The Beginning of a Bigger Strategy

OpenAI describes Jalapeño as the first product in a multi-generation chip platform rather than a one-off project. The company says it is considering expanding future generations to support AI training as well, potentially reducing its reliance on external chip suppliers even further. The announcement follows last year's partnership between OpenAI and Broadcom to build large-scale AI infrastructure over the coming years.

Why It Matters

The AI race is no longer just about building better models. Increasingly, it is also about controlling the hardware that those models run on.

Google has TPUs. Amazon has Trainium and Inferentia. Microsoft has Maia. Meta has MTIA. OpenAI now has Jalapeño.

For years, Nvidia has been the undisputed supplier powering the AI boom. With OpenAI joining the growing list of companies designing custom chips, the industry is moving towards a future where the biggest AI players build more of their own hardware stack instead of relying entirely on one supplier.

Jalapeño will not dethrone Nvidia overnight. But it does signal that OpenAI wants to own more than the world's most popular AI chatbot. It also wants to own the silicon that powers it.