The chip, co-developed with semiconductor company Broadcom, is designed specifically for AI inference – the process of running an AI model to generate answers for users, as opposed to the lengthy and expensive process of training a model from scratch.
“While OpenAI is still measuring final performance, early testing shows that Jalapeno will deliver performance per watt substantially better than current state-of-the-art,” the company said in a blog post.
The company said its own AI models helped design the chip, tightening the timeline to develop high-performance semiconductors.
The chip is designed to work with a broad range of AI models, not just OpenAI’s own products, and will be deployed at data centres operated by Microsoft and other partners beginning in 2026.
By controlling its own chip design, OpenAI aims to reduce its dependence on outside suppliers such as Nvidia.
Broadcom chief executive Hock Tan called the chip “just the beginning,” with the two companies planning successive generations of products.
The companies last year announced their partnership to design and build their own specialised computer processors for artificial intelligence in an effort to break a dependency on Nvidia’s technology.
The chip was the latest in a series of announcements by OpenAI as it looked to strengthen its position as the preeminent company in generative AI against increasing competition from rivals Anthropic and Google.
Powerful chips are the engine of AI, and most companies building AI models depend on technology from Nvidia and a handful of other suppliers.
Tech giants including Google, Amazon and Microsoft have pursued similar custom chip strategies in recent years to cut costs and boost performance.
Nvidia’s chips – originally designed for video games – are uniquely suited to the heavy computing demands of training AI systems and have made it the world’s most valuable company.
