Nvidia has reached a $6 billion technology licensing deal with the AI startup Poolside, according to sources familiar with the matter. Nvidia will leverage Poolside’s Model Factory technology and talent to accelerate the development of more advanced open-weighted AI models, surpassing its Chinese AI counterparts and ensuring America’s leading position globally.
On Saturday, August 22nd, as reported by The Wall Street Journal, Nvidia has agreed with Poolside to pay $6 billion for Poolside’s technology licensing (non-exclusive) and hire most of its engineers (over 100) to join Nvidia’s Nemotron team. This move aims to develop advanced open-weighted AI models surpassing China’s DeepSeek and Kimi K3.
The agreement between Nvidia and Poolside is not an acquisition deal. While many employees will be hired by Nvidia, Poolside’s founders Eiso Kant and Jason Warner, among other top executives, will continue to work at Poolside. Nvidia will also invest an additional $1 billion into Poolside.
In a letter to shareholders, Poolside stated that this agreement allows more people to participate in development, avoiding the scenario where Artificial General Intelligence (AGI) becomes a “closed technology controlled by a few.” AGI refers to AI that can perform tasks across various domains, like understanding, learning, reasoning, and problem-solving in multiple fields like humans.
Established in 2023, Poolside focuses on AI coding agent development and automated model development. Its core concept is to enable AI not only to write code but also engage in data generation, training, evaluation, and improving next-generation AI models. Poolside’s founder Warner previously served as the Chief Technology Officer at GitHub and contributed to incubating GitHub Copilot. By 2024, Poolside had trained its models on 130,000 real code repositories and millions of tasks.
Poolside’s technology is believed to significantly enhance AI model development efficiency, allowing for faster and lower-cost model iterations. Poolside’s automated model development, combined with Nvidia’s Graphics Processing Units (GPUs as the core hardware for AI training) and AI infrastructure, will assist Nvidia in hastening the creation of the next-generation Nemotron models.
In 2025, Poolside had plans to acquire over 40,000 Nvidia GB300 GPUs through a partnership with CoreWeave and build the 2GW AI computation center park Project Horizon in Texas. CoreWeave specializes in providing GPU cloud computing power and data centers for AI needs. On October 15, 2025, CoreWeave announced its collaboration with Poolside. However, Poolside’s financing and computing capacity expansion program did not progress as planned, leading them to strike a deal with Nvidia.
Last October, Nvidia CEO Jensen Huang stated at the GTC technology conference in Washington, D.C., that the U.S., while possessing strong proprietary models, also needs robust open models, in which Nvidia will invest. He said, “The U.S. must also maintain leadership in the open model field.”
Nvidia’s GTC technology conference was initially centered around GPU computations, but with Nvidia’s business expanding from GPU graphics processing to AI, the conference has now become a large-scale tech event revolving around AI, accelerated computing, robotics, AI infrastructure, models, and development tools.
Previously, Bloomberg reported that China is rapidly narrowing the gap with the U.S. in the field of AI. Despite the U.S. restricting the supply of advanced AI chips to China, Chinese AI models are beginning to match U.S. models in capabilities and are significantly lower in development costs.