Nvidia is planning to use a $6 billion deal it struck this week to build one of the world's most powerful open-weight AI models, one that would compete with Chinese heavyweights like DeepSeek and Kimi K3, according to people familiar with the matter.

The chip giant's licensing deal with the AI startup Poolside is also set to present a direct challenge to frontier U.S. AI companies including OpenAI and Anthropic, since open-weight models are generally far cheaper to operate and allow easy customization.

Although Nvidia counts big labs like OpenAI and Anthropic as some of its closest partners, the move reflects Chief Executive Jensen Huang's efforts to position Nvidia for success in multiple AI battlegrounds simultaneously, ensuring the company is able to hold on to its dominant market position.

Poolside, which was founded in 2023 by Eiso Kant, a software developer, and Jason Warner, the former chief technology officer of GitHub, sent a letter to shareholders Thursday announcing a dramatic restructuring of its business.

Nvidia will invest $1 billion in the company at a pre-money valuation of $12 billion, and more significantly, will pay $6 billion to license Poolside's technology and hire the bulk of its engineers, according to a copy of the letter reviewed by The Wall Street Journal.

Poolside's founders wrote that the Nvidia deal was meant to secure a future where artificial general intelligence, or humanlike computer reasoning, "would not be a closed technology controlled by few but one built by many out in the open."

The startup's leadership-including Kant, Warner and operations executive Margarida Garcia, won't join Nvidia-and will continue working on unspecified research projects. The deal was first reported by Newcomer, an independent tech newsletter.

More than 100 Poolside employees, including engineers, will join Nvidia and will work on the chip-designer's Nemotron project-an effort to develop open-weight models that Nvidia debuted in 2023-according to people familiar with the matter. The Poolside engineers will help improve Nemotron's largest and most sophisticated models under development, these people said.

Although Nvidia's Huang has long endorsed open-weight models, which can be free to download and allow users to alter them for specific needs, the American AI industry has poured far more resources into proprietary, or "closed" AI models, which don't share their source code or numerical weights with developers. These include startups that have become household names, like Anthropic and OpenAI, as well as the research arms of established tech firms like Alphabet's Google.

The failure of U.S. AI labs to give priority to open-source models has created concerns that businesses and countries around the world will turn to Chinese open-weight models instead.

Over the last six months, however, Nvidia-known best for designing GPUs, the advanced computer processors that have powered the AI boom-has stepped up efforts to develop open-weight models, partly to counter the rapid rise of open models developed in China.

In March, the chip giant launched the Nemotron Coalition, an association of leading open-model developers including Mistral, Thinking Machines Lab and Perplexity, that it said will share data, expertise, and computing resources.

The same month, the Journal reported that Reflection AI, another Nvidia-backed startup involved with the Nvidia coalition that investors have described as "DeepSeek of the West" for its sophisticated open-weight models, was in advanced talks to raise $2.5 billion at a valuation of $25 billion.

Last month, in his first-ever post on the social-media network X, Huang published a letter titled "Open Weights and American AI Leadership," in which he argued that whether or not the U.S. wins the AI race "will be judged not by one frontier AI model, but by whether the United States builds a strong, open ecosystem that diffuses into every sector."

When DeepSeek released its first low-cost, open-weight model in January 2025, it caused carnage in U.S. tech stocks because investors interpreted it as proof that Chinese models were on track to match or surpass American AI technology. Venture capitalist Marc Andreessen called it a "Sputnik moment" for AI.

The U.S.-China AI war has continued since then, as new players in China have emerged and built open models that trail frontier U.S. AI capabilities by only a few months. They include Moonshot's Kimi and Z.AI's GLM family of models, among others. In licensing Poolside's technology and scooping up the bulk of its engineering talent, Nvidia is betting that it can produce a version of Nemotron that rivals the best frontier models in the next year.

Earlier this month, Nvidia released Nemotron 3.5 Lightning, a "lightweight" version of its open weight model, and is developing its largest version of the product yet. It is rumored to have more than a trillion training parameters, putting it in the same league as some of the largest AI models ever designed. Still, some Chinese open-weight models are much larger.

The Poolside deal also puts Nvidia in the delicate position of competing with some of the biggest customers for its chips-the large frontier labs that make closed models, like OpenAI and Anthropic-as well as with other open-model designers that Nvidia is backing. Some experts have pointed out that most of the large AI labs are also designing custom silicon processors to power their models, and are therefore already in direct competition with Nvidia.

Although Poolside and Nvidia have a longstanding relationship, the deal came together quickly, only in the past few weeks. Nvidia made its offer for the investment and licensing deal to Poolside around the same time that the startup released its latest model, known as Laguna S, which has emerged as one of the most popular open-weight models in the West.

Poolside's Thursday letter referenced a failed fundraising effort in 2025 that caused a reckoning for the startup and indirectly led to the deal with Nvidia.

"At the end of last year, we had a 6 week window in which to raise $2 billion dollars to pay for a 40,000 GB300 cluster coming online in January," the letter read, referring to a common type of Nvidia server used in model training. "We didn't close it in time, and we lost the cluster."

Poolside leaders said "we dusted ourselves off, and went back to work," the letter went on, but soon realized that the company would run out of computing power as soon as next year, without access to the capital and data centers it needed. Nvidia was the perfect partner, Poolside wrote in the letter, in part because they have fewer computing infrastructure or cash limitations.