Nvidia has entered into a significant agreement with the AI startup Poolside, committing $6 billion to develop a powerful model featuring open weights that can rival Chinese large language models such as DeepSeek and Kimi K3. This information was reported by the Wall Street Journal, citing its own sources.
On August 21, Bloomberg reported that the chip manufacturer plans to pay Poolside $6 billion for licenses to AI models, offer jobs to more than 100 of the startup's employees, and invest an additional $1 billion. According to the article, Nvidia has recently focused on supporting open-source models that developers can freely use and customize. The company aims to make the technology more accessible and avoid concentration among a few large players. Nvidia also links these efforts to bolstering future demand for its chips and AI systems.
The Poolside deal is structured similarly to Nvidia's recent acquisition of chip developer Groq, for which it reportedly paid $20 billion for licenses and hired most of the startup's employees and founders.
According to the Wall Street Journal, the agreement with Poolside will also pose a direct challenge to leading American AI firms, including OpenAI and Anthropic, as models with open weights are generally much cheaper to operate and allow for easier customization.
In July, discussions revived within the Trump administration regarding measures that could complicate the use of Chinese open-weight models. Twenty-five American companies, including Nvidia, Meta, and Microsoft, urged authorities against imposing restrictions. In an open letter, they warned that such bans would not strengthen the technological leadership of the United States but instead shift the market towards a few closed developers.
The intensification of discussions around potential restrictions was largely triggered by the release of Kimi K3, whose creators were accused by the White House of distilling Fable from Anthropic.
As a reminder, ForkLog examined the phenomenon of the Chinese model from Moonshot AI in a detailed article.