Nvidia is moving beyond selling the chips that power the AI boom and into building the models that run on them. The company has struck a roughly $7 billion deal with AI startup Poolside, licensing its technology and bringing on more than 100 of its engineers to build one of the most powerful open-weight AI models yet, a direct challenge to Chinese open-source leaders DeepSeek and Kimi K3.
Coin Bureau reported that the deal is meant to help Nvidia challenge DeepSeek, Kimi K3, OpenAI and Anthropic simultaneously, while Bull Theory added that Nvidia is licensing technology and acquiring key engineering talent from Poolside specifically to develop the open-weight model. Bloomberg, which broke the story, reported the arrangement includes a roughly $1 billion investment in Poolside at a $12 billion pre-money valuation on top of the $6 billion technology license, with the combined engineering team folding into Nvidia's existing Nemotron open-weight model program.
Why Nvidia Wants to Own a Model, Not Just Sell Chips
The logic is defensive as much as offensive. Chinese labs have spent the past year releasing increasingly capable open-weight models that developers can run without paying for API access to a closed model, and that dynamic threatens the broader ecosystem Nvidia's hardware sales depend on: if the best free models come from labs that favor domestic Chinese chips or optimize for non-Nvidia hardware, it chips away at Nvidia's own moat over time. Building a leading, Nvidia-optimized open-weight model gives the company a way to keep developers building on its stack even when they choose an open, self-hosted model over a paid one from OpenAI or Anthropic.
An Unusual Structure
Whale Insider also flagged the deal, framing it as Nvidia's answer to Chinese open-weight rivals per the Wall Street Journal's own reporting. What stands out about the structure is that it's part investment, part acqui-hire, and part technology license rather than a straightforward acquisition, letting Nvidia absorb Poolside's team and code without formally buying the company. More than 100 Poolside engineers are expected to move to Nvidia's Nemotron effort, effectively giving the chipmaker an in-house foundation-model team overnight.
Related: 35% of Post-ChatGPT Web Pages Are AI-Written, Pew Finds
The Earnings Timing Isn't a Coincidence
The deal surfaced just days ahead of Nvidia's August 26 earnings report, a print investors are already treating as a pivotal moment given the stock's compressed valuation multiple heading into it. A strategic move into open-weight models gives Nvidia a growth narrative beyond raw chip sales to point to on the call, at a time when questions about AI capital-expenditure sustainability and Chinese competition have both been weighing on sentiment toward the stock and, by extension, the broader risk-asset complex that includes crypto.
