Nvidia is reportedly in talks to acquire Hugging Face, the platform for sharing open-weight AI models and benchmarks, in a deal valued at $13 billion. If confirmed, it would be the latest in a wave of major acquisitions targeting companies that build or distribute open-weight AI technology. The deal follows Nvidia鈥檚 $6 billion agreement with Poolside, an open-weight model builder whose employees would mostly move to the chipmaker, and Stripe鈥檚 purchase of OpenRouter, a provider of open-weight models to businesses, for over $7 billion.
The reported Hugging Face negotiations come as Nvidia seeks to reduce its reliance on deals with major hyperscalers and frontier labs. That push is driven partly by OpenAI and Google developing their own inference chips, including OpenAI鈥檚 chip announced this week, which could make those labs less dependent on Nvidia hardware. Nvidia already offers its own Nemotron family of open-weight models, but adoption has been limited, and acquiring Hugging Face would give the company direct access to a large developer community it could steer toward its chips and standards.
Open-weight models remain a small slice of the AI market. A survey of spending data by Ramp found that only 6 percent of companies use them, while a separate survey by Jellyfish, a developer tools maker, put the figure at 2 percent of software engineers. Nik Albarran, AI product lead at Jellyfish, said open-weight models are mainly used by companies with repeated, high-volume inference workloads, such as customer service chats. Those tasks allow a tuned open-weight model to cut costs, he said.
Stripe framed its OpenRouter deal in similar terms. Patrick Collison, Stripe鈥檚 cofounder and CEO, said in a statement that tokens are the central currency for AI companies and that real economic potential will depend on using scarce compute resources well. For coding and agentic tasks, however, Albarran said frontier models often win out because proprietary labs offer easier access and sometimes a token subsidy. He added that companies currently turn to open models mainly for control and configurability, not for cost reasons, but that if frontier lab prices keep rising, more firms may reconsider.
Lin Qiao, CEO of Fireworks, a leading router and host for open-weight models often mentioned as a potential acquisition target, said her company processes 40 trillion tokens a day, more than either Gemini or OpenAI鈥檚 APIs. She said Fireworks is betting on model diversity, with more companies training models for their specific needs. She said every app company should consider hiring an in-house researcher and using product data to build custom models, describing the future as specialized intelligence where each company has its own model per use case.
The flurry of deals highlights how early the AI market still is, and that the current dominance of OpenAI and Anthropic is not guaranteed. As tech giants hedge their bets, the appeal of open technology is proving strong.
More AI news from TechManNews.
