Nvidia and rivals race to snap up open-weight AI companies

Nvidia is reportedly negotiating a roughly $13 billion acquisition of Hugging Face, the platform where developers share open-weight AI models and benchmarks, often described as a GitHub equivalent for the AI industry. This would mark the third major deal in the space recently: Nvidia already agreed to a $6 billion deal with open-weight model builder Poolside, under which most of its staff will move to the chipmaker, while payments company Stripe acquired OpenRouter, the leading distributor of open-weight models to businesses, for more than $7 billion two weeks earlier.

Why Nvidia wants open models

Nvidia is seeking to reduce its dependence on deals with major hyperscalers and frontier AI labs, especially as companies like OpenAI and Google develop their own inference chips — OpenAI unveiled its Jalapeño chip's capabilities this week. Nvidia already produces its own open-weight Nemotron model family, though adoption has been modest. Gaining control of the largest U.S. open-model developer community would give Nvidia access to a large user base it could steer toward its chips and standards.

Adoption still small but growing

According to spending-data surveys, only 6% of companies use open-weight models (per Ramp) and just 2% of software engineers do (per Jellyfish). Jellyfish AI product lead Nik Albarran said these models are mainly used by companies with high-volume, repetitive workloads, such as customer service chat, where a tuned open model can answer cheaply. For coding and more complex, reasoning-heavy tasks, proprietary frontier models still tend to win, partly due to easier access and, in some cases, subsidized pricing.

Fireworks CEO Lin Qiao, whose company processes 40 trillion tokens daily, said the future lies in specialized models, predicting more companies will eventually build models tailored to their own products and data.