Nvidia targets Hugging Face in $13bn deal as open-weight AI draws major capital
A wave of multibillion-dollar acquisitions in the open-weight AI sector signals a strategic shift by tech giants to secure model-building capabilities and reduce reliance on frontier laboratories.
Nvidia is reportedly negotiating a $13 billion acquisition of Hugging Face, the leading platform for sharing open-weight artificial intelligence models. The deal would mark the latest in a rapid consolidation of open-source AI infrastructure by major technology corporations.
This follows a $6 billion agreement between Nvidia and open-weight model builder Poolside, which will see most of its staff join the chipmaker. Two weeks prior, payments giant Stripe acquired OpenRouter, a top provider of open-weight models to businesses, for more than $7 billion.
For Nvidia, these moves represent a strategic hedge against growing dependence on hyperscalers and frontier laboratories. As companies like OpenAI and Google develop their own inference chips, such as OpenAI’s recently announced Jalapeño, the chipmaker is seeking a direct foothold in the model-building ecosystem.
Although Nvidia already produces its Nemotron family of open-weight models, market uptake has remained limited. Gaining control of the largest United States developer space for open models would provide Nvidia with a massive user base to steer toward its proprietary chips and technical standards.
The surge in capital reflects mounting industry anxiety over the escalating costs of AI inference. Businesses are increasingly exploring cheaper alternatives, including models developed by Chinese firms such as Moonshot, DeepSeek, and Alibaba, though current adoption remains niche.
According to spending data from Ramp, only 6 per cent of companies currently use open-weight models. Similarly, developer tools provider Jellyfish measures adoption among just 2 per cent of software engineers.
Nik Albarran, AI product lead at Jellyfish, notes that open-weight models are primarily adopted by companies managing repeated inference workloads, such as customer service chats. These high-volume, repetitive tasks allow businesses to tune open models for cheaper, more efficient responses.
Stripe has explicitly framed its OpenRouter acquisition around this economic reality. Patrick Collison, Stripe’s co-founder and chief executive, stated that tokens are the central currency for companies building with AI, and real-world economic potential depends on making good use of scarce compute resources.
For complex coding and agentic tasks, proprietary frontier models still dominate due to easier access and occasional token subsidies. Albarran observes that the primary current driver for open models is control and configurability, rather than immediate cost savings.
However, this dynamic is expected to shift as frontier laboratory prices rise. Albarran notes that more companies will be forced to consider open alternatives, adding that self-hosting models makes the most sense once AI-driven workflows mature.
The scale of this emerging market is already substantial. Lin Qiao, chief executive of open-weight model host Fireworks, says her company processes 40 trillion tokens daily, surpassing the API volume of both Google’s Gemini and OpenAI.
Qiao argues that the future of artificial intelligence lies in specialised, in-house models tailored to specific corporate data and use cases. This growing appetite for model diversity suggests that the current dominance of a few frontier laboratories is far from inevitable.
