Nvidia is expected to announce a $13 billion purchase of Hugging Face later this week, a move that would bring the open-weight model platform under the chip giant’s ownership.
Hugging Face serves as the central hub for developers creating large language models outside the control of major labs. It functions similarly to GitHub for the current generation of artificial intelligence.
Speculation about this deal follows Nvidia’s recent agreement to acquire Poolside for $6 billion, which will relocate most of the open-weight model builder’s staff. Just two weeks prior, Stripe bought OpenRouter, the leading business provider of open-weight models, for over $7 billion.
These transactions represent a significant influx of capital into a sector built on distributing technology freely, mirroring current shifts in the industry.
Nvidia seeks to reduce its reliance on hyperscalers and frontier labs. This concern grows as major builders like OpenAI and Google develop their own inference chips, such as OpenAI’s Jalapeño, which was announced this week. If model creators manufacture their own hardware, Nvidia aims to secure a portion of that business.
The company already produces the Nemotron family of open-weight models, yet adoption remains limited. Controlling the largest US developer space for open models would give Nvidia access to a vast user base it could direct toward its chips and standards.
Questions regarding the cost of AI inference are also driving interest. Firms are looking at cheaper models from Chinese competitors like Moonshot, DeepSeek, and Alibaba. Currently, usage is modest but rising; Ramp data indicates 6% of companies use open-weight models, while Jellyfish found only 2% of software engineers surveyed utilise them.
Nik Albarran, the AI product lead at Jellyfish, told TechCrunch that open-weight models are mainly used by companies with products requiring repeated inference workloads, such as customer service chat platforms. These high-volume tasks allow an open model to be tuned to answer questions cheaply.
That is the angle Stripe uses for its OpenRouter acquisition. Patrick Collison, Stripe’s co-founder and CEO, stated: “Tokens are the central currency for companies building with AI, and it’s clear that the real-world economic potential will depend on making good use of scarce compute resources.”
For coding and agentic tasks, however, varying requests and complex reasoning often favour frontier models. This is partly because proprietary labs provide easier access, and sometimes offer token subsidies. Albarran notes that as companies refine their AI workflows, switching to open models will become simpler. Currently, the primary driver for choosing these models is control and configurability rather than cost.
“There are not many companies where that is the case yet … [but] if the prices continue to go up from the frontier labs, more and more companies will be forced to at least consider it,” Albarran told TechCrunch. “When your AI-driven workflows are much more mature, that’s when it makes sense to invest in self-hosting models.”
Lin Qiao, the CEO of Fireworks, leads a router and host for corporate users that often surfaces in acquisition discussions. Her company processes 40 trillion tokens daily, exceeding the volume of either Gemini’s or OpenAI’s APIs.
Fireworks relies on model diversity. As large language models proliferate and improve, companies will find it easier to train them specifically for their needs. “Every single app company should consider hiring an in-house researcher,” Qiao told TechCrunch last week. “They can use their product and product data to build their own model. The future is actually specialized intelligence. Literally, every single company should have their own model per use case, and that will happen automatically.”
It is easy to overlook how early AI is as a tool and a business. The dominance of OpenAI and Anthropic is not inevitable. As technology giants hedge their bets on the largest labs, the appeal of open technology proves difficult to ignore.
