Analysis
Pulse reported Nvidia's roughly $13 billion agreement for Hugging Face earlier this week. What has changed since is the shape of the pattern around it. TechCrunch's Tim Fernholz reports that the Hugging Face deal is one of three -- Nvidia is also acquiring Poolside, with most employees moving over, and Stripe announced a purchase of OpenRouter two weeks ago. Combined, the three deals commit more than $26 billion to companies whose products are largely built on giving models away.
Here is the enumeration of the consolidation, with what each company actually does:
- Hugging Face -- reported ~$13B, acquiring: Nvidia: the default repository and hub for open model weights and datasets, founded 2016 by Clement Delangue, Julien Chaumond and Thomas Wolf; last privately valued at $4.5 billion in 2023. It is the distribution front door for open models.
- Poolside -- $6B, acquiring: Nvidia: code-generation foundation models for enterprise software teams, founded 2023 by Jason Warner (former GitHub CTO) and Eiso Kant. Most employees are moving to Nvidia. Competitors: Cursor, Cognition, Magic.
- OpenRouter -- $7B+, acquiring: Stripe: a routing layer that lets developers call hundreds of models through one API and one bill, founded 2023 by Alex Atallah, co-founder of OpenSea. Competitors: Together AI, Fireworks, Vercel's AI Gateway. "Tokens are the central currency for companies building with AI, and clear economic potential depends on making good use of scarce compute resources," Stripe CEO Patrick Collison said.
- Fireworks AI -- not acquired, the volume benchmark: processes 40 trillion tokens daily, exceeding both the Gemini and OpenAI APIs, per TechCrunch. Founded 2022 by former Meta PyTorch engineers led by Lin Qiao.
“- Poolside -- $6B, acquiring: Nvidia: code-generation foundation models for enterprise software teams, founded 2023 by Jason Warner (former GitHub CTO) and Eiso Kant.”
The adoption gap
The striking thing is how little of this rests on measured usage. A Ramp survey puts open-weight model adoption at 6% of companies; Jellyfish data shows 2% of software engineers use them. Buyers are not paying for today's revenue -- they are paying for position in the layer where developers choose, route and pay for models, on the theory that the choice layer is more durable than any single model.
What the acquirers get
What none of these deals settle is whether open weights become the default. Ars Technica's report on the Hugging Face agreement framed it as Nvidia buying the community layer of the ecosystem; the 6% enterprise adoption figure says that community is still mostly researchers and hobbyists rather than production buyers. Acquirers are paying today's prices for a scenario in which open models take a meaningful share of inference over the next three years.
For Nvidia, owning the hub and the code models is a defense of CUDA's developer moat as inference workloads migrate to alternative silicon. For Stripe, OpenRouter turns token routing into a payments and metering problem, which is Stripe's actual business. The risk in both cases is the same: neutral infrastructure loses its neutrality the moment a strategic owner buys it, and developers route around vendors they no longer trust. Hugging Face's value was that it belonged to everyone.
