Nvidia is close to acquiring AI model hosting platform Hugging Face for $12.9 billion — nearly 3× its $4.5 billion valuation in 2023 — with an announcement expected as early as this week. What Nvidia is buying is the front door that every AI developer opens each day — and it's about to bear Nvidia's name.
An AI toolshop about to be renamed Nvidia — first, it must hold onto its people
Nvidia and Hugging Face are nearing an all-cash acquisition deal worth $12.9 billion, with an announcement possible as early as this week. Bundled into the deal is a retention package covering key engineers and researchers — tying the core team to the company is written into the terms from the start.
According to Bloomberg, the two sides are in the final stages of negotiations. The Information adds that beyond the deal itself, Nvidia is also negotiating a roughly $1 billion employee retention package. Keeping the talent is one of the deal's core terms.
The 86× P/S ratio (price-to-sales ratio — a company's market cap divided by annual revenue; a higher multiple signals greater market confidence in future growth) is based on the roughly $150 million annualized revenue figure cited by The Information. No independent third-party verification of that revenue number has been seen, so the multiple is for reference only and shouldn't be taken as a direct pricing benchmark.
One front door, locking down three chains: models, frameworks, and deployment
2 million+ models, 500,000+ datasets, 1 million+ apps — these numbers show Hugging Face is the unavoidable first stop for AI developers. What Nvidia is buying is a gateway.
The deal numbers are in the data cards: the premium and P/S ratio far exceed those of a typical SaaS company. But Nvidia isn't buying revenue — it's buying position.
How critical is that position? Hugging Face is essentially today's AI world's "GitHub + App Store" hybrid. Developers come here to find models, researchers come here to publish datasets, and enterprises connect here to run inference (putting trained models to work answering questions or generating content). It doesn't dominate any single link in the chain — it sits at the intersection of all of them. Whoever owns that intersection simultaneously controls the model, framework, and deployment chains.
For Nvidia, the math works. Its core business is selling GPUs — the graphics/compute chips that AI model training and inference depend on most. How many GPUs it sells depends on how many people downstream are running models on its hardware. Hugging Face is precisely the first stop where downstream developers "pick a model, test a model, deploy a model."
Once that gateway carries Nvidia's name, competitors will find it hard to route around it. Convincing developers to switch to AMD GPUs, for example, or to adopt a toolchain that doesn't rely on CUDA (CUDA — Nvidia's proprietary compute framework for its GPUs, essentially the "operating system" for AI training), would all come with far higher switching costs.
What Nvidia is buying is a traffic gateway — it doesn't compete directly with OpenAI or Anthropic in the model business, but it ensures every model business passes through its yard. Models are the travelers; the platform is the gate.
The price of that gateway also includes a retention plan (see the data card). For a company built on community and developer culture, talent loss can hollow out a platform fast. That money shows Nvidia understands: buying the shell is easy; buying the soul is hard.