A three-person web agency in Ghent runs its client chatbots on an open AI model it downloaded, free of charge, from a site called Hugging Face. So does a legal-tech startup in Lyon, a logistics planner in Rotterdam and — according to Nvidia's own announcement — more than 200,000 companies worldwide. On 3 September, Nvidia agreed to buy that site for $12.93 billion, roughly €11.2 billion. If any part of your business touches an open AI model, this deal is about you.
The deal, in plain terms
Hugging Face is the shelf where open AI lives. Founded in 2016 by three French entrepreneurs, it began life as a chatbot app for teenagers and grew into the place where AI models are shared the way code is shared on GitHub. Nvidia's announcement puts the numbers at more than 18 million developers, around 3 million models, 500,000 datasets and over a million applications hosted on the platform. When a European team says "we run an open model", the odds are high it came from here.
The price tells you what open AI is now worth. According to TechCrunch, Hugging Face was valued at $4.5 billion in 2023 and brings in roughly $150 million a year in revenue — and it turned down a $500 million investment from Nvidia only last year. Nvidia is now paying almost three times that 2023 valuation, a figure Bloomberg and CNBC both confirmed on 3 September, for a company that was only just approaching profitability.
Why a chipmaker pays €11 billion for a model library
Nvidia sells shovels, and open models dig. TechCrunch's read on the deal is blunt: the big AI labs — OpenAI, Google, Amazon, Anthropic — are all designing their own chips, which threatens Nvidia's core business. A thriving open-model ecosystem, by contrast, runs overwhelmingly on hardware that people buy from Nvidia. Owning the hub where those models are found, downloaded and deployed keeps that flywheel spinning, whatever the big labs do.
Nvidia was already the hub's biggest contributor. The company says it has released more than 500 open models and 250 open datasets on Hugging Face — more than anyone else. Buying the platform is less a change of direction than a doubling-down: Nvidia's interests and open AI's interests are, for now, pointed the same way.
Why this matters more in Europe than anywhere else
Open models are Europe's practical answer to GDPR. For a small firm in Antwerp or Munich, the quiet appeal of an open model was never ideology. It was control: download the weights, run them on a European server or on your own hardware, and customer data never leaves your infrastructure. That posture makes GDPR conversations with your accountant and your clients dramatically simpler, and it makes the transparency expectations of the EU AI Act easier to meet, because you can actually inspect what you deploy.
Hugging Face is also a European story. As Euronews reports, the founders are French and roughly half of the company's workforce is based in Paris. One of the few European-founded pillars of the AI world is now set to be fully owned by an American chipmaker — and Euronews expects the deal to raise eyebrows among competition regulators in both Paris and Brussels.
The sovereignty debate cuts both ways. The case for optimism, laid out in the Euronews analysis, is that open models are precisely what lets smaller European players compete without OpenAI-sized budgets. Hugging Face co-founder Clément Delangue told US lawmakers back in 2023 that open source AI "distributes economic gains by enabling hundreds of thousands of small companies". The case for concern is concentration: the world's most valuable chipmaker now owns the front door to open AI, and front doors can be narrowed.
What Nvidia promised — and how to read it
The commitments are on the record. In its announcement, Nvidia pledges that Hugging Face "will remain an open platform for the entire AI ecosystem", that developers keep free choice of models, frameworks, clouds and computing platforms, and that Nvidia hardware will not be required to build or deploy through the hub.
Promises age with ownership. The precedent worth remembering is Microsoft's purchase of GitHub in 2018, which was greeted with similar anxiety. GitHub stayed open — and it also became the launchpad for Microsoft's paid Copilot products. The realistic expectation is not a locked gate but a free hub with ever more paid rails built around it: hosted inference, enterprise tiers, hardware bundles. That is not sinister. It is how a $12.93 billion purchase pays for itself, and it is worth pricing into any long-term plans.
When this isn't the right moment to change anything
If you use AI through subscriptions, carry on. A bakery in Utrecht paying for ChatGPT, a consultancy in Lille on Microsoft Copilot — nothing in this deal changes your tools, your prices or your contracts. This is infrastructure news, not product news.
If you self-host open models, don't panic-migrate. The weights you have already downloaded are yours, under the licence they shipped with, and the hub remains free to use. The sensible response is housekeeping, not flight: write down which models your business actually depends on, keep local copies of the weights and their licence files, and glance at the pricing of any hosted endpoints you rely on over the coming year. Move only if the terms actually move.
If you were about to pick a platform, add one question. Teams choosing between a closed API and a self-hosted open model this autumn should still decide on cost, quality and data control — but it is now fair to ask any provider, Hugging Face included, what happens to your setup if their ownership or pricing changes. A good answer is portable weights and standard formats. That answer still exists here.
The deal will take months to close and years to show its real effects, and most of the coverage will stay at the level of geopolitics. For a small business, the useful question is more modest: which AI tools does your work actually depend on, where do they run, and what would you do if one of them changed hands or changed price? If you want a structured way to answer that — before the headlines answer it for you — Cresly's AI Readiness Scan maps where AI genuinely fits in your business, open models included, and where it doesn't yet.