Nvidia’s French Stakes Bind Europe’s AI to Its Chips

Nvidia announced planned minority stakes in four French companies on October 13 and a $250 million UK compute gift, putting Europe’s new AI work on its own chips. Jensen Huang, the company’s chief executive, named the French targets at a Place Vendôme reception with Economy Minister Roland Lescure, then saw President Emmanuel Macron at the Élysée.

Huang said the stakes would “launch the artificial intelligence revolution in Europe.” What Europe actually receives is more GPUs, more Nvidia cloud partners, and a still-open bid for Hugging Face.

Four French Companies, One Chip Supplier

Nvidia’s French blog listed planned investments in four French startups: Doctolib, Entalpic, Genesis AI and Pigment. Amounts were not disclosed. Prior French holdings already include Alice&Bob, Gradium and Scintil Photonics.

The new names are not model labs. They are a clinic network, a materials shop, a robotics builder and a planning suite, the kind of buyers that keep GPUs busy after the training run ends. Genesis AI called Nvidia a strategic investor. Doctolib’s chief executive said the chipmaker will be a minority holder with no board seat.

THE FOUR FRENCH STAKES

A chipmaker that owns a slice of its customer does not have to win the next hardware bake-off in an open fight. The cheque is small next to a data-centre order, and that is the point: the stake is a door into the workload, not a trophy valuation.

Huang’s $10 Billion Europe Tally

Paris was the first stop on a run that also took in London, with Berlin still ahead. Huang told the room Nvidia had put at least $8 billion into Europe over the past year and would announce about $2 billion more within a week, a sum the company itself rounds to nearly $10 billion of commitments.

HUANG’S EUROPE TALLY

• $8 billion: What Huang said Nvidia deployed in Europe over the past year.
• $2 billion: Further European deals he said would be named within a week of the Paris stop.
• $250 million: Compute access for UK agent-safety research, spread over five years.
• $12.93 billion: The Hugging Face purchase agreed on September 3, still awaiting close.

He also used the Paris stop to argue that open-weight models are how Europe closes the gap with the United States, cheaper to run and easier to bend to local languages. Mistral is the exhibit he keeps pointing at. The French lab has Large 4 in private preview, trained on 4,000 Grace Blackwell GPUs inside European data centres, with weights due on October 27.

The same blog that named the four stakes also described how far the stack already runs in France. LINAGORA and Docaposte won the CANUT framework, making Twake.ai, built on Nvidia Nemotron, the sole referenced digital workplace for more than 2,500 French public bodies. Vera Rubin Ultra racks built in Angers are planned for 2028, with trays produced in Europe after a factory opening on October 2. The sovereign label is local assembly. The silicon is still Nvidia’s.

Doctolib Will Train In-House Models on Nvidia Iron

Doctolib is the name everyone in the room already knew. The company books appointments and stores records, and in August it opened a research programme on data from its 50 million users. A secondary sale in April valued it at €3.6 billion ($4.0 billion), 38 percent below the €5.8 billion it carried after its 2022 round.

Stanislas Niox-Château, Doctolib’s chief executive, said Nvidia’s money would speed hiring and the build-out of in-house models for patient and clinician assistants, plus the paperwork that eats clinic time. The patient assistant is already in the hands of 400,000 people in France, with a full rollout promised in 2027. Doctolib has been running on Nvidia cards for a year.

The funds will help it speed up hiring and development of its in-house AI models that will power personal assistants for both patients and doctors, as well as help with administrative tasks.

Stanislas Niox-Château, chief executive, Doctolib

He said those models, and the patient data, would stay under Doctolib’s control rather than sit inside OpenAI or Anthropic products. That is a real break from rented chatbots. It is not a break from the GPU vendor. The models Doctolib wants to own will be trained on the same chips Nvidia is now buying a slice of the company to supply.

A $250 Million Gift That Runs on One Stack

In London the next day, Nvidia published a compute donation valued at $250 million over five years for UK work on agent safety and security. That is $50 million a year of access, not cash. Most of it is meant for the wider research community. A share is reserved for three named labs. Provider choice, architecture and commercial terms are still to be agreed. The machines will sit in the UK through Nvidia Cloud Partners.

THE NAMED UK LABS

• BOLD, Oxford: Jakob Foerster’s lab on systems that learn, adapt and work alongside people.
• SOFAIR, UCL: Open-source architectures and cheaper ways to build and run models.
• Imperial College London: Safety engineering and robotics, including physical agents that act in the world.

UK AI Minister Kanishka Narayan called the gift a vote of confidence and tied it to a “distinctly British AI future” and to the UK’s G20 presidency. Business Secretary Jonathan Reynolds said people need to feel the technology is safe. Huang put the engineering claim in one line.

AI capability and safety must advance together. The U.K. brings world-class scientific rigor to this essential engineering challenge. With this $250 million commitment, NVIDIA will put accelerated computing in the hands of U.K. researchers to develop, test and strengthen safeguards for increasingly powerful AI.

Jensen Huang, founder and chief executive, Nvidia, UK newsroom statement

The tests will run on Nvidia GPUs, through Nvidia partners, using tools Nvidia has already shipped, including the OpenShell runtime and Sentry reference design from September 28. In London, Huang also said that, in a hypothetical, Nvidia would stop shipping chips to developers building unsafe systems, after a conversation first. The company that donates the safety cluster still holds the power switch.

Brussels Is Paying for Factories Nvidia Offered to Fill

While Huang was working Paris and London, the Commission’s call for up to seven AI Gigafactories was already open. It launched on July 30 and closes on November 12. Public money of up to €10 billion is meant to pull in at least €20 billion of private capital, more than €30 billion in all. Eighteen member states have signed the joint buying pact, France among them.

WHAT THE GIGAFACTORIES TENDER SETS OUT

• Lot 1: Up to four projects, each able to draw €100 million in phase one and €400 million in phase two.
• Lot 2: Up to three larger projects, each able to draw €200 million then €800 million.
• Hardware letters: The Commission signed letters of intent with Nvidia, AMD and Qualcomm so bidders can reach chips; the letters are not purchase contracts.
• Clock: Awards expected in early 2027, with sites meant to run no later than 18 months after signing.

Huang’s sovereignty pitch lands on that calendar. Europe is paying to own buildings, power contracts and operating rules. Nvidia has already offered to put its accelerators inside those buildings, and it is taking equity in the French software that will queue jobs on them. AMD and Qualcomm signed the same kind of letter, so the tender is not a single-vendor award. The company touring capitals this week is still the one with the fullest software lock and the fattest order book.

European cheques are still being written without Santa Clara attached, including a raise of $400 million to stay with European AI firms. Those funds can keep a founder from selling the whole company. They cannot, on their own, stand up a competing GPU.

The Hugging Face Deal Still Waiting on Regulators

On September 3 Nvidia agreed to buy Hugging Face for $12.93 billion, about $11.9 billion for investors and up to $1 billion in equity retention for staff who join. The New York platform was founded in 2016 by French entrepreneurs Clément Delangue, Julien Chaumond and Thomas Wolf. Huang’s own note put more than 18 million developers on it, sharing more than 3 million models, 500,000 datasets and 1 million applications, with more than 200,000 companies using the hub. Nvidia says it has put more than 500 models and more than 250 open datasets there.

Huang wrote that Hugging Face will stay open, that builders can pick their models, clouds and chips, and that Nvidia compute will not be required to ship through the platform. The close is aimed at 2027 and still needs merger review. Until that review ends, Europe’s main public square for open weights is under contract to the same firm now buying French product companies and donating UK safety cycles.

Open weights are how Huang tells European ministers they can catch up. They are also how Nvidia keeps the free-model world from drifting onto someone else’s accelerators. The French stakes this week sit downstream of that bet: if the models live on a hub Nvidia owns, and the hospitals and robots that fine-tune them run on Nvidia cards, the “European” stack has a Californian basement.

Place Vendôme to Berlin on the Same Circuit

GTC Berlin runs October 20 to 22, and Huang’s keynote is set for 11 a.m. CEST on October 21, inside the window he gave for the extra $2 billion of European deals. Mistral’s Large 4 weights are due six days later. The Gigafactories clock runs to November 12. None of those dates changes the mechanics already on the table.

France gets capital and a path to models it can keep in-house. The UK gets a five-year safety cluster on home soil. Nvidia gets more trained demand, more political access, and another set of workloads that default to its boards. The revolution Huang described in Paris is underway. The meters on it still read Nvidia.