France in the AI race: Mistral, energy and talent

On September 8, 2026, Mistral AI announced the largest funding round ever closed by a European tech company: €3 billion. A week later, TotalEnergies signed a program worth more than €100 million with it. Add nuclear power and tens of billions promised for data centers, and the story writes itself: France is in the race. We went through the facts one by one before giving you our take.

In short: France has real strengths: a champion, Mistral AI, valued at more than €21 billion; electricity that is over 95% low-carbon; and researchers that labs from around the world come to Paris to hire. Its limits are just as real: sums that don’t compare with those in the United States, dependence on American chips, and data centers that are still largely on paper. TotalEnergies did not buy into Mistral: it is a three-year partnership. Our take: France can matter in specific niches (industrial, sovereign, open AI), not compete dollar for dollar.

February 2025: the summit that set the numbers

On February 10 and 11, 2025, Paris hosted the AI Action Summit at the Grand Palais, co-chaired by Emmanuel Macron and Indian Prime Minister Narendra Modi. The number that stuck: €109 billion in private investment in France, announced by the French president. This isn’t public money but pledges spread over several years, mostly from abroad and mostly for data centers.

Investor	Amount announced	What for
United Arab Emirates (MGX fund)	€50bn	an AI campus of about 1 gigawatt
Brookfield (Canada)	€20bn by 2030	data centers, including one in Cambrai
Fluidstack (UK)	€10bn	a 1-gigawatt supercomputer
Amazon (US)	€6bn by 2031	data centers and cloud
Digital Realty (US)	€5bn	new sites in Marseille and the Paris region
Evroc (Sweden)	up to €4bn	a data center in Mougins
Iliad (France)	over €3bn	data centers
Mistral AI (France)	several billion	data centers, including in Essonne

Breakdown: franceinfo, February 2025.

On February 11, Ursula von der Leyen launched InvestAI: €200 billion to be mobilized for AI in Europe, including €20 billion for AI “gigafactories,” which she compared to a CERN for AI. The final declaration was signed by 58 countries, without the United States or the United Kingdom.

Energy was at the heart of the pitch. Answering Donald Trump’s “drill, baby, drill,” Macron said: “Just plug, baby, plug.” Electricity, he added, was available and ready. Three days earlier, minister Clara Chappaz had presented 35 “ready-to-use” sites for data centers, with high-capacity grid connections promised from 2027.

Mistral AI: from €105 million to €3 billion in three years

Mistral AI was founded in Paris on April 28, 2023, by Arthur Mensch, formerly at Google DeepMind, and Guillaume Lample and Timothée Lacroix, both from Meta (Wikipedia). Its funding rounds tell the story of the acceleration:

• 2023-2024: €105 million in June 2023, €385 million in December, €600 million in June 2024 (valuation: €5.8 billion).
• September 9, 2025: a €1.7 billion Series C at an €11.7 billion post-money valuation. Dutch company ASML, which builds the lithography machines every advanced chip depends on, put in €1.3 billion for roughly 11% on a fully diluted basis, plus a seat on Mistral’s strategic committee.
• September 8, 2026: a €3 billion Series D at a valuation above €21 billion, led by Samsung Electronics, with the EQT-managed Scaleup Europe Fund and PSG Equity as co-leads. ASML, Nvidia, a16z and General Catalyst also came back in.

What about revenue? Mistral doesn’t publish its accounts: Pappers, the French company registry service, shows none. At Davos in January 2026, Arthur Mensch claimed more than $400 million in annualized revenue, according to Agentland. At the time of the Series D, the company was targeting €1 billion in revenue for 2026, according to franceinfo. These are company figures that no outsider can check.

The money goes to compute first. After launching Mistral Compute with Nvidia in June 2025, Mistral raised $830 million in debt from seven banks in March 2026 for a 44-megawatt data center in Bruyères-le-Châtel, south of Paris, with more than 13,000 Nvidia chips, scheduled for the second quarter of 2026 (TechCrunch). Then came €1.2 billion in Sweden with EcoDataCenter for a site opening in 2027, and a target of 1 gigawatt in Europe by 2030. Mistral is also a partner, alongside MGX, Bpifrance and Nvidia, in the 1.4-gigawatt “Campus AI” near Paris, expected in 2028.

On the customer side, it is betting on industry: Airbus, BMW and CMA CGM in May 2026 (AFP via France 24), and a five-year framework agreement with EDF on nuclear engineering and the future EPR2 reactors (ActuIA). In July 2026, Microsoft made a multibillion-dollar commitment to use part of Mistral’s European compute capacity.

TotalEnergies and Mistral: what was actually signed

Plenty of headlines talked about “TotalEnergies investing in Mistral.” The press releases say something else, in two stages.

1. June 12, 2025: TotalEnergies announced a collaboration with Mistral AI and a joint innovation lab, with no amount disclosed. On the list, among other things: an AI assistant for the group’s roughly 1,000 researchers.
2. September 15, 2026: the two companies announced “a 3-year joint program representing an investment of more than €100 million”. The goal: frontier AI models to help TotalEnergies’ geoscientists explore, characterize and develop oil and gas reservoirs. A joint scientific lab will bring together the group’s subsurface teams and Mistral researchers.

The takeaway: this is an industrial partnership, not an equity stake. The release doesn’t say how the €100 million is split, TotalEnergies is not among the Series D investors, and Maddyness also calls it a partnership. And here, French AI will be used to find oil and gas, a long way from the “clean AI” image that nuclear power helps sell.

The energy advantage: real, but not automatic

RTE’s figures are solid. In 2025, France generated 547.5 TWh of electricity, 373 TWh of it nuclear, and 95.2% of its output was low-carbon. It exported 92.3 TWh net, a record for the second year running.

The announcements followed. In March 2025, EDF opened its own land to data center developers, in Seine-et-Marne and Moselle. At the 2026 Choose France summit, SoftBank announced up to €75 billion for 5 gigawatts of data centers, including €45 billion for 3.1 GW in northern France (Hauts-de-France) by 2031.

But RTE adds perspective: data centers connected to its transmission grid used only about 1 TWh in 2025, and roughly 30 GW of grid access rights allocated to large industrial and digital projects are slow to materialize. As early as February 2025, RTE’s Thomas Veyrenc estimated that fewer than 15% of projects had confirmed their commitment (LeMagIT). And Next points out that the €93 billion of Choose France 2026 mixes previously known announcements with investments spread out until 2032.

Talent: the real raw material

Arthur Mensch’s path sums up the French pipeline: École polytechnique, Télécom Paris, the MVA master’s at ENS Paris-Saclay, a PhD at Inria, then almost three years at DeepMind in Paris (Wikipedia). American companies saw it early: Facebook opened its FAIR research lab in Paris in June 2015 and signed an agreement with Inria to recruit researchers and PhD students.

The ecosystem has grown denser since. Kyutai, a non-profit research lab launched in November 2023 and funded by Iliad, CMA CGM and Schmidt Sciences, publishes its voice models as open source. Yann LeCun left Meta at the end of 2025 to found AMI Labs, headquartered in Paris, which raised $1.03 billion in March 2026.

The European framework: rules and gigafactories

The AI Act applies in stages: obligations for general-purpose models since August 2, 2025, and transparency duties (telling people they’re talking to an AI, labeling deepfakes) since August 2, 2026 (Gibson Dunn). However, the Digital Omnibus regulation of July 8, 2026 pushed the rules for high-risk systems back to December 2, 2027, and to August 2, 2028, for AI built into regulated products.

For gigafactories, an informal first call drew 77 proposals from 16 member states. The official call, published on July 30, 2026, closes on November 12: up to seven gigafactories will be selected in early 2027.

The weak spots: capital, compute, chips

• Capital. According to Stanford’s 2026 AI Index, private AI investment in the United States reached $285.9 billion in 2025, more than 23 times China’s. Mistral’s €3 billion, a European record, is barely more than 1% of that annual total.
• Compute. The same report counts 5,427 data centers in the United States, more than ten times any other country.
• Chips. According to Stanford, TSMC manufactures almost every leading AI chip, and Mistral’s data centers run on Nvidia hardware. ASML is European, but it sells machines, not processors.
• Foreign money. Samsung, MGX, SoftBank, Brookfield: a large share of the billions comes from outside Europe. According to franceinfo, Europe still holds two-thirds of Mistral’s capital and three-quarters of the board’s voting rights.

Two readings that clash

The optimists: Emmanuel Macron presented the Series D as a step toward “a third way in AI,” according to TechCrunch, and Arthur Mensch keeps repeating that Europe must build its own infrastructure (TechCrunch).

The skeptics: Régis Castagné, head of Equinix France, was already asking in 2025 for stable electricity prices and proof rather than announcements (LeMagIT). ActuIA notes that Mistral’s partnerships with EDF, BMW and Airbus come with no public amounts or timelines.

Our take: can France really matter?

Yes, but not everywhere. Our take: France won’t win a gigawatt or dollar race against the United States. It can, however, hold three niches. Industrial AI: Airbus, EDF, ASML and TotalEnergies want models built for their sensitive data. Sovereignty: public and regulated customers pay to know where their models run. Energy, as long as grid connections keep up.

The risk? That France mostly becomes a location for data centers funded by foreign money, running models designed elsewhere. Three signals to watch: whether the Bruyères-le-Châtel site actually ramps up, the gigafactory selection in early 2027, and Mistral’s revenue, if it is ever published.

What it changes for you as a student

• Jobs, not just for researchers. Mistral now has more than 1,000 employees (Maddyness). Data centers will need engineers and technicians in power, networking, cooling and cybersecurity. Choose France 2026 promises 15,600 jobs, mostly at full capacity around 2030 (Next).
• Hybrid profiles. The TotalEnergies-Mistral lab will need geoscientists who speak AI; the EDF deal, nuclear engineers comfortable with models. Your own discipline plus real AI literacy is an edge.
• Schools and internships. The research track (top engineering school, specialized master’s, PhD) produces founders, but it isn’t the only way in: before picking a data or AI master’s, check how its graduates actually fare. Kyutai hosts research interns and PhD students (team).
• Tools. Mistral’s Le Chat became Mistral Vibe in May 2026; the pricing page lists a $5.99 monthly plan for verified students instead of $14.99. Use it as a coach that explains and asks you questions, never to hand in work in your place.

FAQ

Did TotalEnergies invest in Mistral AI?

Not in its equity. TotalEnergies announced a collaboration in June 2025, then, on September 15, 2026, a three-year joint program worth more than €100 million with a shared lab. The group is not among the shareholders listed in the Series D.

How much is Mistral AI worth in 2026?

More than €21 billion after its €3 billion Series D, announced on September 8, 2026, and led by Samsung. It is the largest round ever raised by a European tech company.

Why does nuclear power matter so much for AI?

A data center uses a lot of electricity, around the clock. France’s electricity is more than 95% low-carbon and it exports large amounts, which attracts investors. The bottleneck is grid connection, not generation.

Do you need a PhD to work in AI in France?

For fundamental research, often yes. For engineering, data, infrastructure or energy, no: companies also hire engineers and technicians.

Further reading

• French Tech and AI: the French startups to know in 2026: who’s hiring, their funding, their tools.
• Yann LeCun and AMI Labs: the “world models” bet, made from Paris.
• Why AI costs so much: chips, energy, data centers.
• The insertion-pro connector: real graduate outcomes for a master’s program, before you commit.