“We’re turning electricity into intelligence.” That is probably the phrase that best sums up Arthur Mensch’s testimony before the French parliamentary committee of inquiry. For nearly an hour and a half, the co-founder and CEO of Mistral AI outlined an extremely industry-focused vision of artificial intelligence, a far cry from mere consumer chatbots.
Mistral AI itself already gives an idea of the scale it has reached. Founded on 28 April 2023, the company now employs 1,000 people, is targeting €1bn in revenue by the end of the year and has a valuation of €12bn. Arthur Mensch also stated that the company is investing around €1 billion this year in research and development.
1. AI is no longer just a sector; it is digital technology itself
“The digital services of tomorrow are essentially artificial intelligence,” explained Arthur Mensch. “The cloud is artificial intelligence.” In his view, Europe is making a strategic mistake by continuing to treat the cloud, digital infrastructure and generative AI as separate entities. Traditional hosting and storage activities are becoming low-margin commodities, whilst value creation is now centred around AI models.
His reasoning is very straightforward. “You can’t build a scalable business starting from a position where margins are low,” he insisted. European players must therefore position themselves directly within the most profitable technological sectors.
2. The battle is now about energy
Arthur Mensch described artificial intelligence as an energy-intensive industry. “We’re turning electrons into tokens,” he repeated several times. The figures cited give an idea of the scale of the change. Building 1GW of AI capacity would require around $50bn in investment over five years. With this level of infrastructure, an operator could generate around $20bn in annual revenue from AI services. According to him, the United States has already fully embraced this approach. “The United States is saturating its electricity grid to convert it into artificial intelligence,” he stated.
Arthur Mensch believes that Europe is still completely underestimating the scale of future needs. His estimate is that AI could account for around 1kW of computing power per person within five years. This would mean an additional 40GW of power to be generated in France and nearly 400GW for Europe. “That amounts to around €20trn in investment,” he told MPs.
3. The window of opportunity is extremely short
“It all comes down to the next two years,” insisted the head of Mistral AI. In his view, the main risk is that US players will secure electricity capacity and infrastructure before the Europeans have organised their demand. “Once the supply is monopolised by US players, we’ll effectively have no supply left.”
France currently has an average electricity surplus of around 9GW. For Arthur Mensch, this surplus is becoming a strategic global asset. Companies capable of securing this capacity quickly will gain a potentially irreversible lead. Mistral AI has, in fact, set its own target: to reach 1GW of capacity by 2029. But Arthur Mensch himself acknowledges that this will be “not enough”.
4. The topic turns to macroeconomics
Arthur Mensch has repeatedly highlighted the risk of a massive trade deficit if Europe were to import the bulk of its AI services in the future. His reasoning is based on a simple assumption: if artificial intelligence accounts for around 10% of the European wage bill, the market could be worth around €1 sextillion. “If we import this non-European technology, that’s a €1 sextillion trade deficit on top of our existing trade deficit in digital services,” he warned.
According to him, Europe faces the simultaneous risk of a surge in the digital trade deficit, inflationary pressures linked to electricity, and job losses or drastic changes to the world of work. “It’s revolutionary,” he summed up.
5. Europe must use public procurement as an industrial weapon
Arthur Mensch explained that the United States and China make extensive use of public procurement to shape their technology ecosystems. “It’s a tool we’ve always been afraid to use in Europe,” he lamented.
Mistral AI currently generates around 20% of its revenue from public sector contracts, with around 10% of that coming from French public sector contracts. The company works with organisations such as Dinum, France Travail, the Caisse des Dépôts and the French Ministry of the Armed Forces.
Luxembourg was cited as being more advanced than France in certain areas of deployment. “Luxembourg has significant framework agreements through which we are rolling out services in central government,” said Arthur Mensch.
The executive believes that public authorities should significantly step up their adoption of generative AI in order to offset demographic and budgetary constraints. “Central government bodies should view generative artificial intelligence as a means of generating the productivity they need.”
6. European fragmentation is holding things back
Arthur Mensch gave a very detailed account of the operational hurdles faced by European tech companies. “I’ve signed hundreds of documents to set up companies in around ten European countries. I’ve opened dozens of bank accounts.”
The CEO of Mistral AI criticises a continent where tax, social security and regulatory systems remain largely fragmented. In his view, this situation significantly hampers the growth of European companies compared to their American competitors. He also gave a striking example from the telecommunications sector: Europe is said to have around 60 telecoms operators, compared with just three major players in the United States. As a result, their investment capabilities are incomparable.
7. European regulations risk strengthening the American giants
“Regulation favours the big players,” said Arthur Mensch. The head of Mistral AI believes that large US platforms have the financial, legal and lobbying resources needed to absorb European regulatory costs. European start-ups, on the other hand, are seeing their fixed costs skyrocket much more rapidly.
He cites, in particular, the GDPR, the AI Act, data regulations and copyright law. “All these things deal with similar topics but require different documentation.”
Mistral AI already has a five-person compliance team dedicated to these issues. Arthur Mensch points out that this cost is manageable for a company valued at €12bn, but much less so for European start-ups.
8. AI is becoming a critical component of defence and cybersecurity infrastructure
Mistral AI is already working with the French Ministry of the Armed Forces and several of France’s allies. Arthur Mensch highlighted, in particular, the growing use of AI in military applications. “Artificial intelligence is now at the heart of military operations centres.” He also believes that the new models are now capable of detecting software vulnerabilities, orchestrating cyberattacks or automating certain offensive operations.
“You can’t get hold of the French army’s code bases that have been scanned by Mythos,” he said.
9. Productivity gains are already massive
Arthur Mensch explained that engineers at Mistral AI “no longer write lines of code” in the traditional way. “These days, you’re no longer a craftsman; you’re a manager.” According to him, developers now oversee agents capable of generating code themselves. For some users, productivity gains could range from five to twenty times, depending on the use case.
Mistral AI currently estimates its own internal AI expenditure at around 10% of its total payroll. Arthur Mensch puts the average cost at around €10,000 per employee per year for a dedicated, permanent AI assistant. “€10,000 a year will get you roughly 1kW of GPU power.”
10. Mistral AI rejects the logic behind the US takeover
Arthur Mensch confirmed that American investors regularly approach the company. But his response was very clear: “Our mission is to remain independent.” American investors currently hold less than 30% of Mistral AI’s capital. The CEO explicitly advocates a strategy of European independence and a future stock market listing. “If you get bought out, in a way you’ve failed.” The conclusion of his hearing ultimately sums up his entire geopolitical vision of AI. “We absolutely must do this, because otherwise we will become a vassal state.”
