True to the strong winds that inspired its name, French start-up Mistral AI took Davos by storm in 2024, having delivered a world-class artificial intelligence model with a fraction of the usual resources.
The Paris-based start-up, less than a year old, was on a high. It was freshly valued at $2bn and had the backing of AI chip leader Nvidia and prominent venture firm Andreessen Horowitz. Mistral’s founding trio of hotshot AI researchers — Guillaume Lample, Timothée Lacroix and chief executive Arthur Mensch — were hailed as the heroes who would finally put Europe at tech’s top table.
Mistral also had the enthusiastic support of French President Emmanuel Macron, who was drawn in by the start-up’s promise of “sovereign” and more “open” AI, proudly independent of US Big Tech.
But a year is a long time in AI. Excitement about Mistral started to cool as it was seen to be struggling to keep up with its larger rivals in the AI race.
Then, this week, came a blast of cold from the east. China’s DeepSeek stunned Silicon Valley by releasing a cutting-edge open-source model with what it claims is a tiny fraction of OpenAI or Meta’s resources and computing power — beating Mistral at its own game.
Mistral was founded on the idea that it had discovered more efficient ways to build and deploy AI systems than its bigger competitors. Yet seemingly overnight, the little-known Chinese lab had gone even further in terms of efficiency to steal a march on much better resourced US tech companies, prompting a huge market sell off of once high-flying AI stocks.
Though some supporters say this affirms Mistral’s approach, others see it as a threat to its business model of delivering affordable, “open” AI. Europe’s increasingly anxious tech investors worry that its $1.2bn in funding — a huge sum for a French company of its size and age — remains inadequate by Silicon Valley’s standards. Its biggest US rivals now have war chests that are 10 times bigger.
At this year’s Davos, Mistral’s Mensch was forced to deflect questions about whether his company would have to sell to a Big Tech company as many other smaller players have done.
But he insists that Mistral is not for sale and indicates that it hopes to go public one day. “We think that what we are doing is important [to do] as an independent company,” he tells the FT. “So this is not on the table.”
One investor in Mistral is less bullish in private. “They are starting to see the writing on the wall,” says the person. “They need to sell themselves.”
Europe has a lot riding on the company’s fate, just as generative AI begins to reshape the way people live and work.
Although the region is home to promising AI start-ups — such as the UK’s Wayve, Germany’s DeepL and Black Forest Labs, and France’s Poolside — none are currently working on large language models, the general-purpose AI system that underpins ChatGPT. Aleph Alpha, once Germany’s hope for an LLM domestic champion, pivoted away from LLMs last year, leaving Mistral as the only consequential player in Europe.
If Mistral fizzles, then Europe’s businesses and consumers will have little choice but to depend on a handful of American — or Chinese — platforms. For many European leaders and companies, having no sovereignty or influence over a technology that has the potential to affect every corner of work, culture and society is a nightmare scenario.
It would also reinforce growing concerns over the declining competitiveness of the EU economy, just as President Donald Trump seeks to turbocharge US growth with wide deregulation and a more confrontational approach to trade.
Trump has already clashed with EU leaders over the continent’s push to regulate US tech companies, having a local champion could give Europe vital leverage — or a back-up plan if transatlantic relations deteriorate dramatically. Under “AI Diffusion” rules proposed by the outgoing Biden administration, several European countries already face restrictions on how many of the most powerful AI chips they can buy.
“It’s not like we are beating America in LLMs, but at least [in Mistral] there is a European contender, which is really good,” says Niklas Zennström, the Swedish founder of communications app Skype and chief executive of venture firm Atomico, which is not a Mistral investor.
“[Tech] sovereignty for Europe is more important now than it ever was before.”
From its inception, Mistral has been intertwined with concerns about whether Europe could compete with Silicon Valley.
Macron, who has praised Mistral as “an example of French genius”, became the company’s biggest cheerleader. The French president — and other European governments and policymakers — desperately want to avoid a repeat of the early 2000s, when the region was sidelined in the rise of internet platforms and social networks. Its once-strong telecom and broadband infrastructure companies also withered under Chinese competition.
Shortly after his election in 2017, Macron vowed to turn France into a “start-up nation” by encouraging entrepreneurs and stoking venture capital investment. Later, he set a goal of boosting the number of so-called tech unicorns — those with valuations above $1bn — from fewer than 10 to 100 by 2030.
In 2018, his digital minister Cédric O — who later became an adviser to and investor in Mistral — advocated for a controversial step: to actively woo big US tech giants, including Google and Facebook (now Meta), into creating research centres for AI in France. They would capitalise on programmes such as those at École Polytechnique, the country’s most prestigious science and engineering school.
The intervention helped slow the brain drain of skilled engineers and AI scientists to Silicon Valley, including Lample, Lacroix and Mensch, who had worked for such AI labs at Meta and Google before founding Mistral.
“If we hadn’t done that, they probably would’ve been in Palo Alto instead,” says O.
Funding from VCs has roughly tripled since 2017, and the number of unicorns now stands at around 30.
But, by mid-2022, Jean-Charles Samuelian-Werve, the co-founder of Paris-based insurance-tech start-up Alan, began to worry about the emergence of OpenAI and other generative AI technologies being created primarily in the US. In his telling, he “started to panic” that Europe would once again “have no control” over powerful technology.
“What was even more frustrating was that much of the research going into LLMs was actually being done by European scientists,” says Samuelian-Werve, a respected figure in global tech circles.
Alongside another colleague at Alan, he prepared a memo outlining the state of AI technology and sent it the Elysée as well as key people in government, tech and universities. In it, they argued that France needed a publicly funded lab or foundation to help develop its own generative AI technologies, requiring an infusion of up to €3bn in public and private money.
French telecoms tycoon and tech investor Xavier Niel had been thinking along similar lines and reached out to Samuelian-Werve, who went out scouting for talented AI scientists and met Lample, Lacroix and Mensch.
The idea of building a non-profit lab quickly fell by the wayside since the trio of engineers wanted to create a business instead, which was briefly dubbed EuroAI.
Samuelian-Werve and ex-minister O became early investors and advisers; Niel joined Mistral’s first round of financing in June 2023, when it raised an impressive €105mn only a month after it was founded. By the end of that year, its first models had wowed the AI developer world.
“We helped them on early financing as well and structuring the company initially . . . but they were the ones to execute all this,” says Samuelien-Werve. “The idea of strategic autonomy for Europe in generative AI was important, but Mistral wants to be a global champion.”
Mistral’s leaders have always pitched capital efficiency as its greatest asset.
“With 100 times less [computing power than US rivals], we’ve been able to make models that are pretty much on the frontier,” Mensch tells the Financial Times.
The approach won Mistral fans, including Microsoft, which signed a partnership with the start-up and took a small stake. It was Microsoft’s first investment in an LLM other than OpenAI.
Technical benchmarking sites, such as RankedAI.co, place Mistral among the world’s top 10 model developers. But newer rivals, not least China’s DeepSeek, are threatening to overtake it.
“The Chinese undoubtedly now have stolen that baton” as the “fast follower” to OpenAI and its US rivals, says Sean Maher, founder of economic consultancy Entext. He describes DeepSeek’s latest model as a “jaw-drop moment”. “It’s going to change the economics of the whole industry.”
The arrival of DeepSeek also undercuts the assumption that the US big players had built up an insurmountable lead because they had more computing power and key chips; the Chinese lab’s LLM delivered strong performance on the equivalent of a computing power shoestring.
Though the news about DeepSeek is still being digested, Mistral’s proponents argue that its disruption might benefit the company too. They say that DeepSeek’s breakthrough only validates Mistral’s founding strategy that cutting-edge AI can be built for far less. They also highlight the control, privacy and neutrality that Mistral offers to corporate customers, in contrast to DeepSeek, which collects a lot of data and adheres to Chinese censorship. Mistral declined to comment.
Nonetheless, critics maintain that Mistral, not yet two years old but valued at €6bn in its last fundraising of €600mn last June, is in an AI start-up limbo: it has raised too much to fade quietly into the background, yet not enough to keep up in the global AI race. It has around 150 employees, compared with thousands employed by its US rivals.
Asked if Mistral plans to raise more capital this year, Mensch said: “Possibly, although we could do without it. There’s definitely some interest out there already.”
Maher predicts that Mistral will go the way of Adept and Inflection — promising AI start-ups whose talent was “acquihired” by Big Tech. That is, if Brussels antitrust regulators allow such a strategic European asset to be taken over by a US buyer. “The battlefield has changed shape,” says Maher. “[Mistral] needs to figure out where to play or get wiped.”
Europe has long been better known in Silicon Valley for tech regulation — with the EU seeking to set rules on everything from content moderation to competition — than for innovation.
But when it comes to AI, European tech investors, and some governments, want the region to have their own companies that are also competitive on the global stage. To that end, when the EU was debating its first flagship AI regulation in late 2023, Macron and others warned Brussels not to slow the development of the nascent sector with too much red tape.
“We can decide to regulate much faster and much stronger than our major competitors,” the French leader said at the time. “But we will regulate things that we will no longer produce or invent.”
Macron will again press that message when he hosts next month’s AI Action Summit in Paris, a follow-up to 2023’s UK event at Bletchley Park.
Mensch has also talked about the importance of Europe having its own AI champions. Unveiling a partnership in mid-January with Agence France-Presse to provide news for its chatbot Le Chat, he told the FT that “Europe must unite to defend its thriving technological sector”.
But Mistral’s chief executive also knows that that kind of rhetoric is not necessarily the best way to build a global business. As such, the company is rapidly expanding its Silicon Valley offices, both to attract engineering talent and to sell to US customers.
“Our European DNA is never the argument for selling or for getting customers,” he says. “The reason why we started Mistral was to promote a more decentralised AI deployment model.”
Mistral won early fans in the software developer community because its “open source” origins means some of its models are available under a licence that allows users to examine the “weights” that shape the output or make derivative works. But Mistral’s more advanced models, such as a well-received new programming tool, are only available commercially and it has struck cloud distribution deals with Microsoft, Amazon and Google.
Mensch says that customers most value the ability to personalise Mistral’s AI systems, deploy them on any kind of IT infrastructure and “have stronger data governance than what is provided by our US competitors”.
One such customer is the French defence ministry, which recently signed a deal with Mistral after benchmarking its open-source models against those from Google and Meta. The ministry’s AI agency determined that Mistral performed as well as rivals, while also offering the sovereignty and security needed for defence IT systems, which are completely disconnected from the internet.
Mistral also has several prominent French companies as customers, such as bank BNP Paribas, the shipping company CMA-CGM, and the telecom operator Orange. But Mistral insists that it is global: a third of its revenue now comes from the US, where its customers include consumer giant Mars and tech companies IBM and Cisco. European customers include online retailer Zalando and enterprise software maker SAP.
Backers say that Mistral’s revenue growth has been quick for such a young business, albeit coming off a small base. Investors familiar with Mistral’s finances say that its annualised revenue run rate — a measure that extrapolates from its most recent monthly performance — is in the tens of millions of dollars. Meanwhile, Anthropic reportedly made close to $1bn in sales last year, while OpenAI generated almost $4bn.
A study by Menlo Ventures, a Silicon Valley VC firm, ranked Mistral fifth in the enterprise AI market, with a market share of just 5 per cent last year — less than half Google or Meta’s share and far behind OpenAI.
Some European tech founders and investors argue that focusing on efficiency was a tactical error at a time when there was nearly unlimited capital available for frontier LLM developers.
“It’s extraordinary what they have been able to achieve, but they are the last gasp of the old paradigm — trying to play the scale game with a tenth of the resources of their rivals,” says one UK tech investor, who does not own shares in Mistral.
But Anjney Midha, general partner at Andreessen Horowitz, who sits on Mistral’s board, argues that the company’s efficiency is why they backed it in the first place: “It has allowed Mistral to strike and execute with a speed and precision unlike anything I’ve ever seen.”
Mensch is also adamant that resource constraints are a feature, not a bug. Technical efficiency keeps prices low for customers and a lid on Mistral’s costs, he argues, as well as being a forcing function for innovation.
“If you have unlimited flops [a measure of computing power] to spend, you end up doing a lot of useless stuff,” he says. “Necessity is the mother of invention.”
Additional reporting by Arash Massoudi, Stephen Morris, Cristina Criddle, George Hammond, Madhumita Murgia and Ivan Levingston
Read the full article here