Paris-based Mistral set out to develop a top-tier AI model to rival OpenAI and anthropic. That didn’t go as planned. But it turns out many customers don’t care if the AI is cutting edge — as long as it wasn’t made in America or China.
When Arthur Mensch, co-founder and CEO of mistral, France’s leading AI company, took the stage at the AI ACTion Summit in New Delhi in February, he drew only a small crowd. Most attendees preferred to hear openai’s Sam Altman or Anthropic’s Dario Amodei preach the promises and perils of superIntelligent AI. But the select group of executives and researchers who listened to Mensch heard a very different message: the rest of the world should control its own AI destiny, not Silicon Valley. And Mensch can help them do it. “AI should be a tool for Empowerment, not dominance,” he declared.
Mensch’s vision for Mistral, and for AI itself, can be summed up in one word: independence. Unlike its black-box Silicon Valley rivals, most of Mistral’s AI Models are “open weight.” With this open-source APProach, customers can get under the hood, customize the AI using their own data, or download it for free to run offline or on a laptop.
The message resonates. Traditional executives are unsettled by the world-consuming rhetoric of OpenAI and Anthropic, along with the emerging threat of Chinese AI firms. Mensch’s focus on control and sovereignty is reassuring, as is his pitch that Mistral will deploy engineers to set up and run the Technology for CLIents. Their data doesn’t even need to leave the office, let alone the country.
“We are really the only company that allows building core Business Automation and products on top of an open stack, and that is something that is valuable everywhere in the world,” says Mensch, 33, from Mistral’s offices in Paris’s trendy 10th ARRondissement, as children play soccer in the courtyard outside.
European organizations are particularly interested in Mistral. At a time when a German state government is replacing Microsoft Office for official business, and France is rolling out its own alternative to Zoom for video calls, there is a clear opportunity for Mensch to step in with secure, proudly European-built AI.
Donald Trump is another driver of business. The president’s trade war, threats to annex Greenland, and promises to shield American tech companies from regulation have fueled concerns about reliance on anything American — from software to data centers and now AI. “The independence we provide to our customers is critical for our product,” Mensch says.
Mistral needs every advantage it can get. Mensch and co-founders Guillaume Lample and Timothée Lacroix are among France’s finest technical talents, having worked at the Paris outposts of top American AI labs. Yet Mistral has fallen further behind in AI performance rankings. According to one popular benchmark, Mistral’s best model would lose in a face-off against a version of Anthropic’s claude released nine months earlier. It is also outperformed by a new crop of open-weight models from Chinese startup DeepSeek and tech giant Alibaba.
That is not surprising. Mistral’s American rivals are loaded with cash, willing to spend more each year than the $3.1 billion Mistral has raised to date — including from French institutions like BNP Paribas and Bpifrance. Chinese competitors claim they can train AI models more cheaply, though they are widely suspected of distilling secrets from American AI giants by Prompting Claude and ChatGPT millions of times to train their own models.
In an industry obsessed with performance, that should make Mistral an also-ran. But Mensch bets that a smaller, cheaper model made in Europe is better suited for governments and global companies than a vastly more powerful American closed-source large language model. Moreover, it is too risky for serious Western companies to rely on Chinese models, says Mistral investor Jeannette zu Fürstenberg of venture fund General Catalyst. The strategy has generated around 80 million monthly by December, though high compute and data costs mean the company is not yet profitable.
“The question to ask is, ‘Is Mistral at the top of the independence leaderboard?’” says Anjney Midha, who led Andreessen Horowitz’s Investment in a $415 million round for Mistral in 2023 and now runs his own AI investment firm, AMP.
Mistral has secured deals with London-based HSBC — Europe’s second-largest bank, with over 70 billion), and CMA, the world’s third-largest shipping company by capacity (100 billion company, it’s only because they screwed it up,” zu Fürstenberg says.
Silicon Valley observers might dismiss the startup as little more than a System Integrator, with a chunk of its revenue coming from Palantir-style consulting deals rather than cutting-edge AI. But that hardly matters. The strategy is working. Mistral now has the backing of ASML, Europe’s most valuable tech company (2 billion round into Mistral in September and signed a deal to use Mistral’s AI in its products and research. That round valued the startup at 1.8 billion.
To keep growing, Mistral just needs to fully exploit its niche as a safe haven from the AI Superpowers in the U.S. and China. It cannot, of course, give up on improving its models entirely. At some point, OpenAI and Anthropic’s models might become so advanced that some large customers will trade safety and sovereignty for raw performance.
Mensch was born in a Paris suburb to a physics teacher mother and a father who runs a small computer server company. A third-generation computer scientist — his grandfather worked on health data systems — he crossed paths with Lample, 35, at Paris’s elite École Polytechnique. In 2016, while studying for a PhD in AI at Pierre and Marie Curie University (now part of the Sorbonne), Lample landed a job at Meta’s AI research Arm, where he worked with Lacroix, 34. After earning a PhD from the University of Paris-Saclay, Mensch completed two years of postdoctoral research before joining Google’s Paris office to work on DeepMind in 2020.
There, Mensch contributed to a groundbreaking paper showing that large language models could be built far less expensively than previously believed. Lacroix and Lample applied those ideas to help build low-cost open-weight models at Meta’s Fundamental AI Research lab. When the project, called Llama, launched in February 2023, it was an instant hit: small, cheap, and powerful, perfect for academic researchers and startups on tight budgets.
The three quit their jobs soon after. “We had already started to think about what we could be doing here in France,” Mensch recalls.
For decades, through the dot-com boom, the explosion of social media, and the rise of the cloud, Europe has been stuck in the slow lane of tech. Mistral’s co-founders were convinced the continent needed its own AI models and that its socialistic governments and industrial giants would pay for them. They named the company after a powerful winter wind that sweeps across the Mediterranean. Silicon Valley VC firm Lightspeed led a $115 million seed round, at the time Europe’s largest, in 2023.
Mistral’s first models, released later that year, also made a splash. Mensch had delivered on his promise that great AI could be built and run at a fraction of the cost of OpenAI’s chatgpt. The company later launched its own ChatGPT-style app — dubbed Le Chat, naturellement — which scored 1 million downloads within its first seven weeks, the majority in France, according to Appfigures.
But Mistral was soon Dramatically outspent and outpaced. OpenAI and Anthropic have raised more than 840 billion and 13 billion in revenue, while Anthropic generated some $4.5 billion. According to a Menlo Ventures survey of 500 American enterprise executives, Anthropic holds a 40% market share, OpenAI has 27%, and Mistral has just 2%. (It is worth noting that Menlo Ventures is a major backer of Anthropic.)
By 2024, it looked as if Mistral was running out of wind. Critics said the company was losing ground, and its revenue that year was reportedly well under $50 million. Mensch concedes the team had “learned on the job” after working at research labs with little focus on commercialization. But revenue steadily ramped up as slow-to-negotiate yet sizable deals began to land.
“If we’re successful, Europe will be successful.”
Mistral CEO Arthur Mensch
To win blue-chip customers, Mistral borrowed an idea from Palantir: “forward-deployed engineers.” Mensch isn’t just selling an AI model; he’s dispatching highly Skilled staff to solve business problems. Mistral’s engineers will work with any open-weight AI model, not only their own, though Mensch says customers often prefer Mistral’s because “they just have more confidence in how they run and what kind of biases they may have.”
Mistral now has a team deployed in HSBC’s London offices to create AI tools that allow the bank’s 200,000 employees to automate repeatable tasks like compliance checks, says CIO Stuart Riley. He notes that HSBC works with multiple AI models, but Mistral fills a valuable niche for workflows involving sensitive data: “We obviously need to make sure these models and the data reside in exactly the right geography.”
It appears to be a running joke among Mistral’s 700-strong staff that the company’s future depends on challenging Palantir, which has ballooned to a $330 billion market cap. posters around the office play on Palantir’s name and the French word poulet (chicken). One appears to show Palantir’s billionaire CEO, Alex Karp, with a rooster’s head; another presents “Poulantir” going public on the NYSE.
Mensch acknowledges Mistral and Palantir share some overlap in target customers, but he likes his odds. Palantir has become increasingly controversial in Europe since Karp began echoing Trump-like rhetoric and securing a series of large federal contracts to build surveillance technology for the U.S. government. Yet Mensch must also compete against OpenAI and Anthropic, which are assembling their own teams of forward-deployed engineers.
One advantage that neither OpenAI nor Anthropic can replicate is what Mensch calls “community solidarity” with Macron and other European leaders. He underStands that Mistral’s fate is inseparably linked to Europe’s. “If we’re successful, Europe will be successful,” he says.
It is not just the continent. Mensch says around 40% of Mistral’s revenue comes from the U.S. and other non-European clients. There, the selling point is less about patriotism and more about control and cost. Boardrooms in America also have reason to worry about the giant AI companies’ ambitions. “I don’t think the Europe-versus-America Prism is the right one,” Mensch says. “I think the right one is open-source versus closed-source models.”
Inside Mistral’s office in Paris, Mensch is excited to discuss a new project: AIs that can control robotic arms. This is part of Mistral’s push to help Europe’s industrial giants regain ground as China and the U.S. race ahead in robotics. The company is also continuing to launch small, specialized AI models, including one released in February designed for superfast voice transcription.
To this point, Mistral’s young CEO has undeniably played a bad hand well. In the AI race, companies with vastly more resources — Microsoft, Amazon, Apple, and even Elon Musk’s xAI — have all floundered against the threat from OpenAI and Anthropic. Meta spent roughly $70 billion on AI last year only to delay the launch of a new model. xAI has blown through billions, yet its Grok chatbot has been outclassed by competitors and has run into trouble with global regulators.
For now, Mistral effectively holds a monopoly on European-built open-weight models. Meta appears to have wavered on building open successors to Llama, and OpenAI’s 2025 open-weight companion to GPT-5 has not seen the Same adoption, according to data from AI model library Hugging Face. But such open space may not last long. One of Mistral’s backers, NVIDIA, has begun releasing its own open-weight models and is pouring billions into their development.
The biggest risk for both Mistral and older U.S. tech companies is that Anthropic, OpenAI, and google’s lead in AI Coding will allow them to supercharge new models that can autonomously improve themselves. The big three already have AIs that code faster than human developers with strikingly few errors. The next generation of models could surpass large numbers of other white-collar professionals.
In response, Mensch is doubling down on independence. “A lot of our customers are telling us, ‘Can you provide me with Artificial Intelligence that is not running on anything owned by hyperscalers like Microsoft, Google, and Amazon?’” So Mistral is developing its own data centers, starting with one outside Paris. Mensch projects it will have 200 megawatts of capacity by the end of 2027. Power from France’s state-owned nuclear plants will help, but the buildout could still cost an estimated $5 billion. Mensch has tapped oil-rich Abu Dhabi and reportedly sought debt financing to help pay for it.
Being a local champion also comes with strings attached. It is unlikely that French or European antitrust regulators would approve a sale of Mistral to a foreign company, despite talk swirling last year that Apple planned to make an offer. “We have received solicitations, but we have shown that Mistral has a path to being a big, independent company,” says Mensch, noting his customers are “looking for a decoupling from their historic providers.”
That, after all, is Mistral’s bet. It is not looking to outspend its rivals in San Francisco. But as their power grows, Mensch’s opportunities might just grow with it.
Comments & Questions (0)
No comments yet
Be the first to comment!