Mistral AI CEO denies IPO plans, touts renewed focus on open source to best deeper-pocketed competitors

The IPO rumors began following an interview with Bloomberg at Davos in January, in which Mensch emphasized that Mistral was “not for sale” and that an IPO was “the plan.” 

“I was asked a question about our future, and what I said is that we intend to remain an independent company, [so] the natural path is to get to an IPO at some point,” he told Fortune. “Just to clarify, we’re not looking towards an IPO [right now].” 

Mistral burst on to the AI scene less than two years ago—its cofounders veterans of Google DeepMind and Meta’s AI research lab and emerging from a few months in stealth mode with $113 million in seed funding—the largest seed funding round in European history. Its first AI model, Mixtral 8x7B, released in March 2024, was widely praised for its innovative design and delivering excellent performance.

But, since then, many industry observers have questioned whether the AI foundation model company has what it takes to go the distance against deeper-pocketed competitors such as OpenAI and Anthropic, not to mention tech behemoths such as Meta, Google DeepMind and Microsoft. Although it has raised $1 billion to date—including a $640 million Series B round that valued the company at $6 billion—that amount is dwarfed by the $18 billion OpenAI has raised so far (with Softbank poised to invest another $40 billion) and the $8 billion in Anthropic’s coffers.

The cash burn needed to stay at the cutting edge of general purpose AI models—in terms of purchasing computing power and hiring and retaining top talent—and the speed with which any performance edge becomes commoditized mean that it’s unclear how even these better-funded AI startups, like OpenAI and Anthropic, will ever turn a profit, let alone smaller competitors such as Mistral. And unlike OpenAI (backed by Microsoft) or Anthropic (backed by Google & Amazon), Mistral lacks a Big Tech patron that can guarantee it long-term access to thousands of graphics processing units (GPUs), the specialized computer chips used to train and run AI models. 

The most open of the ‘open weight’ companies?

Mistral has tried to position itself as the most “open” of the companies offering what are known as “open weight” models. In contrast to the proprietary AI systems that OpenAI and Anthropic offer, in which a user can only submit a prompt to an AI model through an application programming interface (or API), and receive an output, open weight companies allow users to freely download the core “brain” of the AI model itself, as well as the code for running the model.
A partnership with Microsoft in 2024 spurred criticism that Mistral had abandoned this open source ethos as the company released several closed proprietary models and made them available on Microsoft’s Azure cloud service.

But Mensch insists Mistral remains fully committed to open source—while still generating revenue through its premium proprietary models, an AI infrastructure platform called “Le Platforme,” and pro subscriptions to “Le Chat,’” the company’s AI assistant.

In fact, a new billboard on Highway 101 in Silicon Valley slyly notes that most of Mistral’s models are “actually” open. That’s a dig at Meta, which is perhaps the best known company to offer free, open weight models, but which Mensch notes carry more restrictive licensing terms than Mistral’s Apache 2.0 license. (Almost none of these open weight model companies, Mistral included, make the datasets they used to train their models public, which has drawn criticism from open source purists who insist that the models can’t be called “open source” without providing such information.)

“It’s as open as you can be,” Mensch said. “We share the weights, we share the inference, we share a lot of findings around how we built it. There’s obviously some trade secrets that we keep, because how we bring our core value to work with customers.” 

Mistral’s enterprise plan

Those business customers, including Axa, Mars, and Cisco, are the fruit of an 18-month quest to target the competitive enterprise market. Mistral’s business customers pay for “Le Plateforme,” which provides access to Mistral’s open and closed-source models, as well as to a host of AI tools and infrastructure meant to help technical teams build, customize, and deploy AI across the organization. Mensch told Fortune that Mistral’s revenue has increased 25x over the pat year—although he declined to specify either the current sales figure or the base from which Mistral had experienced such rapid sales growth.

As its revenue has grown, so has the company’s headcount and geographic footprint. From just a few dozen employees, predominantly in Paris, a year ago, Mistral now employs 200, including 60 researchers, with offices in Paris, London, San Francisco and a newly-opened outpost in Singapore. Mensch admitted that he’s had to grow into his CEO role as the company has expanded. “For four or five months, I was still coding and doing science,” he said. “Now, I’m mostly focused on sales and product.”

A Trump bump

Many argue Mistral is benefitting from not just the capabilities of its models, but also from geopolitical tailwinds. European countries, and France in particular, are increasingly talking about the need for “sovereign AI” that would enable them to escape dependency on U.S. or Chinese AI systems. While this sentiment was present in 2023 when Mistral launched, it has grown more stark this year in light of the combative approach the Trump administration has taken towards European tech regulation and the ratcheting up of tensions between the U.S. and China. An Economist article last week said that “In the fast-growing world of artificial intelligence (AI), Mistral, a French startup, may be a beneficiary of the transatlantic tempest.”

Mistral has always been a favorite in France, where French President Emmanuel Macron has frequently spotlighted the Paris-based startup as a paragon of French ingenuity and proof that the country can nurture fast-growing tech startups that are competitive with those being launched in Silicon Valley. Like many European politicians, Macron views AI—both as a growth industry itself and in terms of the productivity AI can bring to other sectors of the economy—as a potential cure for years of moribund economic activity. It doesn’t hurt either that Cedric O, a Macron confidante and former secretary of state for the digital economy, is now a “cofounder” of Mistral and advisor to the startup.

But now the advantage of being seen as a “homegrown hero” may apply to Mistral’s commercial position in Europe as a whole.

“European companies are looking to partner more closely with European technology,” Mensch said. “They want an AI partner that can drive transformation, independent of geopolitical tensions. Our regional presence gives us an edge that others simply don’t have.”

Mensch said he has seen a “tremendous increase” in Mistral’s commercial traction in Europe over the last two months, though he claimed not to know whether it was specifically related to Trump’s inauguration. Mensch has been outspoken about the importance of Europe remaining competitive in AI: At the Mobile World Congress in Barcelona earlier this month, he said “It feels like the conversation around AI is in the US and China, and Europe sometimes gets left out of that conversation.”

Mensch told Fortune that he was not sure that those in the US fully understood how much Europeans were starting to become more conscious of the need to push back and assert itself. “If Europe is mistreated, Europe is reacting,” he said, adding that “there’s definitely some pretty strong momentum around uniting, around technology, around automation, on AI.” 

Ambition and a scrappy culture—but does it have the capital?

For now, Mensch said he is focused on continuing to take Mistral from scrappy startup to major AI player, in an era when experts question whether any AI model startup can keep up with the top leaders in the space: OpenAI, Anthropic, Google and Meta. While Mistral remains far smaller than those three, Mensch pointed to the best practices Mistral’s three original cofounders brought from their roles at Google DeepMind and Meta – shaping a mindset focused on shipping fast and maintaining rigorous scientific standards.

“We have created our own culture, which is low-ego and scrappy,” he said. While Mistral likely cannot match the sky-high compensation and multimillion equity stock grants of the foundation model giants, Mensch pointed out that the company’s open source ethos is very attractive to the AI research talent it seeks. 

“When you’re a scientist, you really want to contribute to the community, at the end of the day, you’re usually less interested in the business success of the company,” he said. “So [open source] has been a great advantage, and that’s something that we will continue to promote.” Mistral is also hiring more researchers to work on fundamental AI research that is not necessary linked to product. “There are so many things that need to be figured out, and new ways of thinking about architectures,” he said. “If you’re only thinking about the product, you don’t have the time to think about these things.” 

It is unclear how Mistral intends to fund that research. Mistral has not publicly disclosed, for example, receiving direct funding from the French government for its R&D activities, even though the French government has said it is committed to advancing AI in the country.

Mistral has secured European AI cloud platform, Fluidstack, which is building what it says will be Europe’s largest supercomputer. Mistral is slated to begin using that AI computing cluster later this year. It has also partnered with AI chip firm Cerebras, whose hardware allows Mistral’s AI assistant, Le Chat, to respond at unprecedented speeds. 

For now, Mensch says that he and the rest of the team are keeping their eyes on the prize: “We started very ambitiously, but we need to continue being very ambitious.”