Is Mistral worth $6bn?

0:02Hello and welcome to the Sifted podcast, the show where we help you get to know the brightest and boldest people and companies in Europe's startup ecosystem. I'm Amy, Sifted's editor and your host. And today I'm joined by my colleague Daphne Le Prince-Ronguet, Sifted's French tech reporter, to get under the bonnet of a company that took 2023 by storm, raising an€105 million seed round without even a pitch deck, just a seven-page memo. That startup is, of course, Mistral, the Parisian AI unicorn which began building large language models, or LLMs to rival OpenAI and is now moving into all parts of the AI value chain.

0:40This year it announced that it will be building its own data centre in France and also plans to launch a compute platform backed by NVIDIA. It's also reported to be looking at raising as much as $1 billion in funding this year. That's in part because it does of course have a mountain of competition from the tech giants like OpenAI, Anthropic and Google in the US and plenty of other upstarts over in Europe. So what's next for Mr. Al? What might be coming for the company soon? And can it ever achieve all its dreams? Daphne, welcome to the pod. Hello. How is it in boiling hot Paris today? It is boiling hot, nearly 40 degrees in Paris today.

1:22Very nice. So let's wind back the clock to a cooler time. In 2023, when Mr. Al launched, why was everyone quite so excited about this company? It was around the same time, actually, in June 2023. It was a time of great excitement, for sure. I feel like there's definitely a before and after Mistral in French tech. And funny enough, it's also when I started working for Sifted. So I started getting familiar with the ecosystem when everyone was at peak excitement levels. And I kind of got me thinking, is this the norm? It's not the norm. There's many reasons that people were very excited about Mistral, I think.

1:58And the first reason is definitely the team. So Mistral was launched by three co-founders, Arthur Mench, Timothée Lacroix and Guillaume L 'Empre, former deep mind and meta scientists, high profile researchers. And they decided to launch what was a very ambitious company in Paris, France, rather than in Silicon Valley, USA. And I think that in itself was reason enough for a lot of investors to get really excited because until then they had seen quite often top French talent going to the US to launch this type of company. and what Mr. R was saying was that they were going to do it from Europe. The second reason people were excited was obviously the proposal.

2:36As I said, Mr. R was proposing to build a very ambitious company. I was reading over their pitch deck again, and it's quite straightforward. Pitch Deck quite directly mentions OpenAI by name and says that they are going to be competing against OpenAI and they're going to be building this company again from Europe. So a very ambitious proposal. And I think lastly, the reason people were excited was the method that they were proposing. So firstly, they were talking about building this European champion despite some of Europe's constraints like funding or lower compute power and fewer resources when it comes to building AI model.

3:11And they were saying that they were going to build models that were more efficient, that required less compute power, that were less costly. So again, kind of going back to this European champion theme, they were saying it was possible to do it in Europe with their team and only with their team. They were also proposing to build an open source model, which was relatively new. Meta was already building an open source. But Mistral was really insisting on this, again, being aligned with European values, enabling to be more protective of privacy, protective of data. So again, proposing a different approach and essentially saying, we're going to compete against the biggest companies out there.

3:48We're going to do it from Europe and we're going to do it in a way that boosts European values. And I think that really got people very excited about what they could do. And how do we think they're doing two years on? What's the assessment against that initial goal that you've just laid out there? I think in terms of the products that they've delivered, Mistral have delivered on at least what's in their pitch deck from 2023. They have delivered high-performing AI models that are efficient, that require fewer resources, and that are open source. Going briefly through what they're offering now, They have both open source models and premier models, which are AI models that you pay for.

4:29They target largely enterprises and enterprises have a whole scale of how they can access these models going from using pay as you go API on the public cloud, which enables the model to be a bit less tailored and a bit less protective of privacy, all the way to deploying a model on their own infrastructure, accessing the source code of the model, tailoring it to exactly their use case and having full control over where the data is. I think they've definitely delivered on that promise. Today, they have pretty high profile customers, and they keep building and they've been praised many times for their ability to execute.

5:02They've gone even further now, as you mentioned, they've announced that they were going to launch a compute platform and become a cloud company, essentially, in partnership with NVIDIA. So that's a whole new paradigm for them to be building in. We'll see where that goes. It definitely still ties in with those sovereignty goals that they were pitching at the start of the journey. The big question now is it's only been two years. Where will they be in a year? Where will they be in five years? Are they going to be able to deliver? Their revenues are reported to be around€100 million for 2025, or at least on track to reach€100 million for 2025.

5:38That compares to$10 billion for OpenAI for 2025. So it's obviously still a small company compared to very, very big competitors, and there's still a lot to build. what do you think's behind the move into data centers into the compute platform into kind of this full stack ai which artemans has spoken about on stage do you think that's a real push for them to be generating much more revenue or is it a kind of i guess it's just such a crazy fast race in ai isn't it at the moment and do you think it's coming from more we need to be doing everything because otherwise everyone else is going to come along and people will start using them and they'll forget about mistral there's been mixed reactions i think to that announcement it certainly is very timely ai infrastructure is the hot topic of the moment it started at the ai action summit in paris in february where french president emmanuel macron announced 109 billion euros worth of ai investments And that was obviously in response to the Trump administration announcing$500 billion worth of investments with the Stargate project.

6:46And then Mr. L arrives with this announcement in partnership with NVIDIA. It's obviously very timely and it's what everyone wants to hear. What I've been hearing is obviously to look at the other side of the argument as well, is that it wasn't necessarily the right battle to pick. It's a very ambitious battle. It's a big bet that they're making. They're up against very big US-based tech giants. And what some people have been saying is that instead of trying to replicate what these tech giants are doing, maybe a smarter move would have been to be one step ahead. So, for example, one thing that's been put forward by some of the people I was speaking to was build custom-made chips for AI or work on developing custom-made chips for AI, which would be optimized for their models, would enable them to be even more competitive on costs and resources, and ultimately not rely on NVIDIA chips, which in Mistral Compute's proposal, the baseline is still that we are going to need NVIDIA chips, and NVIDIA is a US-based tech giant.

7:38So it kind of points to what some people are saying about Europe and Europe generally in the AI race and in the tech race. Europe tends to try to replicate what's going on in the US in order to be part of the race when it could or should be trying to catch the next technological innovation wave. So perhaps that was where Mistral should have focused its efforts. Ultimately, we'll just have to wait and see. Do you feel what I was thinking about as I was preparing for this episode is, it's almost like Miss Strau, the buzz is gone around Miss Strau, it feels like to me, you know, there's these other like new hot companies that everyone's excited about, like Lovable, the Swedish AI agent company, which we spoke about on the podcast a few weeks ago.

8:19So, you know, there's even like 11 Labs, which has been around for a while now, but is doing voice AI. It feels like kind of people are moving on. And more significantly, perhaps, these companies, especially the AR agent companies, are making really significant revenue really quickly. So Lovable is saying it's already making like 50 million, 60 million in ARR dollars. Whereas last year, you know, Sifted reported that Mistral made 30 million euros. And you're saying this year is on track for 100 million euros. But that to me maybe doesn't seem, is that enough for a company that's been positioned as like a champion of AI in Europe and is doing so many different things already?

8:59So there's probably a degree of it being normal that after two years, people aren't talking about Mr. Al all the time. They were talking about Mr. Al a lot for the last two years. So obviously the company is building, it's less present in terms of media activity and stuff like that. So the response I kind of get when I ask that question is just let them build and let's see what happens. In terms of the numbers, 30 million ARR, we reported that for 2024. They had started commercialising in 2024, which is, you know, it's not it's not lovable numbers, but it's still pretty impressive. And tripling that for 2025 is an encouraging sign.

9:35I think one of the issues Mr. L might be facing at the moment is that what they were pitching back in 2023, which was, and I think this is literally from their pitch deck, that the value in generative AI was going to be in the foundational models is perhaps not so true anymore in 2025. I think it's increasingly become obvious that foundational models are going to become a commodity and that the real value lies in the application of AI and the application layers, which is what other companies might be focusing on. So for Mistrial today, the challenge is to manage to unlock that value as well. And I think that's the work they're trying to do when they're bringing their models into enterprises.

10:14They're dedicating a lot of the team to customizing their models for specific use cases within enterprises and making sure that this application layer, the end use case for the customer is as smooth and straightforward and as easy to use as possible. And I think the focus has gone maybe more on this application layer and a bit less on the performance of the models, which is what it was pitching in 2023, because now it's becoming obvious that the application layer is where the value is. And another one of the challenges that businesses building foundational AI models typically face is the difficulty of making the financial equation work.

10:49So obviously building foundational models is very expensive. And I think a lot of the companies are still working out how to make significant revenue from it. It's actually one of the reasons that another company that was building LLMs in Europe, Adafalpha in Germany, eventually pivoted away from building foundational models because, as they told Sifted's reporter, Tim Smith at Sifted Summit, that the financial equation was too hard to make work. Interesting. And talk us through some of the clients that it has. It has some pretty big names, doesn't it? It's previously talked about having Orange, the telecoms giant, AXA, the insurer, Total Energies, the big energy company, SNCF, the French rail company.

11:32What's going on there? And are these kind of like pilot projects or are these sort of serious clients who, you know, investors and analysts think Mr. Al can kind of grow with and provide, you know, a lot of AI services to? So because of the nature of what Mr. R is proposing, so the open source approach and the control that you can have over your data and the customised, tailor-made use cases, it's promising. It all ties in with the sovereignty angle again. And the bottom line is that it's very adapted to use cases for companies in fields like banking, healthcare, defence, which have sensitive data.

12:11And that's where Mr. R has been focusing. until a year ago I'd say we were mostly looking at pilots and now I've seen a few headlines about these companies carrying that pilot into production and you know going into the next step in terms of expanding the technology company-wide there's been some high-profile clients as you've mentioned BNP is another one of them there's been recently an announcement that Mistral signed a 100 million multi-year contracts with the shipping and logistics company CMA CGM, which actually saw Mistral send a team to Marseille where CMA CGM is headquartered to work directly with the company on building out more use cases.

12:52A trend that we've been seeing here at Sifted that's interesting as well is that it seems Mistral is increasingly getting close to governments and focusing on defence use cases. So it's got contracts with the French government. It's talking to the UK as well. It's got a partnership with the Singaporean Defence Ministry as well. So it's obviously a field that makes a lot of sense for Mistral. It's got a lot of sensitive data, requires total control over the data. So that's definitely one to watch. And I think one that might become a key use case for Mistral going forward. And I guess, obviously, again, something we've spoken about a lot on the podcast recently you know defense and the kind of geopolitical turmoil is something on everyone's minds and we're seeing a huge push from european policy makers investors also unicorn founders to chat about tech sovereignty and the importance of building sort of tech for europe in europe and there was an interesting partnership wasn't there earlier this year mistral signed a strategic partnership with helsing the german defense tech giant which we have also spoken about on the podcast recently.

13:59Do you think that whole shift is basically just good news for Mr. Al? I think it is good news for Mr. Al. As I said, it makes a lot of sense. The issue with signing defence contracts, as with any company that's signing defence contracts with governments, is how much is going to lock that company in, in terms of being able to not necessarily expand, but is it going to close certain opportunities in other countries because your signing basically deals with European governments. That might be a question for Mistral going forward, especially because Mistral is very keen to expand to other countries, particularly the US.

14:36So again, we'll have to see how it plays out. But generally, I think it makes a lot of sense for them to be locking in these contracts. Who would you say Mistral's competitors are? Or who does Mistral think its competitors are? I think if we go back to the original competitor, the number one competitor that Mr. R was talking about in its pitch deck, its open AI, obviously. You could talk about all the other big tech players in the field who are developing foundational models, working in generative AI, increasing the AI agents. So Google is one of them who's advancing pretty quickly in the field.

15:12Meta, although arguably is facing various different challenges of its own. But now that it's going into compute, you could argue that it's also facing competition from US cloud giants like AWS or Google Cloud or Microsoft Azure. So those are big companies as well. Mistral is obviously offering something a little bit different, but it's worth keeping in mind that it's up against some pretty big and deep-pocketed companies across the Atlantic. And it's not only cloud companies based in the US, big tech giants based in the US that are building cloud services and providing cloud services today. There's also a number of startups in Europe, including companies like Nebius, who are providing these services as well.

15:50So the competition exists in their home ground too. Do you think the pressure's getting to Arta Mensch, the CEO? I feel like he's still on stage quite a lot, isn't he? He was on stage at the AI summit held by an investor, First Minute Capital in London, with a big bank that he's clearly about to do a partnership with. I feel like he's still popping up quite a lot, isn't he? He is, yeah. And I can only imagine it's been a big learning curve for him. he is still out there and he's getting asked a lot of pressing questions about how Mr. Al can compete what it's going to do next is it worth its valuation Mr.

16:26Al's valuation currently is nearly six billion euros its revenues are as reported potentially 100 million euros for 2025 so it's quite big expectations to live up to and I think he must be under under pressure he does look quite cool in interviews so I'm sure he's handling it very well also worth noting he's well advised one of the more discreet co-founder on the company is Jean-Charles Sémion, the founder of French Insurtech Allen, which is a huge company in France. And it's well known that he's been quite instrumental in terms of advising and pointing Mr. L's founding team in the right direction.

17:03So I think that has also been quite a critical piece of the puzzle for Mench and his co-founders as well. My favourite story of a founder going through a crazy time was this time last year, Artem Mensch was coming to Founders Forum in the UK I think just after having been maybe at Viva Tech in Paris or maybe he was going to Viva Tech afterwards anyway he managed to lose his passport on the way to the Eurostar and I think it was also a few weeks after he'd had his first baby and I was just like this guy is really leading a a hectic life at the moment yeah he did yeah he became a father quite early on let's not forget it all started at Sifted Summit where Artem Mensch gave us a fireside discussion so yeah amazing he's been everywhere and the companies while he's very prominent the company has a bit of a kind of elusive comms strategy doesn't it do you think it's is it genius or is it just silly how quirky their announcements are so they've got better at it i think inevitably it's like 200 more employees now so i'm pretty sure they've got a comms team and marketing team that's working on their slick logos and communication strategy.

18:11But early on, it's true, most of their announcement would just be cryptic links on Twitter. And by the way, their Twitter photo is like just Mr. AI written in what looks like word art. And that's not changed, that's still the case. But at the time, they would announce their models like that, just drop a random link that only developers would be able to understand. I'm not a developer myself, I did not understand it. And, you know, it spoke to developers, not to the mainstream. It's changed a lot now. You can see that they are increasingly going into the mainstream. And actually, they have a consumer facing application as well that's called Le Chat.

18:46It's a wordplay on chat, which in French, chat means cat. Le Chat is a consumer facing application. So I think they have to obviously have a PR strategy that aligns with that as well. And talking of that, I'm going to ask Luchat a question in a second. My apologies to everyone for any French pronunciations I have made during this podcast. So there are rumours swirling that it's considering raising a billion dollars more this year. What have you heard? And I'm going to ask Luchat what it said. Yeah, let's see what Luchat says. I mean, to be fair, Mr. R announced Mr. R Compute last month. But before that, at the AI Action Summit in February, it had announced that it would be building a data center.

19:31At that time, Mr. R only said it would be for its own purposes. And then with Mr. R Compute, it became obvious that this is a whole new part of the business they're launching where other companies will be able to access the compute power they're serving. But my point is that back in February, they already said that they were going to build AI infrastructure and that it would cost billions. And I think they were quite elusive on how they would get that money, but it was quite obvious they'd need to raise more money. It's emerged recently that they would potentially be looking at raising a billion euros by the end of the year.

20:01It would probably align with what they're trying to do. It's expensive to build data centers. They're in partnership with a number of other investors, including MGX and BPI, which is the French public bank. but they will need more money so it only makes sense for them to be raising again pretty soon if they want Mr. Al Compute to come about but let's see what Rasha has to say about it it says as of my last update Mr. Al AI has raised significant funding but there is no confirmed information about them raising one billion in a single round and then it tells me about some of their past rounds so what else should we ask it other big questions that people are asking of is mistral going to ipo it says no confirmed information how about is mistral going to be acquired that's a big one and then it says mistral is not going to be acquired instead it plans to go public through an initial public offering so you can't get your answers yeah so so arthur told bloomberg davos earlier this year i quote we left big u.s tech to create a company in europe to show that Europe has something to say that's still the project and so the independence we have is something that we value dearly and then when he was prompted by a journalist to ask if Mr.

21:14Al would IPO he said of course that's the plan so that's apparently where they're heading. What other big challenges finally Daphne do you think might lie ahead for Mr. Al? I think the top challenge will be to prove that it is worth its valuation. I think that's the main thing that keeps coming up and it's the main reason people are speculating that it might get acquired. If you take the viewpoint that you don't think that they'll be able to scale into a company that's as big as what they had in mind and that their valuation is inflated, it only makes sense to, again, speculate that their investors will be wanting to get out as soon as possible at the highest valuation possible and that they'd be pushing for an acquisition.

21:57That is entirely rumours that I'm hearing on the market. It's nothing that I've heard from Mr. Al themselves, but I think they're going to have to show that their revenues can increasingly get nearer to lower multiples in terms of their valuation. That will be a big challenge, I think. Mr. Al Compute will be one to watch. As I said, it's a big bet. It's going to cost a lot of money. Will it deliver? Will they manage to convince European companies and enterprises around the world that they should be using Mr. Al Compute and not the easy option, which is inevitably based in the US. And ultimately, I think that's how someone I was speaking to put it to me quite recently.

22:35If you go back to the foundational models issue and the fact that it's going to become a commodity, there's only going to be a few winners there. Is it ultimately just down to how well they market themselves and make sure that people pick their solution over others? Perhaps that's just it, but we'll just have to see who emerges as a winner in that race. and lastly a challenge that Mistral is likely to increasingly face in the future is the fight for talent which has never been more intense we're hearing of reports that companies like Meta are offering up to 100 million dollars in signing bonuses to poach top talent from open AI it's only likely to become even more intense and that's something that Mistral is going to have to manage as well watch this space and you can follow everything that Daphne writes about Mistral and other French tech companies.

23:24Where can people find you, Daphne? On sifted.com. Sifted.eu, sorry. We should keep that in or we can scratch it. We'll decide. Which are some other companies on your radar, Daphne? Quite a few, actually, here in France. There's obviously a lot going on in AI. I've been following H company very closely recently. H is another company that raised a monstrous seed round in 2024, early 2024. they raised 220 million dollars in equity and debt and since then they've had a few challenges including quite a few of the co-founders leaving i'm a very big follower of quanto who i think was on the pod recently as well he seems to be doing very well depending on when this is published i can say that they've just applied for a banking license which is something that everyone was expecting for a long time.

24:17So it will be interesting to see how that goes. And lastly, another company I've been following for a long time is Insect, which farms and transforms insects into protein ingredients for animals and for humans. It launched a long time ago and it raised a lot of money. And because of the nature of what it was building, which was a huge industrial plant, it faced a lot of challenges in scaling and ultimately had to file for insolvency a few months back. I think it's a good example of some of the challenges that industrial players are facing in France and in Europe more generally so one to watch not just insects but also the repercussions for the wider industry.

24:53Amazing thank you so much Daphne for sharing all of that insight into Mistral. Everyone listening please let us know what you thought of this episode if there's something you'd like to throw in about Mistral please get in touch I'm amy at sifted.eu Daphne is daphne at sifted.eu we really love hearing from our readers and our listeners and we especially love getting tips from you guys as well so get in touch with those and as always please rate and review the podcast to help us find new listeners and spread the love and this podcast was produced by tim smith thank you