Exclusive: Mistral’s first investor on what their €3B raise really means

0:04It's another interview with PathFounders with me, Mike Butcher and remember to join the PathFounders Network, which is launching imminently. Go to pathfounders.com slash network if you want to join. Europe has spent the last few years asking whether it can produce any AI company capable of competing with the giants of the US and China. Mistral may be its best shot. The French AI company has just raised€3 billion in a Series D round, valuing it at more than€21 billion. The largest equity fundraising ever completed by a European technology company. But this round is more than about money. Mistral is evolving its strategy, it would appear, still building frontier AI models, but also investing heavily in compute, infrastructure and sovereign AI, positioning itself as an alternative for governments and companies that want more control over their technology and their data, and perhaps crucially, not with the US.

1:08So is Mistral still trying to become Europe's open AI? Or is it building something very different and potentially more defensible? To unpack the round, the valuation of what it means for the future of European AI, I'm joined by Paul Murphy, an early investor in Mistral when he was at Lightspeed, and definitely the guy to comment on this latest move. Thanks very much for joining, Paul. You bet. Well, you were actually, just give us a quick background because you were actually one of the first people ever to recognize the potential at Mistral with Artur. Tell us a little bit about that backstory just quickly.

1:50Yeah, so, I mean, just as context, I was the first investor on the board of the company since inception. I recently left Lightspeed and therefore had to step off the board. So anything I say is not from an internal perspective, it's more market perspective. But yeah, the Lightspeed team and I met Arthur in 2023. We knew from their backgrounds that they were a special team. And when we kind of dropped everything and went over to Paris to meet them, And we knew within a couple hours that we wanted to figure out how to work together. So we did that. I think we did the whole thing in two days and kind of got started and been on this crazy journey with them since.

2:36Well, that's amazing. I'm sure you're going to have a few stories to tell about that. Let's get into the actual news right now. So three billion euros at 21 billion euro valuation, pretty extraordinary by European rounds. what's um what what are your thoughts your initial thoughts right now about um where they're heading with this you know your overall kind of zoom out view well i mean i think the reality is there's one story which you alluded to you know that's happening in the u.s around anthropic and open ai um which you know are incredible businesses and incredible technology but there's there's another story that I think is arguably more important, which is what's happening in China and specifically around open weight models.

3:29Some people refer to them as open source models. And that's, I think, actually the area that Mistral is uniquely positioned to be the kind of Western counterpart to what's predominantly been taken over by the Chinese. so does that mean they're drawing away from building up frontier models or does it mean they're playing a different kind of ball game i mean the um arthur is sort of like to say publicly that what he's trying to do is convert electrons into intelligence and solutions and i think that's a really good way to think about what ai when it comes to what ai can actually do for an enterprise as an example.

4:12The general purpose models are kind of only of so much value. Really, what you need to do is build solutions on top of that. It turns out that to build most of the solutions that create value for an enterprise, you actually don't need to be at the frontier. And so I think there's been a little bit of a, you know, there's been this sort of race towards, you know, being number one on the benchmarks and the leaderboards. But if you talk to, you know, CTOs and CIOs at large enterprises, they don't really care about that. They care about how their business is being transformed. They care about hallucinations.

4:47They care about latency. They care about token pricing. So there's a whole lot of other factors. And I think most importantly, they need something that actually solves a problem for them in their business. So I think that's the kind of frontier race is never something that was, in my view, new kind of front and center for Mistral. Their view was we have to be, you know, close to the frontier. In some cases, we'll be at the frontier. But it was never a metric that we talked about, you know, are we always the bleeding edge model? That was kind of never really the point. It is interesting, isn't it?

5:23Because it does strike me as being a slightly more Chinese approach to AI, in the sense that China doesn't seem to be particularly interested in super intelligence and artificial general intelligence and, you know, creating a, you know, a new super being more like, can we make this robot arm more efficient? Or can we make an enterprise more efficient? Would you agree? I don't, I don't think I agree. I think that there's very much a set of people in the Chinese AI ecosystem that are racing just as hard and just as fast as the US frontier labs. I think people give them a hard time as, oh, they're just distilled models and there's no real innovation happening.

6:07And I just don't believe that that's the case. I think there's incredible innovation happening in China. At the same time, it's a big ecosystem and there's many providers. And certainly some of the providers are focused more on here and now kind of practical solutions. And some of those models like Quen are getting just phenomenal traction in the kind of public sphere amongst developers today. Well, it is also interesting that there's a kind of pivot in this running around towards infrastructure, building out data centers, and sovereign AI as well. What's your comment on that aspect of it? Yeah, I mean, models are only of so much use.

6:49You ultimately have to run the models. Obviously, most of the frontier models have an approach which lets you host the models or run the models on their platforms. There's a subset of folks within the public sector or even large enterprise that want a bit more control. They want to run that on either private cloud or on their own kind of own devices or on-prem. So I think that's actually historically played really well for Mistral because they had the ability to not only could you run their models, but actually could run them efficiently. One of the lesser known critiques of a lot of the Chinese models, despite them being open weight, is they're actually really complicated to run.

7:30They require enormous amounts of RAM. They're not super efficient. And so I think Mistral's always had a unique edge there in the open weight community. I think the bit around compute is interesting because I think we all underestimated how much compute was going to be needed. Because even though we knew this technology was incredible, I think we sort of underestimated how many enterprises and consumers wanted to adopt the technology. So there's sort of, you know, a very well known race towards building up compute capabilities, not just for training, but also for inference. So it's, I think Arthur and the initial team saw that early on, and they thought, you know, not only do we just need to have the raw compute, both to serve our customers and to train our models, but also from a sovereignty perspective, it turns out there's a real reason to have some of that compute or a large portion of that compute in Europe.

8:28And so that's been a unique focus for them. And I think, you know, everyone's going to win with the more compute they can build in Europe. It is interesting on that point, Samsung leading the round also kind of feels kind of important. I noted that we had Macron, President Macron, talking about France and South Korea leading into this strategic aspect of it. Do you, what's your thoughts about Samsung? What does Samsung bring beyond money? Yeah, I mean, you know, I think there's, it's a very kind of interconnected supply chain when you look at the kind of AI compute stack. So obviously, Mistral already has ASML on the cap table.

9:15But you look at, you know, everything from the memory providers, the chip manufacturers, the fabs, it's all it's all pretty closely connected. And then of course, NVIDIA is, you know, is sort of ubiquitous. they're kind of in most or all of the relevant companies. So I think in this case, what I took away from Samsung investing was most people that were close to Mistral understood how incredible the business was becoming and what they were achieving. What was hard is that sort of, you know, that didn't sort of necessarily make the same headlines that some of the U.S. model providers would make, or even some of the Chinese model providers would make.

9:58So there was a fair amount of education that I think had to be made, both within the corporate side as customers, ironically, that was easier because they were more focused on practical solutions. But I think with the investor community, it just required more work, just more education. And so what I really like about this round is the fact that you have a non-European, non-French capital provider coming in. In this case, they're strategic as well. But this is much more than a European story. This is clearly Samsung saying Mistral is going to be an incredibly strategic company globally for AI. And they're locking arms in a pretty unique way.

10:35I think, you know, I don't know exactly what they have planned, but I suspect there's going to be partnerships on a whole host of fronts with Samsung over time. There is a certain irony that Mistral is going to be hosting third-party models, including Chinese open models. And we have Samsung as an investor now. So it's quite sort of a geopolitical hodgepodge, isn't it? Yeah. I mean, I think, again, like Mistral's view is always that they just want to be focused on solutions. And so if there was a better open weight model, there wasn't religion within the company about using their own models. It's just their models were the best at serving this, you know, and achieving this certain kind of benchmarks for tasks that needed to be achieved.

11:21But I think they're actually quite, you know, going back to why they wanted it to be open weight to begin with, is they really believe that over time, pre-trained models are going to become more and more commoditized. And that the kind of quote unquote, innovation is going to happen, you know, post training, fine tuning, reasoning, or, you know, probably over the next couple of years, it's going to be specific to the solutions built on top of that infrastructure. And it's going to build one gigawatt of European compute capacity by 2030, which is quite a capital intensive business. How do investors think about that kind of build out?

12:04Yeah, so I mean, there's a vibrant debt market to finance a big chunk of that, if not all of that build out. I think they talked publicly about hitting 200 megawatts by the end of next year, which they appear to be on track to do. They've got a couple of sites up already that they're using for training. So, yeah, I mean, I think the capital markets will support that. I don't think, in other words, capital will be a limiting factor. I think the limiting factor will be energy sites and then to some extent how quickly you can build out those sites. um we saw recently that uh matt clifford from um entrepreneur first has joined anthropic to be their european managing director i think also rishi sunak is involved at that an anthropic we've also got uh other british politicians have involved in open ai um and open ai anthropic google all the hyperscalers, they're all going to be offering products in countries in Europe and around the world.

13:13And obviously talking a sovereignty game, for instance, such as European data residency and such like. How do you think companies like Mistral or anyone, frankly, will be able to talk about or to talk up sovereignty when you've got this big pressure, big competition from these big US giants like OpenAI and Anthropic? Yeah, well, I mean, first off, I think that's an incredible hire for Anthropic. I think Matt's a really special human. Not only does he understand how things work in Europe, but he's also deeply, deeply technical. So that's a big win for everyone, I think. you know, the reason that that is being kind of is a new is a focus area for the US labs, in part is because there's recognition that it's really important for in particular European companies.

14:11And so I think it's, to me, I look at that and I say, well, that's, you know, that's validation of missional strategy really from the outset. And I think it's also really good for those US labs to be thinking more about AI sovereignty. I think that's in everyone's interest. I don't think you want to have any kind of company in any one country developing such vital dependencies on technology, which is truly black box and only kind of lives within the confines of one country. So I think it's sort of good for everyone, including Mistral. I don't, you know, they definitely compete with U.S. labs, but there's more customers out there looking for solutions.

15:02And again, I would say, if you say, what are the kind of the bottlenecks? It's not necessarily or bottlenecks towards growth for Mistral. It's not competition. It's just, you know, moving fast enough, getting enough people to build solutions for customers. So I think overall, it's probably oddly going to be a win for Europe and for Mistral, just having the U.S. labs here. Fair comment. Finally, what do you think is the sort of often misunderstood story about Mistral? Especially, I think I find that the U.S. press often kind of misunderstands the company a lot and that they're often comparing apples with oranges.

15:45what's your sort of takeaway from this big round and how that might be changing the kind of overall trajectory of the Mistral story? Yeah, I mean, I think the US press in particular and the kind of Twittersphere and others, it's a really convenient talk track to sort of downplay what Europe can do and what a French company can do. It was comical when we made our first investment announcement. There was almost like disgust amongst folks in the U.S. that a tier one U.S. fund could write such a meaningful check into a French company. What were they thinking? And I think that just time and time again, actually, it was like three or four months after that, that missile released their 7 billion parameter model.

16:42and everyone was like, holy shit. Okay, these guys are serious. They know what they're doing. And if you look at even DeepSeek's first release, a lot of that was built on the back of the mixture of experts' work that was pioneered at Mistral. So I think it's, they've probably never been as good as they should be on kind of landing their communications and kind of the marketing story globally. It's done well in the US. They just hired an amazing CMO, Brian Hall, who I'm convinced is going to help improve that. But the more people, like I said at the beginning, the more people dig into the company, the better they understand how special this business is.

17:21And where they are relative to the other labs is all by design. It's not because they can't do more. If they had wanted to raise another 10 billion to run compute and try to be at the bleeding edge of frontier, the capability is there. It's just that's not kind of the business strategy. So, yes, I'm incredibly excited about the company. I can't wait to see what they do. Unfortunately, I'll be somewhat on the outside looking in, but it's exciting nonetheless. Well, thank you very much for giving us a little bit of an insider insight into into the Mistral story. It's really interesting, isn't it?

18:00And certainly as a journalist, I'm certainly fascinated by the moves that they make. But thank you so much for joining us. Paul Murphy there, formerly of Lightspeed, now doing something else. I'm sure we'll hear about it in the near future. But thanks very much for joining Path Founders, Paul. And remember, join Path Founders Network. Go to pathfounders.com slash network where we're assembling Avengers Assemble. We're assembling the best and brightest of European founders, investors and ecosystem players. Join there.

18:40We'll see you next time.