As Mistral acquires Emmi, Europe’s AI race moves into physics and industry

0:00Guillaume Decugis:Hello and welcome to Path Founders with me, Mike Butcher, where we like to unpack the code, the capital and the consequences behind Europe's tech startup ecosystem. Today on PathFounders, we're joined by Guillaume Dekujis, who's a general partner of Serena Data Ventures and an investor in MEAI, the Linz Austria-based deep tech company, using AI to make industrial engineering simulations run in seconds instead of days. Now, the reason why we're chatting today is because in April last year in 2025, EMI raised a 15 million euro seed round, reportedly the largest seed round for an Austrian startup at the time.

0:47Guillaume Decugis:And it was tackling one of the least glamorous but most important bottlenecks in industry, which is how physical systems are designed, tested and trusted. Now, Guillaume, you were an investor in this company, and the news out this week is that they've been acquired. Shall we unpack the story? Yeah, happy to. Let's get it out of the way. First, they've been acquired by Mistral, the Paris-based AI company, one of the very few in Europe. And do you want to reveal how much they bought it for? This trial is not disclosing the transaction. It's a significant one, but it will honor their wish to remain confidential in this.

1:37Guillaume Decugis:Right. Well, obviously, it's a little bit of a roll up. Would you characterize it as an acquirer? No, definitely not. It's true that you've seen some transactions where people are buying a team for minimal return to investors. um you know the the signal i would share is that you know we invested just a year ago uh the team is doing fantastic there's been a lot of validation we got from the the bet we met a year ago uh and quite frankly uh there's a couple of reasons we we took the deal so early and the economics uh one of them i mean as vc investor as vcc investors we typically don't invest for just a year we invest for five, seven, eight, ten years.

2:23And yeah, the economics were just fantastic. But also the industrial synergies between Mistral, which is at the core, a horizontal player, and EMI, which is a vertical foundational model, are fantastic. And, you know, we're very happy to be part of this and to roll over and move forward with them. Right.

2:47Guillaume Decugis:Well, I'd love to talk to you a little bit about your fund in a second, but let's just chat a little bit about Emi. And for listeners who are not engineers, what was the sort of big problem that it was solving? And how do you feel that it will fit inside Mistral's ecosystem? Yeah, so what I think we've all seen what, you know, AI could do to language with large language models. Our conviction as investors was that AI was also going to be applied to other formats, and physics is just one of them. And when you train AI to kind of learn the laws of physics and be able to infer from those models how physics would behave, you can do a lot of different things.

3:34For instance, you can simulate how the wind would evolve, the air would evolve around an airplane wing. you can simulate the fluids dynamics around a formula one car or the heat dissipation in a semiconductor those are all things that you cannot do with lms because the language is not the human language but it's the language of physics that's what uh emmy is doing they're building what we what they've called large engineering models which are applying ai to the physics world, which has multiple applications. And if you think about how we design airplanes, how we optimize Formula One cars, how we do crash tests in automobile, or how do we design semiconductor chips, it's all down to simulation.

4:23It's all down to understanding the laws of physics. Now, this is a field that's been existing for 50 years, but that hasn't been disrupted. and it's one where there's been a few attempts, early attempts to tackle it with AI that I think were very specialized and didn't generalize well. And so it's a bit of like, you know, when we moved from machine learning to chat GPT, we had this universality that we've all been, I think, the witness of. It's a bit of the same thing which is happening now with the type of models that Amy has been building. Right.

4:56Guillaume Decugis:I see. Why do you think investors care about AI for physical engineering as opposed to software and chatbots? Yeah. So if you think about industrial companies, obviously, they can use AI like any company to optimize their legal processes, business processes or, you know, do software faster. But the core of their business is R &D. And R &D is about designing objects in the physical world. And when it takes 48 to 72 hours to simulate, again, an airplane wing or the results of a car crash, then you have to do things in a certain way. So if you're a design engineer, you're going to drive, you're going to design something, you're going to give it to the simulation guys, you're going to wait two to three days to have the results, and then you're going to iterate that design.

5:50Now, imagine if the results of your simulation were real time, you can completely transform and change to where you're going to design things. And so we think that's the kind of technology that could change the timeframe to build an airplane from maybe 10 years to three. So this is very promising technology. And at the core, all industrial players want to apply AI to their core business, which is designing and, you know, building new products.

6:19Guillaume Decugis:OK, well, in terms of EME, it's raised 15 million euro last year and now it's effectively been acquired. So it's quite a quick journey. Do you feel that they could have, in your opinion, in discussions with them and obviously your investors, in your discussions, did you come to the conclusion that it was time to take this path? Did you not feel that the team and the company, did the founding team not think they could go further, for instance? It's a really good point. And there was a lot of board level discussions on this, as you can imagine. And the initial plan was to actually do a Series A.

7:02The company had raised enough funding, had runway for another more than two years. So we weren't in a rush. But what happened is that we validated a number of things faster than the initial plan, which is good news. So what we realized, I mean, this company has been very good at driving the scientific part with the business part together. So when we invested at Seed, they transformed a POC into a seven-digit contract, and then they signed additional customers. It's still small from a revenue standpoint. Those are good validations that those industrial players, which you know are moving very slow, could accelerate their decision-making process because of the type of technology and outcome that ME was producing.

7:46So based on this, we had decided to go for a Series A, and that was the initial plan. What happened is that even though we had some offers for a series, we also had non-solicited offers from several players for acquisition. And we started to think about, you know, what it meant. And, you know, as much as we felt that the team could continue, can pursue on their own, we felt that this is going to be, you know, the type of player where it's going to be a winner takes all. And there was some really compelling, you know, elements in the, in the Mistral story. One of them was that you have the scientific advantage.

8:25Those guys have been thinking about this problem for ages, for years. They have a strong scientific advantage. But you have to feed that with data. And you have to feed that with proprietary data that comes from your customers. The type of data that you have only when you work with the top airplane manufacturers or the top semiconductor manufacturers. And we felt that, you know, there could be even further acceleration by combining the scientific advantage with the kind of go-to-market acceleration we can have with a larger player like Mistral.

9:00Guillaume Decugis:Right. I see. So, yeah, it's a sort of it's a kind of a vision thing, I guess. They wanted to take the vision further in this direction, especially with the on the research side, it sounds like. Do you feel in your. Yeah. Yeah, it's interesting because just to comment on this, you have some very ambitious founders. And of course, they could have, you know, the initial plan was to do this on their own. But I think the name of the game for them was to do, you know, some, you know, be the leader of that space and kind of, you know, become the way, find a way to dominate the space faster. And so that played a key role.

9:40And the economics, as I said, were the second reason we were interested in that.

9:46Guillaume Decugis:In your opinion, do you feel that this is the start of a strategy that we'll see more from Mistral acquiring these smaller companies to increase its offering overall? Well, you know, I can't comment on Mistral strategy, but I can say a couple of things. You know, we're in a specific position where we, this is our second exit to Mistral. We actually were also a co-lead investor in Koyab, which was their first acquisition. Great company that's going to enable them to bring a lot of applications to their cloud. So they're definitely on acquisition track. Again, won't comment on their strategy, but I think to me that's interesting to see that it's not the first time we're seeing people in the LLM space try to be more specialized.

10:43If you look at what Entropy has been doing over the last few years, they've been specializing in agentic coding. And I think it's been a really great response to LLMs becoming a commodity, and all of those LLMs are the same. So I think we've all been, I think, very impressed by what ChatGPT did a few years ago. And then there's been this question of, are all LLMs equally good or is it a commodity? And now we're seeing this verticalization, this use case facilitation becoming something interesting. So we'll see. But I think it's an interesting move.

11:22Guillaume Decugis:Yes, well, it is very interesting and it's a quick exit for them. I dare say you're probably, I presume you're happy as investors? We're very happy. This is going to augur well for the trajectory of our fund. Again, the fund is – we did our last closing in 2023, so it's a still pretty young fund. We're still in our investment period, and this is our second exit. So, yeah, it's a good outcome. It's your only way. Tell us a little bit more about Serena Data Ventures. Yeah, so I think we're part of this kind of new generation of VC funds in Europe that have a specialization strategy. And by that, I mean that, you know, historically as, you know, someone who comes from the founder side, I've been a VC for just two years and I've been spending half of my time as a founder in the US.

12:13You know, I think I've seen the European VC scene as like, you know, historically, it's been a lot of, you know, fragmented and local funds. By that, I mean, you know, generalist VCs that invest locally in Paris and Berlin and London. And of course, some of them now become big and have access to all of Europe, but most are still investing locally. And when we invested in ME, a lot of my VC friends were telling me, like, hey, how did you guys connect with this company? We have a different approach. We are specialists. We're not generalists. We invest only in what we call infrastructure software, so this foundational layer of software that enables a lot of different tech.

12:55And by being specialists, I think it gives us an ability to identify pattern and to connect in a more meaningful way with founders all across Europe. We also are all partners who have a background as founders. We've all been scaling companies in the US. So I think this gives us a specific angle that's helped us win some very competitive deal like it was the case for Amy a year ago. Right. I see.

13:20Guillaume Decugis:Well, thanks very much for talking to Path Founders. Guillaume Dacujis, General Partner of Serena Data Ventures. And thanks for unpacking for us your perspective on this acquisition of MEAI by Mistral.