E399 | Oliver Holle (Speedinvest) & Ekaterina Almasque (OpenOcean): How to identify the next AI investment sweet spot

0:00Welcome back everyone to another episode of the European PC podcast. Today, we are talking to Oliver Holler from Speedinvest and Ekaterina Almask from OpenOcean. You will very quickly, if you know a little bit about the players in venture, know that these are two very different players. Oliver, of course, is coming from Speedinvest with a 500 million euro fund, investing broadly across the ecosystem with a generalist fund. But you have OpenOcean who are on the other side of the spectrum. I think they're 150 million, if I'm not remembering incorrectly. and they're specialized in AI. So today we're going to dive all deep on AI and we're going to do that based on these two perspectives, trying to see exactly how does a big player that are generalists think about AI and how does a specialist like Ekaterina think about AI.

0:47So I hope you'll enjoy this conversation as much as I did making it.

0:54Tear down this wall. It's more than just an ally. This is a union of values. Let's start acting. This show is not investment advice, and the hosts of this episode may be invested in the funds and companies featured. All right. Welcome back to the European VC podcast. Today, I have Oliver and Katja with us. Maybe just two seconds. Tell us, how do you relate to AI being, of course, Speedinvest and OpenOcean? Oliver, maybe you go first. so i oversee six investment teams within speed invest all focusing on different verticals one of them being focused on deep tech and the infrastructure layer so to say every single team has done multiple ai investments it's a topic that you cannot avoid no matter if you're a consumer if you're in fintech if you're in health and of course if you're in sas so it's a it's a massive theme that will not go away and we're all going to form our opinions around that so it's super exciting to talk about this yeah it's obviously a horizontal or is it katya it is horizontal and it's a lot of things but um for me so i'm looking at deep tech investments and fundamental technologies investments uh across europe at locomotion as a general partner of locomotion but also i'm an investor in ai since 2006 and i actually graduated in ai computer science in the 90s.

2:20So it's a long story. And I have seen all the up and downs of development. It was very interesting that AI became usable back in 2006. And this journey of high investment volumes in AI started in 2006 and accelerated recently. So now we are looking actually at the second wave of AI investments. And maybe we can talk today even about third and fourth waves. coming. Beautiful. Four waves. And we're only on the second now. Okay, so that's going to be intense. So let me ask you the very first big question. Some split AI across the four different layers and think about it as such. And then they say, well, we only operate in the top two layers, the applications and so on.

3:12Or how do you think about it? How do you kind of split up AI in terms of where you play and where you choose not to play and where you think the biggest potential is. Could you talk a bit to that? In our perspective, we try to think along also our teams and then along our organization principles to make life easier. So there is obviously a huge opportunity around, let's say, AI core infrastructure. That means, of course, foundational models, and then a lot of adjacent or complementary questions that need to be solved. These fundamental additional questions can be around safety, can be around transparency, around actually performance questions, edge questions.

3:57So there's a lot of, so to say, topics on top of the foundational models. We as Speedinvest have so far not invested in the foundational models. Why? Because we are afraid of the hardware and infrastructure component there and ability as a relatively small seat focused VC to not get completely washed out over time. I think there's a high risk there. And so I think if you're lucky and you're getting to the pre-seed round of these companies, then it's fun to do it. But anything else doesn't make too much sense if you don't have the capital behind you to defend your position. That being said, there is a lot of opportunity in these adjacent and complementary questions that need to be solved.

4:40There's so much research going on as universities around those. And that's where we have seen multiple extremely exciting cases to invest in. Actually, with MindsDB, we have a great investment in the US together with Katharina that would sit for me in that category. Yes, and then there's all those vertical applications of AI. And again, we have five other teams and all of them are super excited about temperatures that they see there. And there, for me, AI is simply the next infrastructure to build upon, and there is just no way to not do it, not invest in it. Where do you play on the four tech stack layers, so to say, Katja?

5:26Well, actually, almost everywhere, but with a big butt, right? So it's true that some of the plays in the AI are so capital intensive. is almost like building a nuclear power station. It requires so much capital that for a small VC firm, even for a growth VC firm, it's very hard to tap in. We are talking here about an open AI type of place, right, where it's billions and billions invested. And probably if you were very early on, you could somehow enter this game, but not these days anymore. What is important for us is to find those opportunities where we think it's fundamentally changing, like it's disrupting in a certain way, but we also can generate enough revenue short-term and long-term.

6:15Yes, we can wait for some time, but there should be this kind of predictable growth. And usually where we see the largest gaps today is actually in the whole enterprise stack. If you're an enterprise today sitting and trying to apply AI, actually you are in a very difficult situation because you don't have tools even how to deal with your data, efficiently how to connect this data to AI tooling, how to control what the AI is doing, you know, all the compliance, all these things. There is very little tooling done. And even in talks with Google and other big tech players, they confirm the same issue.

6:53So we are talking, we are facing here enterprises that are willing to take AI on, but they don't have the infrastructure for this. And this is, we think, it's a very sweet spot. Because this is where we can invest, create real difference, generate real value. And those companies are most prone to start scaling very fast. So this is kind of like one category. But the other category as well, there are some applications, let's say, that are super useful for enterprises where it's the first AI and very disruptive AI models and they scale super fast because they create a very easy way for enterprises to create something.

7:31And I'm talking about Gen AI in this case. So Gen AI is a big thing. Gen AI, by the way, is the second wave, right? So the first wave we talk about, the first wave was just introducing deep learning and now we have Gen AI. And in Gen AI, we have companies like HeyGen or Colossian in Europe that actually allow you to generate videos very easily. And suddenly you can do marketing at much less cost, much faster. And they scale very fast. Higem is probably one of the fastest scaling companies in the AI space. I think they scale to 35 million within a year or something like that. So it's an interesting space where you say, wow, this is really disruptive.

8:09The team is doing something great and actually is very useful for the enterprise. I think there's an interesting question that I'm asking myself actually on the enterprise space because you spoke about this now. You have, of course, also the examples of Klarna that recently, what the CEO recently said, they're basically pushing out a lot of their SaaS application layer and are actually just working with AI and building their own applications based on that. So I think there's a huge business that is being made at the moment with a lot of system integrators and a lot of solution providers that sit on top or that work with existing APIs like OpenAI and others, Mistral, etc.

8:48and then just do service work, solution work, and thereby solve the needs of those big, big enterprises. So maybe the focus is more on the, I wouldn't say SME only, but the giants probably will build a lot internally based on the core infrastructure. There's a sweet spot there that I think we all have to think carefully to not get eaten up sooner or later by a solution, but a combination of the big saloon providers that make a huge killing at the moment with the password of ai plus the open eyes of the world yeah that's absolutely right i think it's it's it's also to be fair a lot of companies like even you take okay dose of the world right and all building up some ai solutions the question is we have seen this before in other industries at some point it's just more efficient to switch to some platforms that provide it and again like it's an emerging market that's why we all here are guessing a lot and and we are only half time probably right so so but we have all hypnosis right we all have hypothesis 50 would be a lot exactly how do you both think about because now the next big thing is of course agents it's one thing that we've had something that could give us good answers that could help us dig through data But the next big thing seems to be, can you actually make AI act on your behalf and act unsupervised almost, at least on its own, so that you might supervise, but you're not controlling it?

10:26Do you think that that is where we're headed? Or do you think that we're moving a bit too quick in terms of going from the first stage? We haven't really solved it yet. And now we're already talking about letting agents lose in the world. Is that pie in the sky fingers moving a bit too quick because they're keeping the hype train moving quick? Or are we really, is that where you're seeing the next big thing? And if so, didn't many seed investors make investments just a year ago or two that are already becoming redundant or not relevant anymore because the next thing is really agents? So I think you're hitting on a very interesting topic because I think agents is the third wave, actually.

11:13So if I, like, in my simplistic view of the world, forgive me for that, please, but in my simplistic view of the world, agents are coming and it will be an amazing disruptor for many, many automation plays that we have seen in the last 10, 15 years. because the last 10, 15 years was all about automation, simple automation of workflows in the enterprise. And now suddenly we have agents that can actually mimic action of some of their roles and functions in their organization. And you can even create super systems of agents acting on behalf and doing even payments to each other. You can think about all of this.

11:53And again, I think this is a great wave that is coming. And you're right, there are some investments already, But again, what is missing yet is largely the infrastructure layer. So what is coming again, that there is a huge opportunity to invest in orchestration, in management, in compliance, in making sure that there are no risks involved for human beings and so on. So this is all like AI safety and also is coming as a big wave. and actually there are companies that start doing this, but I think they will become really, really large when this agent's economy starts to really evolve. Yeah, I would agree.

12:33I think overall in the whole AI space, it's such an obvious case of where a lot of people today want to invest in yesterday's term sheets. So you always have the lag, right? Everybody now wants to be part of a foundational model company, but the reality is if you haven't invested two years or three years ago, maybe five years ago into these companies, it's over. You have to invest in a new wave. So there is this risk of investors just being late to the game and thereby really losing out. I think that's very clear. With regards to agents, I'm personally a bit less worried because for me this is ultimately extremely verticalized and you need to be extremely close to the customer and to the actual problem that you're solving.

13:19and that will be ultimately for the best companies out there that build a vertical solution on the basis of AI, that core competence, that understanding of the customer, that understanding of the workflow will be the difference and they will always, if they're good, they will always use the latest tech to optimize for that. So I would hope if I look at our best companies there that they will be just as fast as the new startup wave out there to implement agents or implement the next wave of tech. Again, as Katharina mentioned, we have a couple of infrastructure topics to be solved still, a couple of fundamental questions to be resolved.

14:01I agree with this. So there will be an adoption wave, but I want to come back on this one foundation model and just bring maybe as a point or discussion point. It was an interesting conversation I had with Greg at NEA, who was the CTO of some microsystems. And he was like, well, you know, the VCs are all pouring money into foundation models. It's all great, but what people oversee, there is just a gate for more content. And actually, you know, it made me think, and he was comparing this to Explorer or, you know, like this Netscape versus Explorer war at the late 90s where everyone was pouring money into development of, you know, how to access internet, but at the end of the day, this is not where the most value today is generated.

14:52So it's an interesting kind of thinking as well. We need to implement certain things, and yes, we need, I mean, it's great that Microsoft is baking open AI and things like that, but in reality, where we, for us, for venture capital, also enterprises, where the most value might come from, is not from these fundamental layers, but something else that will come on top of it. For me, if I'm honest about it, that's my fundamental belief. At least for us as European early stage investors, that's where we probably have to play. I think everything else is unrealistic. And there are such amazing opportunities.

15:30I mean, speaking of content, the case where I've seen AI at work with extremely immediate monetary impact is a content company. Actually, it's called Inkit. It's a U.S. company. They were now originally out of Germany. They've been using AI to dramatically improve their workflows, dramatically improve their revenue base by just using different formats, audio books, multiple languages, multiple new formats, using, of course, also image and video creation based on the same core human content. And that was wonderful to see. And it's already out there. You can do it with existing APIs. You don't need to wait for agents or anything.

16:13And then their revenues exploded just by implementing these kind of existing solutions that are already out there. So that very much comes to your point of the content layer probably being a huge opportunity here. My reflection on what you said, Katja, is yes, you can definitely liken it to the war for the Explorer. But at the same time, we should not forget that in the end, a lot of the revenue for Google, as an example, comes from their domination with the Chrome browser and everything that plugs in to basically their search business. And as such, you know, I don't think it's a matter of there not being created value.

16:52I think it's just a matter of the value already having been captured because the winners have been picked and it's going to be very, very hard for anyone else to come in. And what then that means is that there's a lot of value also being created on top of explorers, obviously, and the internet, which is then what the rest of us are now searching for, so to say. I wanted to ask you on this search for the value, then on top of the foundational models, because you spoke a lot about enterprises. And I saw Mark Benioff of Salesforce speak at Dreamforce a bit more than a few days ago. And what he said very clearly was that so they are working on something I think he called it agent force or something like that.

17:38And what he described was very much that. AI is there, you know, that's where they play today. That's where he spends all his time. And he's working tirelessly with customers on building models for them where the accuracy is super, super high because he's asked his accuracy. and that makes perfect sense that this accuracy problem, meaning avoiding any form of the models making up shit and the models making the wrong inferences and thus taking the wrong actions when it's an agent. That's where he's very much trying to solve. And what I'm thinking is that a lot of VCs are saying that the existing SAS models and SAS giants are going to be disrupted.

18:25But what he says is that, well, we are the only ones with the infrastructure layer with our clients where we can make sure that the agents that are developed are actually functional because the accuracy is higher enough to be able to go out and act in the world. He said he probably focused more on himself than the others, but by inference, you could realize that what he said was, well, startups are going to have huge accuracy problems for a long, long time because it's very, very hard to just plug into an existing business and take whatever they have and then have very high accuracy. How are you seeing your startup solve that?

19:04How are you thinking about that when you're looking at deals in this space? Because it does feel to me that putting something out in the world, acting, we're a far step away from being able to do that. Yeah, I'll just jump with a few thoughts here. So Salesforce, by the way, is a very interesting case because actually they're building a very interesting vertical cloud, basically, you know, comparable to many other tech players. So it's becoming really, in my view, and I might read it wrong, but in my view, Salesforce is becoming a tech player by itself, very similar to, you know, Google of the world, probably smaller scale.

19:47but the strategy is very much, and I admire the strategy, by the way, because I think there is indeed value in being able to control this tech. Coming back to your question, I think indeed what's the main problem with the foundation models today is that it's a black box in a way. Indeed, enterprises have a hard time controlling input-output. Yes, there are certain ways of doing this, but it forces enterprises to think about their proprietary model. and proprietary models force them to think about their proprietary data and other sources of data and also how to actually do this with the lack of ai talent so if you are like a general you know if you're a general enterprise and you just need to build you know top-notch cutting edge ai stack how do you do this and i just give you an example it's from the past but in the past bmw was launching their first car with software inside.

20:43Okay, it was 2005. And, you know, like you can think BMW is the best, one of the best companies in Germany. What's the problem? But one month before launch, their car wasn't starting. Okay, so they came with this problem to the CTO of Siemens and said, you guys have this corporate technology with top-notch engineers from all around Germany. Can you solve our problem? And the target scene was created. to solve this problem. And it was solved with the top-notch engineers, but the lesson from this story is that even BMW couldn't get the best software engineers to write the great code in the first place.

21:23With AI, this problem actually is worse, okay, because AI is much more complex, and many, even top-notch companies, they just cannot hire these top-notch engineers. So the question is where we are investing, we say, well, you know, our hypothesis is that not every enterprise, even the likes of Klarna, would be able to hire these very, very best AI engineers, but they have to solve very, very top-notch cutting-edge AI problems. So how do they do this? They will go to AI startups that manage to get some best engineers and work on particular, very specific problems, but solve them the best in the world.

22:05And it can be any problem. It can be, for example, a problem, you know, can I trust my model output? Yes, there is a startup or several startups looking at verifying that the output of the model, of the performing of the model, is what the enterprise was expecting and nothing happened in between. And there is accurate output. And if it's not accurate, it tries to detect what happened. And example of the startups, for example, Lettuce Flow in Switzerland, it's our portfolio company. There are a few similar in the US. There is one that is like Kara also that has gained a lot of investment, but they moved into security more.

22:42But, you know, there are startups trying to solve exactly this problem. And I think for each problem, there will be startups. And some of the startups will become platforms all the time because they will be so successful in solving one problem, then the next problem, then the next problem, that with their AI aggregated engineering talent, they will be able to dominate certain verticals or horizontals. You know, you take it. So this is our hypothesis when we invest. And I would be very, very interested to hear from Oliver how they see this space. I fundamentally agree with what you said. I think there is a clear picture that incumbents will benefit and will actually have a lot of value.

23:23A lot of value would go to them, right? And this is different to the browser wars because back then we all started, everybody started almost from zero, right? Maybe with Microsoft is a great example of not starting from zero. But you will have, given the infrastructure investments that you, given the massive synergies that you have on data accuracy that you, Andrea has mentioned, there will, it will be very, very hard to beat the giants that are out there today. I think that's very obvious already. And that means also to be very discerning and careful in picking startups. At the same time, just as you said, I think there are only going to be so many planners of the world which have the talent, have the resources, and have the technical capabilities to solve that for themselves, and that will leave hundreds and thousands of customers out there that will not be able to do it.

24:15And I think that's where very focused, vertically, startups with a clear value proposition to solve a specific problem can make a huge difference. obviously for us for example a big question that we ask is this a mission critical question where ai is left alone to solve then we will be very careful that can happen in the finance industry that can happen in education that can happen also of course in the medical sector so that's that's an area and super careful to to invest in there are a lot of other use cases let's let's stay in the content side entertainment side where a hiccup doesn't make a difference at the end of And then there's the other use case where AI is not left alone, but is always in very close dialogue and control of humans.

25:02We see a lot of this happening in the medtech and biotech space where AI is used, but it's always used very much within a certain workflow element of the whole value chain. where again, the risk of leaving these agents alone is not really there, at least the way they use AI these days. So these are the two key areas. Is it mission critical? To what degree is AI embodied in the value chain? These two questions help a lot in addressing the question you mentioned, Andreas. Now let me tee up some debate. Katja, you're an AI specialist and your fund is, that's all you do. You've argued that AI presents a new business and scaling model that will require the VCs have to change.

25:54So my question to you is, can you tell us a bit about why you think that AI presents a challenge that makes it paramount or necessary for VCs to change how they think and operate? and then also maybe make the case for a vertical specialized player in this and then I'll ask Oliver afterwards to respond to that and say, how do they do it inside Speed Invest? Perfect, Nandes. The only thing I will say, I wouldn't make a case for vertical specialized VC on AI specifically, but I would argue more about general deep tech approach because I think this is what we are talking about here. So, indeed, AI represents a major kind of challenge for typical VC, let's say VC investment approach that worked in Europe for the last maybe, I don't know, 15, 20 years, especially when B2B SaaS became the sweet spot of investment in Europe.

26:56So B2B SaaS, I would argue, and again, please tell me if I'm wrong, B2B SaaS, I would argue, is all about pattern recognition. Basically, you get a certain kind of early stage company, you recognize there are certain revenues, you think, well, they actually do much better than other B2B SaaS, and we predict that it will grow a certain amount. Yes, we look at the market, but the market usually is there and it's only a matter of time and putting resources into marketing and sales to scale this company. Maybe simplified. Again, I'm simplifying a lot today, so please don't judge me for this. In the AI space, it's an emerging market.

27:38It's what NVIDIA's founder, Jensen, was calling as a zero billion market opportunity. Yes, we know already AI is there, but the market, to be honest, is not there yet. So we are creating the market. What does it mean? It means that customers don't really know yet how to use AI. And today is much better than it was in 2006, but still there is a lot of education involved. So when AI company comes early stage, they usually don't really have a product market fit at all. They don't have revenues. they are lucky if they have almost consulting projects where they are paid, which is called POC, proof of concept.

28:18And then for majority of VCs, this represents a major problem because you cannot invest in POC, you cannot invest in consulting, and who knows whether they will ever find product market fit, so whether they will be able. And this requires a very different thinking because it's not about pattern recognition, it's about belief. and this is where actually people who were in the industry for a long time and ideally have built things by themselves in the past, they say, well, you know, I don't really know because I'm wrong 50 % of the time in the best case scenario, but I sense there is something happening and I don't know yet how it will evolve and what the end solution will be.

29:00But I think it's a really interesting opportunity and most probably like the team who's working on it is really kind of, they have credibility that they can try doing this. And I can give you a few examples. It's not from the AI space, but think about Google itself, right? When Google started, and I remember that because I was standing with Sergey Brin in 2001 on the stage when he was telling that the future of searches, the future of the world is all connected data, but they didn't really have revenue model back then. You know, for me, it was a surprise that some of, I think they raised 25, just raised 25 million or were about to raise 25 million, but it was all on belief that they will be able to do something with the search.

29:40And they were not inventors of search itself because everyone back then was talking about content and drive and web. You know, that was like, there were European sponsored projects on this. No, but they did something and, you know, raise money and then started to think, okay, we can monetize and monetize in a different way, completely different, not by selling search, right? And many companies in that space, in AI or deep tech in general, they start somewhere, but then transform in something else. Think about NVIDIA, gaming, and suddenly opportunity comes. Nobody could foresee. People can say, well, I know there is a bottleneck in compute, but nobody really knows how to evolve.