Talking Tech Sovereignty with an AI Optimist

Cristina Caffarra speaks with Yariv Adan, founder of ellipsis Ventures and former Google AI leader, about why he believes AI will expand human potential by making intelligence widely accessible and accelerating scientific and economic progress. They also discuss his AI-native venture capital firm and debate Europe’s pursuit of AI sovereignty, weighing the risks of dependence on U.S. frontier models against the need to build competitive European capabilities.

ESCAPE FORWARD Ep. 11, 7 July 2026

“Talking Tech Sovereignty with an AI Optimist” with Yariv Adan

Founder and General Partner, ellipsis Ventures

Cristina Caffarra

Hello everyone, I’m Cristina Caffarra and this is Escape Forward, the space where I have the privilege of moving forward from what was my original domain, antitrust and competition, and have conversations with people I like a lot, I find fun, interesting, and I learn from. The discussion today will go again into AI and digital sovereignty –  as many know, this is something I’m very involved in.

Major concerns around availability of frontier models have prompted many European fans of these models to suddenly wake up to this ghastly prospect that somehow we may be deprived of these amazing toys for a host of reasons, from capacity constraints to security to leverage. But they are seen as irreplaceable and this has created a lot of reactions. In fact, two major reactions in my view.

One side has been pretty much a strong, very vocal plea to politicians, policy makers, that we simply cannot do without these models. We need in Europe to do absolutely anything in our power to secure access to these frontier LLM top lab models, because all sorts of horrible things will happen if we don’t. And the suggestion is of a trade in which we essentially secure access to these models in exchange for some sort of inducement, incentive to hyperscalers that host these models to build, for example, giant amounts of datacenters across Europe, give them energy cheaply, give them permits, and so on.

And then another reaction has been a pretty aggressive, a backlash against the sovereignty movement that I’m involved in, with an argument that in the sovereignty movement we don’t have a real plan for how Europe is going to become an AI superpower. And as such, we are essentially not very useful, we are feeble, we are useless, and so on.

I think both of these reactions are wide off the mark. I call the call for doing deals in exchange for model access as “AI maximalism”. I have discussed this with a number of people, Luis Garicano and others, who do genuinely believe that we should be doing these deals. My position is that they entrench and worsen the position of dependency that Europe finds itself in. And I do not see this trade with the devil as having any positive upside for Europe. I want to pursue some of these themes today in various ways from various angles. In particular with someone who is not really steeped in the sovereignty conversation, but is very knowledgeable about the AI world. Indeed, including running a business which is premised on AI as the objective, but also using it in the business.

Yariv Adan, you’re welcome. I met you in Davos and I was very struck by the discussion you led about your vision of an AI world, which is very different from many that I have been familiar with. We all know and like Gary Markus and Ed Zitron and Brian Merchant and their view of the world, but you have much more what I would call AI optimistic view, a very positive vision, which I thought was interesting and different, and really painted a picture of AI as something which is not necessarily just to be feared, to be concerned about, the bubble bursting concern, but something that will be integral part of our lives in a way that will improve them and to some extent relieve us from centuries of toil and thankless labor. On your background, you were originally doing startups in Israel, then for 17 years you’ve been with Google, building AI, as Director of Product Management, and Google Assistant and Google Lens were some of the products you led on. So a very extensive experience at Google, which will surprise some people given my history with Google. Google even sued Texas over me, as you know. But this is another part of Google, it had nothing to do with you…  And now you founded a venture in in Zurich, ellipsis ventures, which is investing in early stage AI. But interestingly, and I want you to discuss it, works with agentic AI in a big in a big way.

So over to you. Just give us the positive vision before you go and talk about how you use AI in your business, and then we’ll talk about sovereignty.

Yariv Adan

Yeah, thank you for inviting me. Aalways love hearing you and having a conversation with you is a privilege. So thank you so much. And these are fascinating topics. Yes, so I actually started with giving talks to people about AI in my business, and I got the question so many times, but what about humanity? So I took a step back and really thought what do I think about AI and humanity? And I think as you pointed out, I’m very optimistic. In early stage investments you need to be an optimist in the sense that it’s very easy to see all the way something can fail, but you need to find the ways that it can succeed and kind of follow that path and make that thing happen. And in a way, I think that’s a little bit of my DNA. Like you said, I’m a builder. I was building stuff at Google and in startups in order to do that well, you need to be an optimist.  And it’s always easy to say, it won’t work. The hard way is to see how it will work.

And the reason I’m an optimist about AI is there are three main areas that I see with optimism. Other people are actually seeing a risk. First, think what AI does, and many are worried that suddenly this intelligence is being controlled by a few giants and so forth. But the flip side of that is actually that the cost and availability of very advanced intelligence suddenly becomes a commodity, a utility that anyone can access, and especially anyone can access from their phone. Luckily, AI is coming after the internet and after the mobile revolutions. That means that everyone across the world today has access to phone and to internet. So anything that AI can do, at least in terms of digital intelligence, people have access to that. Today, intelligence costs a lot. If you want to go to the top doctors, it’s a scarce resource, very few people have access to it and very few people can afford it. And this is like medical doctors or really good shrinks and business consultants and lawyers, anything that actually requires expertise today is limited. And if you suddenly open up that to the world, I think we’re living in a much better world. So I think like first the commoditization and the fact that you can deliver intelligence across the world at low cost is great news for humanity.

And I think it’s even more than that – I was recently invited to a discussion on “is democracy ready for AI?! And everyone had in the back of their minds, social network, fake news, blah, blah, blah. I think that humanity doesn’t need the technology to generate fake news and horrible religious, political and other theories that leads to genocide. We’ve managed to do it thousands of years before electricity.  But when I look at what are the real problems of democracy, not the phenomena of the problems, what is the core? It’s inequality, inequality in access to basic services. I think it’s in education. I think it’s in the fact that people are enslaving themselves. And I think it’s in the fact that governing organizations are both corrupt and inefficient. And I think that AI actually has the potential to touch on all of these, to close the gaps on basic services, like I said before, just by offering it at a much lower cost, much higher availability, lower scarcity. I think education is one of the things that can be completely revolutionized. One of the biggest problems in the world starts from that, that we have one teacher for 30 kids. So the first thing you teach the kids is to shut up, hold their arms, sit on the chair so we can actually manage the class. So you kill any curiosity or free will or spirit when they are three already. If you can actually shift resources and use AI over there and recreate an educational system that actually works and can work everywhere, that’s progress. And I think that if we take the capabilities of AI to monitor the governments, to see what they are doing. Imagine you put a microphone out in every meeting. every memo that is being written, anything that the government do and it highlights, this is not what you committed to, this is against this and that, and measure them and publish it, can we maybe even delegate some of the administration to AI? I think it’s much more transparent and powerful.  

So I think democracy is not ready for AI because instead of governments starting thinking, how do we revolutionize our education system, our health system, our governance and administration, instead we are trying to regulate stuff that I think there is no chance and it doesn’t add value. So again here, the things that I’m talking about have a lot to do with the stability of our society and making it a better place.

And then I will tie to something else a lot of people are concerned about, AI is going to take jobs and what do we people do.  And I’m not trying to minimize some of the risks and the importance of working on them, but I’m saying there is also a flip side. I argue that we have been stealing jobs from AI rather than AI stealing jobs from humans in the sense that I think humans are doing work that they shouldn’t be doing. And it’s always hard to do it when you are born into a situation. You assume that whatever is happening here and now is the way that it should be. And any fluctuation from that is so scary and dangerous. But if you move a little bit like, you know, 40 years ago, there was no startup, there was no email, we were completely disconnected. The pace was slow, right? It was a very different workplace. If you look a little bit in other continents, it’s very different.

I always like to go 150 years ago and think about people working in what we currently think of horrible factories, every day, all day, all their lives. And imagine someone coming, the guy from Plato’s cave, and telling them, wait, wait, I have an invention for you, I saw the light, and the light literally looks like light, it’s electricity and mechanization. And then everyone would say, no, no, no, what would we do with all of these people whose purpose in f life is sitting in this factory producing? They are contributing to the society in this way. And you say, no, but actually, why do you need to start working from the moment you are born and until you die? I imagine a world where you study until you’re 20 and somehow we figure out how when you’re 65 you stop working and the economy takes care of you. Why do you need to work seven days a week? I imagine that you can work five days a week and maybe sunrise to sunset, and then you know only eight hours and you take 30 days off every year and then people will say, but what will they do? And they say, oh, they will have hobbies and they will have a life and families will spend time together and eat breakfast and dinner together. That would have sounded crazy.

And I think in the same way, I imagine the world, why do we need to start working at 20? Why not at 40? Why do we need 40 hours? Why not 15? Why 30 days? And if people really need to go to an office and pretend they are working, fine. We can always, like kids, go and we can make pretend. People go to a pretend office and pretend they have a mission in life. Although many of the people I know in the offices are not happy. I don’t underestimate, you know, that suddenly if you don’t have a job we need to solve for that. But I’m saying it’s a better place actually if we work less and shift these resources into education and other places where humans are needed.

One last point on my optimism. As I said I start from we’re going to have intelligence as a utility, which will contribute to the stability of government, better governing. And then more time for a better world.

But the other thing, and it’s very clear now already, is our apish brains have peaked. We have actually reached a point where we are struggling and getting very slow and ineffective in solving some of the most complex problems. And we’ve seen, for example, we had an epidemic. It took us quite a while to actually find a drug that could help us. And it’s unclear until today how efficient that was.  And of course, we are dying and we’re getting old. Climate keeps surprising us and people are dying. We didn’t solve energy. We are kind of stuck in space. We visited the moon a long time ago and it takes a long time to get back to it. We are very slow in deploying robotics yet we need them. So a lot of these very complex problems are very hard for our brain. AI is not exactly like our brain but has some advantages, I think can break the glass ceiling on where we are on semi chaotic and very complex problems. And we have some poster child examples of unlocking a bunch of stuff, like in drug design. And there are so many companies working on a material for better energy and so forth. So I think suddenly humanity can solve problems that were impossible and actually are critical for making this a better place.

And I think it’s so funny. I think a lot of people have the wrong reaction to AI because it’s artificial intelligence. But it’s like as artificial as the nest that the bird builds is artificial. Like it’s part of us. It’s an extension of us. We should think about AI as it’s the same atoms, and it’s an extension of humanity, and it’s just humanity becoming smarter in a way. Sure, there are still risks in fire and in pencils. People have done horrible things using pencils, but we need to learn how to use that.

CC:  So your vision is this expands the realm of possibilities, right? What it allows is not limited to doing routine jobs in a way that is less demanding, so we don’t need the mass of workers doing summaries of the literature or summaries of what’s existing, or doing some boring processes. But you are really thinking of a much broader set of opportunities.

YA: Yeah, you know until now, we haven’t been designing drugs. We’ve been discovering drugs, right? There was no engineering in drug building. We’re like dying from diseases because we don’t understand how our body works and we cannot create proteins that actually operate as we like. We aren’t tapping into space, right? We are killing the resources of a single planet in an infinite universe.  The utilization of energy is super low. We don’t understand what the weather will be in more than two weeks. We cannot anticipate weird stuff in the weather that is like today. We are so basic in certain ways that I think in ten years we will look back and we will just not believe in it. We died much faster than we want. We’re suffering from so many diseases. We do not control the geopolitics. We do not understand the economy and what controls it.

We’re victims of so many things because we don’t understand them. Yes, there is so much headroom here that is low hanging fruit. And I don’t understand why everyone is worried about that. We will automate some pencil and paper work. I don’t think that’s interesting.

CC: But I must ask you this though. There is so much discussion around the downsides and the fear narrative around it and the concern about humanity losing its nature. So you don’t see any of that? Including, at a more prosaic level, concerns about this is a bubble, this is it will explode and it will kind of take us into a recession.  How do you position yourself relative to these other narratives?

YA:   I think there are bubbles within this. Already two, three years ago, I called it “commoditized magic”. And that’s a very confusing state to be in because usually we were used until now to commodity and magic being opposite. You know, magic is scarce, maybe doesn’t even exist. Extremely valuable, costs a lot.  And you’re lucky if you can use it. on the other hand, commodity is something that anyone can do it and you compete on price. And suddenly these models are definitely magical in the sense that things that even experts, not 10 years ago, but two, three years ago, definitely five years ago, would say, oh, completely impossible will not happen, suddenly you can do it. And this is very valuable. On the other hand, it’s completely commoditized in the sense that you can have it for $20 or $200 a month. Right? and everyone can access it via the phone, right? So we have this thing that is fast moving, very useful, is clearly intelligent. And again, doesn’t need to be perfect, there’s a lot of issues with it, but it’s definitely much more efficient compared to the alternative. That that’s a very confusing state of being. And it’s very disruptive. And whenever something is disruptive, and whenever especially when people with a lot of power are invested in the past, it creates a lot of cognitive dissonance, right?

So do I think there is a bubble? I think that at the moment, there is a huge opportunity to automate inefficient human processes in a way that there is a huge margin to be taken. So you actually deploy in legal, engineering, whatever it is, and you’re replacing a human process with an AI, you are actually saving millions very, very quickly. However, and then people are used to translate, that is a recurring revenue. But I argue no, that’s not recurring revenue. Once you automate the process, if it is based on a commoditized technology, the baseline for running it is the cost of actually running that automation process.  If you have a company that did it and now is making money because it was an early mover, but there is nothing sticky there and there’s nothing defensible, eventually margins will go down.

So if you are assessing them based on previous laws, then you’re in a bubble. I think that in the past, if a smart entrepreneur identified the problem that is valuable to a large set of customers and they managed to move quickly and build a solution and get first to market, that was a recipe for a very successful startup or a unicorn. But these days, when they are built on a non-defensible technology, and when there are hundreds or thousands of other teams that can do the same thing with very little resources and expertise and moving fast, that’s not the case anymore. And if you are making investments based on the old rules, then you are creating a bubble. So I think there is a bubble in that sense that there are certain companies that I don’t think are reflecting their future values or even they have a right to exist as a large player in five years from now – and there will be a correction around that. This is the commoditized side of the commoditized magic. But I think there are other companies that are solving problems that are extremely large in the trillions, like in the drug industry, in the energy industry, stuff where there is real scientific, technical, industrial value with real data models, I think that unlocks a lot of value. And I think in that sense, there is no bubble. And I think the value for humanity is clearly not a bubble. I think the fact that humanity will be in a much better place five years from now on the aspects that I described, that’s not a bubble. I think the economy is under threat because our entire economy assumes the opposite. Our entire economy assumes that in order to generate value, you need this very large industrial corporate that is very expensive to run, Opex and CAPEX, so you need to raise money to keep it running. And we are all invested in that market. Our pension, many of the pension are invested in that market.

And we know that actually the reality of the markets change quickly. If you look today’s market, 40 percent is seven companies that didn’t exist pretty much before the internet. Microsoft did. The other ones didn’t. So if 15 or 20 years from now, most of the power is actually with companies that don’t need that Capex and Opex, or maybe it’s centralized by five companies that control compute and energy, what happens to the economy? That freaks me out a bit. But I think that this is like 15 to 20 years ahead. So I’m like, OK, we’ll figure it out. Someone will figure it out.

CC : I want to move on from this broad landscape, and talk about how you have organized your fund.  You invest in AI startups, but what is interesting is not just the kind of targets that you focus on, but the way in which you’ve actually set up the fund, the business. Essentially with a with a bunch of agents running much of the analytics. And you and your partners essentially at the top of the pile. I think this is an unusual extent to which your business is effectively deploying agents. Tell us a little bit about that.

YA: Yeah, so we basically put our money where our mouth is. And we actually started from a place where we said, hey, we do not want to sell our fund to our investors as “we are these great smart people from Google” and we did companies and we sold them, and then have junior people doing all the work. We wanted to keep the organization small, but then of course you’re limited by how much you can do. And we strongly believe that if we build our organization as an AI native organization that is built on AI first principles, we could actually enjoy both worlds and basically have two people running an organization instead of 30, 40, 50 people. And it took us a while to build it. And I think one of the key things that we did is that we were not thinking about how are we going to automate ourselves or not how do VCs work and how do we automate them. Because we believe that if you try to automate people, you’ll get maybe a 1.5x, 2x optimization.

What we did, we took a step back and we said, if AI was running VC, how would it do it? And then you get the full power of AI. And we basically split the work and we have two problems. One problem is from all the startups that are currently happening in the world, how do we find the 10 people that we should speak with this week? So that was kind of one system. And then the second system we said, once we speak with these 10 people, how do we converge quickly to a decision yes or no? And the realization was one that it’s basically a loop of let’s look at everything we know, what are the gaps and questions that we have in order to get into a yes, no decision, close these gaps and loop until we are comfortable. And this is something that AI can do very well. And if you model something and you think in principle how AI can do it, you can leverage the fact that it operates at great scale. You can leverage the fact that it works 24-7. And we build the system. And then we realized that we have enough data to also build marketing agents on top of that, investment relationships, and operational agents. We gave a talk about it and then we started having inbound requests from other funds asking us, wow the system is amazing. Can we actually buy the system from you?

CC: Right. I am aware of that. That’s why I wanted you to to describe it.

YA:  Yeah. And then we said, OK, we don’t want to sell it to you that’s not our business. But we actually, had a bunch of students working for us. And we actually gave them the code. And they are now operationalizing it and taking our system and customizing it and making it available for other funds. Yeah. And it’s interesting because we were trying to think, OK, where is the human required in the process? Should we have a sub fund that is completely automated?

Also at first we were automating only ourselves. And then we were trying to think about, can we go beyond that? So for example, we said, what if we had the best investors in the world working for us? So we actually did a little bit of deep research. We asked, are the best 50 investors in the world in similar stages to us, in early stage deep tech AI. And then we collected all the information that there is on the web. And there is a lot of available information on the web from podcasts and from other stuff where they describe how they think about investment, how they choose teams and so forth. And we basically created a virtual investment committee that works for us. So we actually have these top investors. We created profiles that I believe are more consistent with their thinking then they’re in reality, and they have all the information that we have and then they run. So at random, we choose one person to lead the investment. He presents it to the other agents. Then they have a first discussion, then they argue and then they conclude and they give us their feedback and action item.

CC: So the agents discuss? And then they give you the output.

YA:  Yes, but these agents are replicas of real world top investors. So basically we get a second opinion from others and it’s very interesting to see because different people look at the deals in a different way. So again, I think AI opens, allows you to take any idea that you have and make it into reality.

And the other principle that we had is AI is going to be much better than the human alternative long before it’s good or great. And that’s the question we always ask ourselves. It’s not that we ask ourselves, is this perfect? Which I think in many cases slows down AI adoption. But we ask ourselves, if we didn’t use AI and we kind of use the human, what’s the value that we would get? And of course, when you look at the economics of this automation, at the end, for an early stage fund, the economics are very simple. How many deals can you see? How fast can you move? And at what quality can you do that? So these are the things that we are super rigorous in measuring and making sure that we score high on as we do this automation. And I can tell you that, we’re definitely at a very efficient level.

CC: So this is fascinating to me. But you, with your background at Google and being in charge of much AI there for years, are not your usual CEO of a company. There is at this point this enormous appetite which is exploding in boards everywhere for agentic AI. Everyone is talking about we need to have this, we need to have it tomorrow. It’s got to be a priority, it’s got to happen. And the impression from talking to boards is that they haven’t the faintest idea what it is that they’re talking about. They are trying to feel around, they have  a very approximate notion. This embracing of AI and particularly the agentic dimension of it is at the beginning, because when I when I talk to people, this is just now on the board’s agenda in the last three months. It’s really exploded in the last three months, but it is very real. And how do you see this process of adoption ultimately taking place in Europe in particular? Who is advising these boards? It is a bit of a murky world of consultants. And sometimes you ask yourself, but what do these people know? I mean, they are going around selling advice and they seem to be the blind leading the blind. Seriously, how do you see it?

YA:  So first, I actually think there is a ton of low-hanging fruit in any organization by automating the back office and the front, the go to market side. Both saving and increasing your revenue. I think there is a ton of stuff to do over there. And just like kind of managing the hygiene of how the organization works. So I actually think there is so much simple flows that if you properly automate and delegate to AI, you can actually save a ton of money, be more efficient and scale your go-to-market operations.

Yes, I also completely agree with you that most companies don’t have the right people with the right understanding for making this decision. My tips for people are first develop the muscle. I think that if I look at what is blocking big adoption in large organizations, I would separate between things that are already ready for adoptions and others that are not yet ready. I think there are still some definite technical challenges that need to be solved to completely make organizations run on AI. But like I said, the low-hanging fruits are actually ready. I think organizations don’t have the right talent, don’t have the right skills, don’t have their data in place, don’t have the right processes, cannot operate at the speed that they require. And I think even if you know what you’re doing and you’re putting a good plan in getting the talent and all of that, even in a decent organization, given the size, think that’s a two-year roadmap. But I think that within that two-year roadmap, you need to start developing the muscle.

So my first recommendation is for organizations, don’t worry so much about the ROI and how much you are spending or what you know, manage it to reasonable degree. Because the strategic threat is being disrupted and becoming irrelevant, not spending money on AI. think that’s a tactical thing that you need to take care of. So I think spend the next two years to really develop the muscle, understand what it means to use AI and just like really encourage people to use it, to automate it, find within the organizations where there are people and the right leaders that can do it. What are the right tools? What are the things that work? Don’t try to be too top down and don’t over-optimize. There is so much to learn.

So that would be one big tip for these organizations. Two, educate yourself. Like really go and find who are all the cutting edge startups that are operating in your space. There is so much like even in very heavily industrial, right? People think, I’m producing something physical, I’m protected. I’m saying no, first, like I said, other companies that automate back office and go to market have a huge advantage over you. But then there is a lot of startups that are looking how to improve your R&D, how to automate your engineering, both software and hardware, how to automate and increase the R&D, actually the research. There’s so many self-driving labs and co-scientists and stuff like that. Robotics and whatnot. So really I would encourage these boards, go and meet the startups that are actually operating in your space, understand what they are doing, understand what they’re offering, understand how might you partner with them, whether be as a customer, acquisition or whatnot, but just start having the educated view of what is actually possible because these guys at the forefront, they are the one that actually understand the vision, not the consultants, as you say. So really if you are not spending every day meeting with a founder operating in your space, and your space is like your expertise, but also startups in the generic kind of hygiene of running a large company, I think you’re not doing your job. And that’s where you will actually open your horizons on what happens.

CC: But Yariv, let me jump in. Because I know that many boards will not say I got to talk to a smart startup. The knee jerk reaction is I got to go and talk to Google, Microsoft, and Amazon, and Gertner and Cap Gemini, who will recommend the three above. And everything is contained in that bubble. And therein we go into my final topic if you like, but one that I cannot help spending time on. Because you know, I’m grateful that you suggested that people should inform themselves and talk to the small guys and they understand what they do. But which infrastructure do the small guys run their models on is also a a question. I want to get now to what I hold very dear because this discussion around AI transformation in the last few months is coming exactly at the same time, and it is intersecting this big European aspiration for sovereignty, for capability. Not in an anti-American way, we discussed it many times, and I don’t need to go there again. There isn’t any anti-American sentiment, as you know, in me, but it is much more we need growth, we need value capture, we need European assets, we cannot be building everything on somebody else’s assets.

And so let me lay out a little bit where I think the European dilemma lies. You and I have discussed it before, and I remember you telling me what you are trying to do is tough, you guys;  and it is tough, but at the same time, I refuse to accept that a place like Europe is and will remain forever a colony when it comes to these things.

So let me set the scene. We have on the one hand the European Commission and they are these nice people who tried to give us the ability to train frontier models, all of these HPCs and gigafactories, etc. This stuff around training is not at the moment something which is terribly effective, much as they tried, they underfunded it.  I’m more interested in what we were just talking about, which is inference. So you need you need compute for inference for all these agents that people are going to start running around. And for that, essentially, the reason everyone is so anxious is because the default are the frontier LLM models that come from the top labs and are bundled with US cloud. As we know, the top labs are bundled in their in their distribution with the top cloud, Google, AWS, and Microsoft. They’re not available on European cloud, the top models. European cloud has got open weights, not really these models.

This is really the origin of the anxiety of people out there who says we must do any deal with the devil to ensure access to these models because without them we are completely over. And that is one position. Then you have the siren calls of people who push for Europe to do these deals, which I think are incredibly dangerous because they imply further entrenchment. If we say, okay, in you give us access, you get all these permits, you get cheap surplus French nuclear energy, and you can build all these data centers across Europe in exchange for you giving us access to these models, that is a terrible answer for anyone who cares about European sovereignty.

And then of course there is the chicken and egg problem. I know lots of people in Europe who are willing to build data centers. And they tell me, Cristina, I don’t have a problem with funding. I can get funding from private funders. The issue isn’t money, the issue is demand.  We are at the beginning of this adoption cycle of AI. And at the moment, the demand that exists is entirely captured by the hyperscalers who are actually bundling the models with their cloud. If I’m a European trying to build a data center, which demand have I got that I can actually sell my capacity? Ultimately, the answer is Microsoft can be your tenant. Good luck. And so this is really the risk of taking these construction projects when you don’t yet have enough demand that’s going to come your way. This is also what holds Europe back.

That said, and I’m not asking you to come up with an answer, I do not accept that the only possible future is one in which we need to absolutely capitulate to this destiny. I had a conversation here with Andreas Liebl of Applied AI in Munich. He was very much of the view that we cannot accept that Frontier L LM is the only thing. There are other architectures that are possible, they’re not quite as proven, but Europe cannot give up. Sorry for the lengthy introduction, and I know you know this is my baby, and I care a lot about it, so please treat it gently.

YA: Yeah, at the end, it’s all about risks, right? And we are talking about “Europe”, but there are so many different players here, are we talking about what should governments do, what investors should do, what should small startups do, what should industry do? Each one here has a very different role. And talking as an investor, it’s a risk game. To your point, you’re saying, hey, there is a risk where we will not have access to the latest and greatest model. And I’m saying, OK, that’s a risk. And I assign to it some number. I think that for running most of SAS, I don’t need the latest and greatest model, to be honest. And then there is the risk of me using a smaller player that goes bankrupt or using models that are 100% of the time, not as good. At the end, it’s a risk game. I think there are a lot of companies are beginning to look at it kind as an insurance thing: I’m still using the big US, but I need as an insurance also fallback. The two beneficiaries of that might be Mistral and Cohere. Funny that Canada, could be the big winner of that. And my biggest complaint on Europe is a little bit like this person that prays every day to God, help me win the lottery, and then God says, buy a ticket.

Anthropic and OpenAI are making huge bets and huge risks, they can go bankrupt if they are wrong. The others as well. If Europe wants to do it, Europe needs to understand that it needs to move as fast, that to take the risks that are as big because you cannot be super conservative and also take long term. I think Europe dug itself a very deep hole and there is no magic solution in the near term to get out of it. And then the question is, okay, what is our strategy that is short term? And maybe the short term strategy is the thing that you don’t like and saying, okay, let’s ensure access. Okay, we can tell these guys if you want to be present here and make money from our own European things, okay, we’ll give you the energy, but we need to make sure that we are not just deepening the hole, and in parallel build the mid and longer term solution here.

Europe at the moment is so behind on multiple levels in the game here. Like you said there is no cloud compute infrastructure. Then there are no frontier models. Then if you look at how many leading global startups are happening in Europe vs happening in the U.S., we’re so behind. And then if you look actually at the governments and the industries in Europe, how much are they willing to take risk and bet on European technology vs go the easy way and buy US technology, we’re so behind.

CC: Yeah, I know all of this. And it is our endemic problem. But I think what you call potentially a short term solution is horrifyingly unacceptable and dangerous, and I also think that these contracts are in any event not enforceable. If you enter into contract that say we want model access in exchange for you carpeting Europe with data centers, they would say, Yeah, yeah, yeah, sure, sure, we’ll sign this piece of paper. And it is completely unenforceable, in addition to which, you know, this is one-way bets. They got in and they carpeted Europe with data centers and we’re out of the door. Europeans are not in play.

YA: Yeah, I agree with you, not saying…

CC: We are entrenching further. We are further colonized. I mean, you need to make a judgment that access to this model is so absolutely existential that it is worth trading ultimately agency and ownership and control for. We would not control these assets. You have layers of infrastructure that are not controlled by us. It is not acceptable.

YA: But I would like ask you, can you stop that? What’s the mechanism to forbid that in a democratic, open market to block from all the different players in Europe? So I’m saying, how can we focus that energy, not to try to stop that, but what needs to happen quickly to think deeply the alternatives? And the other thing I would think is: if I am betting, if I look back on the internet and mobile and social, the companies that were big in 1999 are not that big anymore. Like should Europe actually come and say, instead of worrying about automating the giants of today, can we focus on the giants on tomorrow that currently are very small? Right? Like let’s find the top and most promising startups. Let’s bet on the future. Let’s see how we can support the compute and model. And I think it’s an easier problem. It’s a risky one. But if we have conviction that the landscape will look much more different, it’s a smaller problem. That’s a direction that I would like to look.

CC: I think there is no question that no one in the sovereignty movement believes that we can go to the mat with American frontier labs and create an alternative in Europe. The conditions aren’t there. Europe is not going to suddenly create an alternative, throw billions and trillions at it. You know, this is not in the realm of possibilities. On the other hand, there are other things that Europe is can do, potentially smaller model for different applications. Not everything needs to be a frontier LLM. Do we need these models for every conceivable application? Certainly not. Not to do to do many, many things. What is the gap between these and what is perhaps a couple of steps behind?

YA  Yeah. I think there are two opposing powers. I think that to me, I think there is a huge advantage in having these super powerful model that then you can distil down to do specific things. By the way, when you have a super powerful model that you spent, you know, trillions into training, distilling from it specialized, cheaper, faster models, assuming that the data exists within that and they’re spending a ton of money on the data, it’s easier. And I do think that the stack is getting much simpler, right? Like if you look today, the SaaS stack, it’s very complex. And it allows for many players and for a lot of advantages. I think the future stack is very simple. And unfortunately, it relies a lot on ability to do compute extremely efficiently and intelligence extremely efficiently. And these are the places where the frontier models and the hyperscalers excel. I don’t see how to escape that.

CC: But this is like saying that this kind of scaling, scaling, scaling architecture is the only one?

YA: No, I think that at the end most of the value in my opinion will be captured in the compute and in the core intelligence. I think then there is a second question of when you do inference, can you be a market with more players? The inference is a very different game on how to serve that. And over there, you want to utilize all the possible hardware. You also want to extend the lifetime of hardware. So I don’t need every two years to buy on the latest NVIDIA and stuff like that. I want to be able to run stuff locally. So I think there is a lot of advantages to run not necessarily on a hyperscaler. And I think there is a real opportunity here.

The way I would approach it pragmatically is to come and say, okay, what are the first verticals, players, companies, segment geographies that are best set for success to start that game as kind of arrowhead or landing beach to start building it and then expand? Like I wouldn’t try to come with a plan to solve everything, come and say, okay, I’m betting on running inference. There is already a lot of companies looking at it and saying, how can you manage your small hosted group of computers as if it’s Google hosted? There is a lot of them, when you don’t have the skill, when you, know, and you don’t have the hardware and so forth. So I think that’s interesting. And to be honest, my knee jerk has been…when companies told me we’re doing this because of European regulation or European availability, I don’t think that should be a reason.

I come from an intelligence background. In intelligence, you always separate between capabilities and intents. Because a capability, whether someone can or can’t do something, changing that takes very, very long. Intents, whether someone wants or doesn’t want to do something, it’s a flip of a coin. You cannot bet on that one. So I think whether European companies or others want to adopt American or don’t want to adopt American technology is something that can flip in a day. Suddenly the president changes or God knows what happens. So I think a company that advocates to run stuff locally should have a solid business case and value that is super competitive compared to whatever hyperscalers can offer. Even when they are competing against each other, they should have a strong reason. If they don’t have a strong reason, I don’t believe they have a good reason to exist. The fact that “we are Europeans” as an investor I don’t like it. I think there is some kind of backwind that supports it now, but how long will it last I don’t know. At the end, price, availability, and comfort, you cannot compete with it. If something is better and cheaper and easier to access, at the end we’ll do it.

CC: But as you know the focus on Europe is not unconditional. I certainly do not advocate that we adopt certain solutions just because they’re Europeans. I feel very patriotic about being European, but there needs to be a business case and cost and efficiency need to play out. I mean, there’s no question that these things are as important. But at the same time, you may feel that this is not something which may not have durability, but I think my perception going around Europe as much as I do is that the vision that Europe needs to power itself up and reduce its dependency is very, very established by now. And so whether we have the capability and the intent to act on it, I don’t know, in full, but there is a big shift. I mean, seen where I am, the mountain is shaking.

YA: Yeah, but again, it goes to risk, right? Like you’re always asking, who controls this? And what’s the reason that something bad could happen? Geopolitical is one risk, poor operations is another risk, bankruptcies, there’s so many. And again, when you look back 10 years, and you’re saying, how many things we took for granted 10 years ago, suddenly it completely flipped 180 degrees. If I had described to some of the geopolitical things that happened in the last five years, 10 years ago, you would have told me you’re crazy. So as we think 10, 12 years from now, I think a lot of things can slip in. I definitely agree with you that…

There is a trend for concentration of power in the hands of very, very few. And that’s scary for companies whose existence depends on it. And I think that companies need to think very, very seriously about what do I do in such a world? What is my primary strategy? And what is my kind of insurance policy here? And maybe I cannot be without an insurance policy anymore because the risk is so immediate. And I think it’s very different for countries that understand that AI now is part of the basic security and infrastructure of that company. It’s very different from whoever controls energy as a resource. I think it’s very different for the large companies.  And I think every one of them should understand that, yeah, geopolitics, the concentration of power, the dynamics of US versus Europe have changed. And you cannot ignore it, to your point, at least for the time being. And I agree with you that I think when I actually speak with many people, they’re not even thinking about it. And I think that’s idiotic. And it’s a train going into a wall.

And I think what you are doing of raising awareness, calling bullshit, making the conversation educated, getting the perspectives from everyone around and like not letting go with like, know, populist answers. I think it’s the way to go. You know, this is why I’m here. I’m a big fan.

CC: Yeah. This is the note to end on. A bit of praise. I love it. Listen, we’ve spoken for over an hour. So I think I think we cannot speak any longer, but when I speak to you an hour passes quickly. So thank you. I think this was fascinating for me. Great to have you, Yariv Adan. Thank you very much.

YA: Thank you. Yeah. It was fun. Bye.