Is Europe’s AI infrastructure build-out heading for a gigafactory-style bust?

0:01Hello, I'm your host, Amy Lewin, and welcome to the Sifted podcast. Today, we're tackling a meaty topic that we're all going to be hearing a lot more about soon, AI infrastructure, and more specifically, whether Europe has what it needs or a plan for how to get it. Joining me to discuss all of that are two of my colleagues, senior reporters Daphne LePrance-Ronguet and Freya Praty, who've been digging into all of this recently. Hello both. Hello. Hello. Where are you joining us from today? I am joining from Paris, where I am based. I'm in London at the Sifted HQ. So first up, why is the startup ecosystem and why European policymakers quite so obsessed with AI infrastructure at the moment?

0:46I feel like this is one of those things where we get a lot of press releases at the moment, especially from government sort of tech department, but also big groups like the European Investment Bank, the EU want to know, you know, it feels like they want everyone to know that they are doing something about this. Why is that? So I think, to be honest, this is a continuation of a problem that Europe has had for a long time and that started with cloud. So Europe has wanted cloud sovereignty for many, many, many years and I think historically everyone would agree that it has failed to achieve that.

1:26But it's become even more pressing today because those huge US-based cloud hyperscalers like AWS or Google or Microsoft, they are now those who are securing the computes, the GPUs, which we all know of now, which power AI models. And they're building the data centers to host these GPUs and to power AI. And now what's at stake is to be able to at least reproduce, this in Europe and gain independence in that way from the US-based cloud hyperscalers who have kind of taken on providing the infrastructure for AI that everyone desperately needs. And I think European countries have understood this for a while.

2:06I remember one of the early pieces I wrote for Sifted was in 2023. I found it back. And I was already writing about how France was trying to position itself in the race to build AI infrastructure and trying to encourage more private investment into data centers for AI specifically. But then I think there was a real turning point at the start of 2025, when Trump was elected president in the US, when there was a realization that we really can't fully rely on compute power that's stored in US data centers by US hyperscalers anymore. Because what if Trump switches it all off tomorrow? So I think that was the first thing.

2:44European countries saw that AI needs were going to grow massively. And it became a priority to ensure that they could respond to these needs independently from the US. And I'd say the second thing was that at the same time, AI infrastructure became a sort of proxy for how successful a country was, quote unquote, doing AI. And so we really saw this when the Trump administration announced the Stargate project, which was unveiled at the start of 2025. That was a real show of four to 500 billion, well, a commitment to a 500 billion investment over several years, a joint venture with private companies like OpenAI, SoftBank, Oracle, etc.

3:24There were more as well. I think Europe kind of responded to that. And then every European country, as well as the EU as a bloc, started kind of putting out these announcements for how much infrastructure they were capable of doing. And in that way, AI infrastructure kind of became that kind of became equated with how well you were doing as a country in terms of supporting your local ecosystem. So it's a bit of a show of power, I'd say, and particularly in the past year or so. And Freya, can you maybe lay out some of what the different European governments have promised so far? Who's looking best at the moment, as Daphne says, in this kind of display of we're going to be great AI?

4:03Sure. Well, in the UK, where I'm based, last January 2025, the government announced an AI opportunities action plan, a key pillar of which was to build out compute capacity. They designated four areas of the UK as growth zones, which would have streamlined planning requirements to allow data centers and to quickly pop up and skirt the typical planning pipeline, which can be quite long. Those four zones are all in the stage of either planning and permitting or seeking investment partners. So we haven't really seen them get off the ground yet, but they are meant to be, when they come to fruition, home to some big projects.

4:43The UK has also had NVIDIA, Microsoft and Google all commit billions to growing compute in the UK. So the UK has had a lot of attention so far. I think the question for the country is whether our energy prices are a lot higher than other countries. So whether it makes sense economically to build this stuff here is still a big question, even though the demand is here. I think it's the same in Germany. The prices of energy there are higher, so the economics are trickier over there as well. Places like Spain and Italy and the Nordics might make a bit more sense, as well as America, which has cheaper energy than us.

5:21And what's happening in France, Daphne? So in France, the big announcement was made in February last year at the AI Action Summit in Paris. So President Emmanuel Macron announced 109 billion euros in private investments for infrastructure projects in France. So this came from a host of different private players from around the world. that was probably the biggest well it was the biggest announcement in France really and going back to what I was saying it was presented by Macron as proportionally to French GDP equal to the Stargate project in the US so going back to that point of kind of who's going to be deploying most money for this yeah kind of putting everyone in that race mindset I think.

6:06And when this just you know the capital requirements of these projects are so huge and And also, I guess the time it takes to get those kind of permits, acquire the land, build these data centres. This isn't a super fast moving thing. Is this really an area where startups can play and are they trying to? I'd say it's mainly not startups. As you said, it takes really deep pockets. The biggest data centre in the UK in terms of capacity is a site near Newcastle in the northeast. and that's being built by Blackstone, so this huge investment firm. That's the more likely profile we'll see behind data centers or them being tied to the huge US tech giants like Microsoft and Google.

6:53There are some startups working on it. The biggest one in the UK is Nscale, which set up two years ago and raised a big round at the end of last year. They raised $1.1 billion to build a data center in the UK. They've also got some sites overseas as well. So they're one of the bigger startup players, but primarily we'll see this from asset managers or infra funds that are accustomed to building classic infrastructure like airports or car parks. They're the kind of people that know how to put the cap table together for this and probably have the connections and money behind them. Startups, I think, are more likely to sit around the edges in terms of things like tech that could help a process within the building work better or some kind of cooling process maybe.

7:38I think that's probably where we're most likely to see more startup action because building infrastructure in itself is very, very capital intensive. One company I noticed maybe goes against what you've just said, Freya, is Evrock, which is a Nordic player and has raised about 100 million euros so far last round in March 2025. it said it told us at the time it was planning to operate 10 hyperscale data centers by 2030 and employ more than 10 ,000 people across Europe. It sounds pretty ambitious to me and that one is backed by some VCs. Its backers include EQD Ventures, Norskern and Giant Ventures.

8:20Freya I know you know some of those investors reasonably well. What's different about Everrock or why have why have they decided to take a punt on something that's a bit outside of typical VC comfortability there? I don't know Everett's cap table in depth but I imagine they'll be coming on with equity and then probably Everett will look to raise debt or project financing for the build-outs themselves but that's me guessing based on the climate tech infrastructure build-out which I followed quite closely and which followed that kind of setup where you'd have VCs early on usually investing in the parent company or the top co and then they'd each individual infrastructure project would look to some kind of debt vehicle.

9:05The other one that kind of defies the trend I was talking about earlier of this being built by much older companies is a Helsinki company called Data Crunch which raised 55 million last year including from some more sort of VC names. They're already operating data centers in Finland and Iceland which goes back to this question of it's much easier to run them in certain countries which might be the reason a startup's managing to do that already. They have much cheaper and freer flowing energy in Finland and Iceland. What about mistral daphne are they planning to move into this area obviously they're a big um model builder but they're also a company that's very much positioned as you know building sovereign ai for europe what's their kind of connection with the data centers and is that an area they're edging into as well yeah it's actually an area they have edged into already uh last year although So they were always interested, as you say, in sovereign infrastructure.

10:06And it was kind of always part of their pitch that if there were opportunities to secure their own sovereign infrastructure, they would take that opportunity and why not invest in it as well. In the end, last year, they opened their own data center in partnership with a data center operator here in France, which I believe is already operational near Paris. Then last June, they announced Mistral Compute, which is very similar to what a cloud giant would offer, only at the Mistral scale. So obviously, they're not operating massive data centers all over France. But it does consist of essentially leasing NVIDIA GPUs to clients who are seeking a more sovereign proposition than US hyperskaders.

10:55And then in parallel, it's not entirely clear how this project ties in with Mr. All Compute exactly, but they announced a joint venture with NVIDIA, with BPI France, which is the French public bank, and MGX, a UAE fund, to build a huge AI campus in France, which presumably will be a host of quite powerful data centers. So yeah, they are very much involved in trying to build up AI infrastructure. And interestingly, after they announced all this, they raised a 1.7 billion series c at 12 billion euro valuation and at the time one analyst told me that so that valuation was was obviously very significant and quite impressive and what they were saying was that it was why it was largely tied to their commitments to ai infrastructure and building out data centers because that was kind of what they assessed was a bubble within the AI bubble that was developing.

11:57So where investors previously probably didn't care that much about data centers, now there's a lot of interest in that and startup valuations in the sector are really inflated. So yeah, today I think GPU leasing does represent some revenue for Mr. Owl. Whether it will grow significantly or not remains to be seen, but it's definitely contributed to valuing the company quite significantly. And one thing I find pretty interesting about this is, Freya, you mentioned earlier that in lots of ways, these big data centre build outs are comparable to some of the huge climate infrastructure projects we saw a few years ago, like the gigafactories built by companies like Northvolt and Britishvolt, most of which completely went up in smoke.

12:43They raised tens of billions of euros, started building numerous sites, and then things fell apart. Why is that not perhaps more of a cautionary tale for investors getting into the data centre build out? Or why do they think it's going to be different this time? I think people recognise there were particular dynamics around batteries, which is what those gigafactories are producing. facing Europe's entirely undercut by China in that. And although there was a push for sovereign batteries, I think it was perhaps less high up on the political agenda than sovereign AI has become, though it was also quite high up.

13:27So it is an interesting comparison. I had a chat with a banker who's worked on some of these deals last week, and he was saying to me that data centers and infrastructure are the safe play for the big funds because you're not betting on a winning or a particular company winning you're just betting that ai will be big which is quite a safe bet in in this person's view so he thought that's why we're seeing like brookfield announce this like 100 billion dollar fund it's they know that ai in some form will be big and this is how you make that play you back the infrastructure behind it that can kind of deal with all forms of ai um but it's a big question for particularly youngish companies the gigafactories that came before them do offer a lot of warning.

14:13I think the message from those is the companies that crashed were the ones that spread themselves very thin and started working on a load of projects all at once. Hopefully there's some lessons for the AI infrastructure waves, but we'll see. And a question around sovereignty, pretty much all of these, correct me if I'm wrong, data centers are underpinned, powered by, invested in, by in some way or other, Nvidia which is not a European company how do policymakers investors these startups sort of square that circle maybe Daphne on the Mr. Owl's also got a huge partnership with Nvidia what's their kind of response to that?

14:58To be honest it's definitely a question that comes up every time I speak to investors about building out a sovereign European AI infrastructure and I don't think anyone has an answer to it. Ultimately, NVIDIA has probably a complete monopoly over the chips that we put into data centers. And all of these data centers, as you said, are filled with NVIDIA chips, which just means our dependence as Europeans is placed on NVIDIA rather than on AWS or Google or Microsoft. It's a difficult question to answer. You could venture into saying we need to build our own chips, but I don't think that's a realistic answer in the short term.

15:42So, yeah, it's not really an answered question today, but it's definitely I've seen a lot of eyebrows raised at European countries claiming sovereign AI and then building out data centers powered by a US big tech company like NVIDIA. When we're talking about data centers for anyone who's kind of struggling to picture these things, how big are they? Are we talking like the size of a giant Amazon warehouse? Are they bigger than that? What kind of energy requirements do they have? And what impact does that have on the local area? I imagine they're being built kind of in the middle of nowhere. Freya, do you know more about all of that?

16:25I think the size can vary hugely from my understanding. It depends how much compute the site plans to offer. A lot of the big ones are being built far from cities. But interestingly, there are a number popping up within London. I think property developers are realizing the money they can get from this. And so inner city, data centers are becoming more of a common occurrence. There's a really interesting one planned or proposed for a site in Brick Lane in London, which was a brewery. And they're now proposing to turn it into a data center. It's had a lot of backlash. And obviously, one of the big things is these data centers produce a lot less jobs than the alternative things that could be on those sites.

17:07The site that Blackstone are working on in the northeast near Newcastle was the old British Fault site, which would have, I can't remember the number of jobs British Fort Wynter promised they would bring in, I think it was 3 ,000. And I imagine this data center will bring far, far fewer. So it's also data center versus the alternative that we could be using that site for. I think often local opposition comes from that. And I think it's an interesting step back to take. There seems to be a consensus in the tech world that it's a good thing to build out data centers all over Europe. It's something that we want in France.

17:43There's currently a project for a law to streamline applications for data centers so that they can qualify as projects of national interest. So I think it's very easy to take it at face value that these data centers are good. But actually, the reality is that outside the tech bubble in civil society, there is quite a lot of resistance to these projects. Obviously, their energy and water use is a huge constraint on local resources and local communities. So it's not necessarily something we should accept as inevitable. And I think it also brings the question of whether the right strategy is to race to build as many data centers as we possibly can in order to match what's being done in the US.

18:24Ultimately, do we need as many data centers as there are in the US? Or does every single AI company in Europe need to be building their products from scratch in using sovereign AI infrastructure in Europe? Or is there a more strategic way to think about it where we have a safety net without having to invest the same amounts of money and resources as they do in the US? And I think talking with various policymakers, particularly, they seem to say that that has been lacking. So there's been a lot of announcements, but there hasn't been that many announcements of strategic thinking or, you know, planning around where we're going to put these data centers and why we're going to use it and whether we need them entirely or not.

19:07So I think that would be probably welcome for a lot of the public and probably a place to start public debates about them as well. But it does feel like we're seeing a lot of announcements and not much of the planning behind it, at least coming from politicians. Yeah, I guess there's a world in which in a year's time, we all realize that AI is nowhere near as useful as we thought it was, and we don't use it as much, or somehow models become so much more efficient or whatever. And then we get to 2030 when, you know, dozens, hundreds of these centers have been built and there isn't the demand for them.

19:46I mean, I guess it might be hard to find that many people who would strongly argue that that's how things are going to pan out. But as you say, if there's governments plus private players plus, you know, organizations like the EU all piling in on this, has anyone really done the maths and said, you know, this all adds up to X number? Do we actually think we need this? I think there will be demand, but is there a need for all of this to be sovereign? In very pragmatic terms, you look at the numbers in the US, we're talking about trillions of dollars being invested over the next few years. Europe is obviously not matching that.

20:25So is there a more efficient, smart way to think about this? I think is a question that's coming up a lot. Perhaps the most intriguing company in this world is Nscale, which, Freya, I know you've looked into a little bit. It's less than three years old, founded by a guy called Josh Payne, who hasn't really got heaps of experience in this world. he kind of evolved out of a crypto mining company and it's raised billions. What do we know about N-Scale and why are so many people taking a punt on this business, which doesn't maybe have the kind of serial founder, worked in this and this and this before that you often expect from such important and well-funded companies.

21:19Yeah, as you say, they emerged quite quickly and quickly gained a spot within the UK's AI strategy. So they've been mentioned in a lot of government communication around the UK's effort to build out its AI capacity. I was originally, like many journalists and people in the sector, slightly like, whoa, where did you guys come from? I've had a few more chats with people recently who've talked to me about the crypto aspect in particular. Some people see it as quite a long-term play. They gathered these energy assets, which you need for data centers. They then use them for crypto mining to reap revenues while they were planning the data center build-out.

21:58And now they're ready with the energy assets to produce data centers. So some people view it as actually a quite like under the radar, but clever move. Others view it as you came from crypto, where's your data center expertise. I think that although the founder is quite young and there have been accusations that the team doesn't have a huge amount of expertise, they've recently hired a new COO, chief operating officer who comes from Palantir, which although doesn't make data centers, I think that person seems to have a good profile. They've also acquired a company that works on the engineering and management of AI infrastructure.

22:38So I think they are making moves to bring in that expertise. And they raised a big sum of money. They raised, as I said earlier,$1.1 billion. So people do kind of trust them in that way. They are just unique in that they're a young company competing with, as we've said lots, huge US tech giants or asset managers. So they're intriguing. The jury is out, I guess, until the data centers are built, but they're one to watch. We briefly mentioned earlier the sort of less costly, more startup-y plays around AI infrastructure. Daphne, what are the other elements when we talk AI infrastructure? We don't just mean data centers, do we?

23:21What other technologies are being developed or what What other kind of capacity infrastructure needs are there to power the AI revolution in the years to come? Well, as I mentioned earlier, I think chips, semiconductors is something that a lot of startups are looking at in Europe. So trying to figure out what is a more efficient way to build chips, what are more efficient architectures, architectures that are built for AI rather than leveraging GPUs, NVIDIA's GPUs, as everyone is doing right now. That's a very difficult play. You're up against huge companies and the fabrication process is very complex.

24:07And for now, it's in the hands of just a few producers, mostly in Taiwan. So there are a few of these and they're very promising, but the scaling phase is a huge question i'm personally quite interested at the moment as well in companies uh that are looking at the memory bottleneck in ai so gpus are also contain a component that is uh memory and there is a big bottleneck for that component particularly and it's caused it's actually currently causing huge shortages and price hikes that are already affecting quite a lot of businesses so similarly there's a handful of startups in Europe who are looking at how to make the memory component more efficient for AI.

24:58So yeah, those are just examples, but it's obviously a big equation to solve and there's many parts to it. And obviously we know that AI data centers use up a humongous amount of energy and have sort of there's been a fair bit of criticism around that does this have any ramifications for the currently slightly beleaguered climate tech sector Freya? Yeah as you say climate tech has totally fallen off the agenda since Trump took office and I think that's had ramifications outside of the US across the world as well European climate tech has had far less funding than it once had. So there is a hope among lots of people I talk to that although AI is very energy hungry and water intensive, that it could act as a sort of Trojan horse for climate tech, which is a really interesting and hopeful idea.

25:55The idea that data centers will need to co-locate with some form of energy generation and the energy generation that works best is something green. Data centers can't go offline for even like a thousandth of a millisecond. And green energy typically is better at providing that consistent supply as opposed to the grid and the kind of fossil fuels that we have within the grid. So we saw recently Meta announced a series of nuclear energy deals to power its data centers in America. Amazon has also signed deals with small nuclear reactor providers. So I think we're seeing that those companies are clocking on to the fact that they need power and the best power they can look to is green which is very hopeful could potentially bring climate further up the agenda kind of under the guise of sovereignty and the need for a country to compete within this AI race it's quite a hopeful take but we are seeing the deals to back that up with Meta and Amazon recently making those.

26:56Very interesting. And what gives you confidence, Daphne, that this isn't all a huge bubble? Or do you think it is? I think it might be. But it is. AI is a huge bubble and AI infrastructure within AI is a huge bubble. That's different from saying that it's going to explode and everyone is going to forget about AI. Obviously, it's a revolutionary technology that everyone in Europe is going to use. and we are going to be needing data centers and we're going to be needing data centers in Europe to be able to have some form of independence and sovereignty in AI. So I think that's not going to go away.

27:34There's obviously a lot of political will to support these projects. Capital is being unlocked and more and more promising companies are emerging in the space. It takes time as well. I think a lot of these announcements have been made in the last year, which is in data center time, is not long. So there's still time to see how successful they are, but it will certainly be one to watch and see who are the winners and losers in this. Fantastic. That is all we have time for. Thank you so much, Freya and Daphne, for joining me. Thank you. Thanks. If you want to hear more from Daphne on all things AI, please sign up to her AI and Deep Tech newsletter.

28:15Frequently also features Freya, so you can follow her there as well. It comes out weekly on Mondays. We'll drop a link in the episode description. As always, please rate, review and share the podcast. And this episode was produced by the wonderful Maya Durampal-Bodby.