AI Infrastructure as an Asset Class

All right, thanks everyone for uh for sticking around. uh we're going to turn it up a notch from people and from investment in whatever everyone talks about now going a layer deeper because what happens with the workflows that you build or the when you enter ChatGPT or even what I just heard when you order takeaway at your hook shop the data goes somewhere there's an infrastructure underneath it it's a physical space usually which we all know as data centers but Who owns the data center? What is the data center really? And what is the infrastructure? Why is it so important?

And how do we see it from an investment perspective? So please Paul and Bjarke from the come forward and introduce yourself and let us know what you have to say today. Thank you. Uh good to be here. Lots of familiar faces.

Um for myself, my name is Bjarke Mikkelsen. I moved back to Denmark about two years ago after 17 years abroad. I was eight years with Goldman Sachs in London and then I was uh I built a tech business in Asia, sold it to Alibaba. Uh spent the last five six years in Singapore and then moved back and made a partnership with Lars Tulander who's known by many as the godfather of real estate in Denmark. So we thought tech and real estate together we have to build data centers.

That's the the short version. in addition to a bunch of other stuff. And hey everyone, my name is Paul. Um, my background is within energy. I've spent 15 years developing different energy projects in Europe and lately in Asia where I met Barka.

I lived in Singapore too for um 5 years. developed a big project for TSMC in Taiwan who offtook the the full almost a gigawatt from from an offshore wind farm and and I I know Barker privately and he called me up uh in the beginning of 24 and pitched the idea of a data center to me. I didn't know what a data center was. I think a lot of people were at that position. Maybe it's it's probably different now, but it became clear that energy was a part of it and and it would be even bigger in the future.

Um, so yeah, that's me. Great. Do we have our our presentation? Good. Who's controlling this?

Are you? Okay, great. I'll I'll do the first part. Um, and I I'll warn you, the the first part of this presentation is going to sound like a sales pitch. Uh, it's really not.

The second part is going to talk about why it's not as easy as it sounds. Uh, just so you're warned. We have a bunch of fun facts. I'm sure many uh many people know lots of fun facts about AI, but I think we should actually start with something that's extremely current, which is the the biggest question that's been on everybody's lips up until I would say 60 days ago. Any guesses?

No. Is the AI revenue going to show up? That's been the biggest question, right? The infrastructure investments are there, right? The rollout is happening.

There's no question about it. The usage is there. Everybody's using AI. But is it driving incremental revenue and is it actually generating returns on these huge infrastructure investments? And I'll give you one reference point.

Anthropic, which is, I would say, the leading AI model for B2B solutions uh used by until recently the US military for example. uh they increased their annual recurring revenue by 6 billion just in February and then they added another 6 billion just in March and that means they went from basically nowhere as a reference Nova Nordisk has annual recurring revenue of maybe 50 billion they added half a Novo in two months almost right so the revenue is there right this is B2B but ChatGPT also 20 billion annual uh recurring revenue So the revenue is coming and that's really been the key question that everybody's been waiting for. Uh can we actually make a return on these investments? Um and speaking of anthropic, I thought this is also an interesting opening. So I don't know how how much you know about the the recent fallout between Trump for the US military and anthropic but basically the US government they use particularly Palanteer one of the biggest I'd say hardware military startups uh for equipment like drones and stuff uh and then they used up until recently Anthropic as the AI model.

So, if you want to build a satellite defense system that can take out missiles from space or if you want to build a drone uh army that can attack targets using AI models and be self-reliant, you need AI models. And the US government does not have its own models. So, they have to pick a partner for it. And what happened was they they picked Anthropic, which is part of why uh Anthropic's revenue is exploding. But it became a problem because suddenly you know you start discussing terms the terms of agreement like what happens if there's an unforeseen event and anthropic they say no actually you cannot use our software for that or if there's a call it a rogue developer who suddenly poisons the code and it doesn't work and you're in the middle of a an attack in Iran or you're kidnapping a a president right and the AI model breaks down because it's out of your control and that doesn't work.

So, the conclusion was that Trump basically fired them and he said, "Well, then you're out. Like, I need full control over the AI software. Um, and if you can't give it to me, I'm going to get it somewhere else." And the conclusion was basically that even inside the US, even inside the US, not talking about Europe and US yet, um the government, the government needs data sovereignty. You basically need to control your infrastructure, your AI infrastructure. So the conclusion is that the physical infrastructure you have to have within a secure perimeter.

I'll talk about what that means in a minute. And then you need to be multimodal. So you need to have alternative models that you can work with and you cannot be entirely reliant on one partner because that partner can either move against you or they can suddenly fall behind the curve and get overtaken by somebody else and then you're behind the the game, right? And that model being physically uh secure and having control over your physical infrastructure and being multimodal is something I'm going to come back to. I really think that's the it is the only strategy that a country or a government or a company can have in the future.

So what is AI infrastructure really? It's most people think about it as applications like ChatGPT or the foundational models like GPT 5.3 or Opus. Uh but it's also the three things below which are the physical parts where we are playing in the energy part and the infrastructure part. We're not investing in chips. The customers that live in our data centers, they bring their own chips.

Uh but we build everything else for them so that they can bring their chips and run them in our data centers. But this is basically the five layers. The energy, the chips, the infrastructure and then the models and the applications. And where does Europe stand today? Not not very uh advanced.

So a couple of reference points out of the cloud infrastructure in Europe, 70% is controlled by US companies and only 10% of the AI infrastructure investment went into Europe over the last what is it? Well, since uh 2013. So Europe really needs to get its act together with data data sovereignty. We cannot be reliant on physical infrastructure that is not controlled by us. And that was our strategy.

Basically, let's build the first Danish owned and operated data center platform where we own the steel and people can come put their physical chips into our infrastructure. we can build the security the cyber and the physical security and all the infrastructure around it and then we can use separate models uh that we can use from different parts of the world it could be US could be China could be anywhere um and many people ask me so you're you're a real estate guy so h how come how come data centers is are you really the right person to to do this and the way I look at it is that this is a chart from the US which is I think a a couple of years ahead of Europe, but this is coming in Europe where construction spending in the US for data centers has now and this is 6 months old, but now it has actually overtaken office spend. So from a real estate perspective, this is the asset class. We we like two things. We like residential and we like data centers.

So the I would say the asset class is there. There's no question about it. And as I said, up until recently, the only question has not been about is the rollout happening. Are the tech giants going to keep spending money and investing in this because it's a must-win battle that you cannot afford to lose. But as an investor is actually driving returns and now we have the proof just literally within the last 60 days.

Um, so where is it? Oh, some of the some flags there that are showing different jurisdictions. But basically, um, what we wanted to show here is where is Europe going to build this capacity that we need in order to control our data sovereignty. And what it shows is that there's there's primary markets in Europe, flat D, Frankfurt, London, Amsterdam, Paris, and Dublin um that have, I would say, absorbed most of the first wave of data center investments. But most of these flap D markets are showing significant signs of um maturity where there's either not enough power or there's not enough land or it's just getting impossible to get permits for more data center projects.

So if you look at it here, sorry the chart is a bit messed up, but in 25 basically 34% of the data center capacity in Europe is outside the flatb in what we call secondary markets like Denmark or the Nordics. But in 2031 that's going to be more than 50%. So most of the growth is actually coming right here. Um and when you look at it from an investment perspective, you can see these are uh this is the European data center capacity with the light green is the secondary markets. It's increasing up until 25.

Um everything that is being built is being sold, right? So, it should be a no-brainer if everything that's being built is being sold. If the revenue is showing up and the tech companies are going to keep investing in their infrastructure rollout and Europe needs data sovereignty by definition, otherwise we're just not going to stay secure and independent. Why are we not seeing more data centers being commercialized? There's a lot of talk, but there's not a lot of people who are actually doing it.

So, my point in the beginning was it sounds like a sales pitch. Look, it's a no-brainer, but it's actually really, really hard, right? And so, that's a good segue to to Paul. So as he said he was the first person I called when I said look he's he's delivered a 1 gawatt or 900 megawatt uh infrastructure transmission project on the biggest at that time private PPA to a single client which was TSMC and Paul was a project director. So if anybody can solve I say complex large giga projects is Paul.

Um and that's what he's going to talk about for the rest of the the session. How to really create an investable product that has a risk profile where you can actually get it done. Yeah. I got the hard part presentation. So we saw from the previous slide how Denmark is is nicely positioned amongst the Nordics.

experiencing the most growth in terms of the Nordics the next five years and we believe that there are many underlying good reasons for that and we've highlighted some of the some of the key ones here. Number one being an abundance of green energy. So we source around 75% of our energy from green sources and that energy is flowing through an incredibly reliable grid. We have one of Europe's most reliable grids next to Luxembourg and that's because we've spent the last 20 years burying most of our cables into the ground. Secondly, we enjoy geopolitical stability.

Uh we have a digital society and we have predictable permitting processes compared to our uh competing markets which is great for these kind of capital intensive long-term investments. And lastly, Denmark is really nicely connected through high-speed fiber, not just to the rest of Europe, but also to the US. So those three things together make for a really nice foundation for why you should roll out data centers in Denmark. But there are challenges. Two of the key ones, not just in Denmark, but in Europe, and the world, are access to power and permitting.

We understand those issues quite well, but we'd like to highlight some of the issues that are tougher to solve. number one and yeah again it's a bit messed up here the text but if you look at Denmark and really relevant right now where our TSO and again have closed down for new applications because the grid is just overflown with new applications I mean how do you prioritize that are data centers important to Denmark or should the grid go to new hospitals or how do you prioritize our peak load is 7 gawatt we use four on average and there's applications worth of 60. So they got to clean that up and and where do data centers land in in terms of priority. Number two, we find it to be a difficult conversation when we meet with IG companies and we talk about them in our data centers cuz they need to go out and invest in expensive GPUs. They all have an AI strategy, but it's not that many that are willing to invest in it.

buy expensive GPUs, sign a five-year lease hold or more, and move in. There's also Neoplouds. It's a new phenomenon, hence NEO. Um, they're more flexible, uh, but they're also less creditworthy. So, what do you do?

Do you fill your data center up with them because they're willing to come in, but what happens if they go bankrupt? So, that's the two extremes. And what we've found to be important for both is time to market. That's really, that's how you win. But it's not um it's not easy to go from a early development project where there's a lot of risk to an execution project where it's down to execution risk.

So it's really about how you decipher between the two. And we think solving for time to market can be done if you look at the if you look at the asset in this way. We have structured all of our activities during development on the criteria that you see here. These are our final investment decision criteria which will basically uh be our key focus during um during the maturation of the data center. So you have a site, you have a grid connection, you have your permitting, and once you've delivered on all those criteria, you're basically at a stage where you execute and where you're limited to much more predictable execution risk.

But it really requires that you can bring together four pillars and we've listed them on the left. So if you can develop and construct, if you can solve the energy problem, so you can get your grid and you can get your you can source your electricity, you can solve for your capital where does the money come from and you can find the customers, but you can do all of that simultaneously and make them meet and it's at the nexus of those four that you get your value. Now if you look at how we usually develop uh a data center and this is just for illustration I mean there's variation case by case but development is usually 6 months or more where we where we focus on the the list of criteria that I showed you before. So we find a site we find an energy partner we go through a permitting process with the local municipality and get our local planning in place. We get our grid connection we do our design.

We buy our components and we get our offtake agreement and by the time we enter execution we've roughly spent we try to limit this 5% of our total budget and that's really uh that's really where where there's value as well in balancing spending before you are at that FID gate because that's where the risk is is is is minimized and then you have execution a year to two again case by case before you go into 25 or more years of operation where you have stable uh asset management. So I think if you look at the execution model and making these four pillars meet at the same time there's an additional element where you must succeed which is to bring benefit to local communities and I think there's tons of real evidence as Berka says now we're generating revenue and and there's there's many predictions that AI will generate um a massive boost in productivity. It's going to generate um GDP boosts. It's going to generate new jobs. It's going to make old jobs obsolete, which is really what the first three items are talking to.

There's other elements too that I wish to just bring forward, which is waste heat. We see in our projects, many district heating operators in Denmark having short supply of heat. And you could say 15,000 households are powered today by data centers through their heat, which is maybe a modest number. Um, that's for that's for technical reasons because legacy data centers didn't produce as much heat as today's GPUdriven data center. So, that's going to change.

And then last but not least, you and there's there's a there's a three missing here, but if you look at the data center capacity in Denmark, you'll see a lot of people say, well, they take our energy, but but really facts are that in Denmark, we've stood up more renewable capacity because of these data centers and they consume. And this is a big topic in the US and the primary reason for why many states in the US are now putting a moratium on new data centers because people are worried that it'll take all their power. Uh they uh Trump passed the rateayer protection pledge to say now let's bring our own power uh data center developers and and and that's that's been our strategy from the beginning to team up with those that develop power and have the grid and then make these projects with them. And maybe one more one more point on on that slide which we we didn't put on but basically in all of our projects we have a Danish pension fund as a partner. So in addition to these I would say societal uh benefits we also want that Danish citizens should be making money on the data centers we're building because we're basically you know we're past the point where it's only IT budgets that are getting disrupted.

We're at the point where labor is being augmented. What's driving anthropics growth is, you know, one, it's the B2B solutions, but it's also claude. I'm sure many of you are toying around with a claude agent that's now actually able to do real stuff. So, we're going to take a lot of the outsourced jobs. In Denmark, for example, we have more than 100,000 outsourced jobs that are sitting offshore.

We're going to insource those and they're going to live in our data centers in Denmark and we're going to have the pension funds. So Danish citizens are actually benefiting from from those AI jobs that are living in our data centers. So I think on the capital side it's very important for us that we really focus on local investor base and trying to prioritize the Danish pension funds. Yeah. And that's is not only an the right thing to do, but it's also very politically important because of what Paul said that data centers is a controversial topic.

So if you want to have a long-term successful business where you get prioritized by the TSO, you need to be benefiting or showing benefit to uh to national interests. I'm 100% sure of that. So to summarize uh in five points data sovereignty is important and um and a matter of national security as as Berka has has highlighted. Europe is behind big time on AI models moderately on applications and definitely on chips too. Maybe not so much in the supply chain within chips but on chips but we can win on power and and infrastructure.

Then as Ber also said the flap D primary markets seem to be uh saturated. So the Nordics are stepping in to fill the gap and in particular Denmark. And number four to create an investable product you need four things to meet. You need someone who can develop these data centers and build them. You need the energy and infrastructure.

You need the capital and funding. And lastly the offtake to meet at the same time for a d-risked investment case. And then last but not least, you need to bring some uh economic benefit to the to the societies that you impact. So that's the end of our presentation. Thank you very much.

I go to the next. Yeah, I know. What's um Yeah. So thanks for breaking down a very difficult topic. I think it's still Chinese for many people in the room but uh it was it's definitely made a lot easier understand how you fit in.

Um when we have these challenging topics we invite other experts to challenge other experts and that's why we have invite Adrian who um who advises a lot of families on investment Danish families mostly and one of the key things he is focused on is infrastructure on AI. So we give him the word for a couple questions. I've been told that this needs to be rapid fire. So, I invite you gentlemen to join me for a quick fire chat. And once we get settled here, um I think we're dancing around a very important question here, returns.

What do you expect in the Danish market from data data sensors uh in your products? So, it depends on where you invest, right? So, the the the key point in time is the FID that that Paul talked about. So a and everybody everybody wants to invest post FID when you have the off you have a 20-year lease with with Microsoft and you you have the whole construction budget locked in who doesn't want to so that's the easy part the hard part is where who where does the first 10% or five 10% of the budget come from right and in that fact in that phase you know we can see outsized returns of call it anywhere between 50 and 100%. annually for post FID we're still seeing returns in the I'd call it 40 range 40% annualized um then the other thing that also matters is where do you cut it like there's different types there's collocation uh there's built to suit and then there's a power shell solution and depending on the offtake like some of the customers we're talking to they want to take more of that in-house they say we will bring our own cooling equipment we want to do our own operations we want you to build all of the electrical infrastructure.

So that's okay. Capex is lower uh risk is lower for us long-term contract then we work with call it 25 30% returns but it's it's in that range. So very very high uh high returns. There was a Danish consulting firm that came out with a study suggesting that data sensors would be 15% of the Danish uh electricity demand. Is that a public opinion or is that something you also see from your analysis?

uh sounds about right but I suppose I think it's it's very important our strategy was to go to the private energy companies many of them they have a strategy where they built um production energy production solar wind and so on um and then they had a power toxtake strategy so they wanted to scale those two in parallel they can produce and then you can consume and produce uh liquid hydrogen um but the problem is power 2x is very difficult to make it work commercially. So when we go to them and we say, "Hey, we have an alternative offtake solution for you. You know, it's it's easy." So let's say that 80% of what we're building will probably come from private energy companies is actually not something we're drawing on the grid. And maybe if I can add to that, in in one of our projects, we're talking about standing up 800 megawatts of renewable energy next to the data center. And and there's usually focus on how much will they take?

uh and less focus on how much they'll give which was why I made one of the points something else data centers can participate in in the Danish future grid and the European grid which is let's say having more and more renewable energy replacing fossil fuels is balancing because the grid becomes unstable when you only rely on wind and solar they send electrons through inverters which are really u I'm I'm an engineer too uh like the first speaker uh they're without inertia. So when they stop the grid cannot easily balance down not like a hot oven or a coal fired power plant where you have a lot of inertia but data centers come with big batteries and they come with hundreds of generators if it's in the hyperscaler scale and all of these can participate and help the energy uh the transmission system operators and in solving in solving their problems. So that's how you're solving some of these bottlenecks that everybody within the data sensor industry is talking about. For Denmark specifically, what are the bottlenecks other than the TSO that you're thinking about in your products? Well, I think the main bottleneck is the the grid, right?

as opposed to I think you can't emphasize enough how clogged it is because you know if you're energy net you you you think that people are not just going to pay money if they're not planning to use it but the option if you have a field and you pay let's say put down half a million or a million croner to get a 100 megawatt um application advancing in the queue and then you hope that somebody comes and buy buys your package right that's a good option to have H but many of the applications they are there there's no intention to actually use it. They don't have any plans to develop it and use it. Which means that Energyet they need to solve this clogging, right? That's the biggest problem that you know we we're okay because we have private energy partners. But if you really want data centers to become a thriving e ecosystem, you need to clean out all those free options of people who are not real.

Right? That's that's probably the biggest challenge I think. And maybe if I can add the second biggest that's that's permitting. I was laying out the process uh for a hyperscaler possible tenant yesterday and and there's 15 steps, three public hearings. It's all very transparent.

U Denmark is is good with that. It's all very digital and and you're dealing with people in the municipalities that deal with everything from your driveway to a hyperscaler data center uh through that permitting process and it just takes a long time. So, I think that we could talk about that process and speed it up. Um, I saw an interesting comment this morning that in the Danish elections, no one talks about AI. Uh, they talk about wealth tax, uh, and top marginal tax, uh, and how we can get the rich, but they don't talk about what will most likely change most of the jobs we know today in in in maybe a fivey year time horizon.

So one of the topics that we're here for today is also how to we how to actually pay for this wealth tax in terms of return. So if we can't invest in data centers but you sit in this nexus being a project developer so you know who has the pricing power within this industry. So the second derivative of this who actually has the pricing power um in your products I would say everyone. Okay. So my what I've realized is that the rule of thumb is every everybody's uh margins are like between two and four times higher from the suppliers to the recruiters to the brokers.

So it's not like this. And I you know I believe that that win-win is the best model. Like if if if we go into this and say we want to beat everybody else and and and make them bleed and take everything for ourselves, we're not going to be successful. So I don't see it as a scarcity like if we can crack this the way that we've started there's enough for for everybody to have good returns. So so essentially this extreme amount of demand that there is for AI in turns this forest of suppliers to your projects it doesn't matter the key here is is time or why do they have this immense amount of of pricing power?

Yeah, scarcity, scarcity. But I would also say in terms of the demand, it's, you know, everybody knows that there is going to be a big demand for AI, but if you're, imagine you're uh, it's not like signing up to an office lease where we build an office for you and you commit to, call it a 10-year lease, right? you have to in this case you have to commit to the lease and you have to commit to buy GPUs for probably more than the value of the lease contract and you have to commit to learn how to operate those. So even though the there's a lot of companies that are you know in the process of getting their head around it and making that decision actually pushing the button and saying I'm going to I'm going to sign now and I'm going to put my purchase order in with with Nvidia. Um that's a big decision right?

So much of the offtake in in you know the same way that much of the capital investment is coming from the U US much of the offtake is also coming from the US companies. So the European companies they need to make the decision together with us that we're we're trying to work a lot not only at the I'd say the direct customer level like who's who are the call it the IT or the corporate companies that need GPU power in the future to have their own AI infrastructure but also the applications that they're then that are then running on that software that they're building. Uh so it's a whole like making making the whole ecosystem integrated that's really the hard part. So, I'll uh open up for one question from the audience before we have to wrap up here. Uh and we'll do it the old way.

So, please just raise your hands if you have any any questions in the front row over here. There there's a supply constraint, but there's also um constraint demand because there's very few tenants. And so it's kind of a different real estate because the assets quickly become obsolete. You're very dependent on very few tenants. So you could say the risk is low because there's a huge demand in the operation.

But if something happens to one of the tenants, you could say sort of an event risk regulatory change. We don't want the US whatever it is or how do you how do you consider this sort of strength in in the negotiation between the very few tenants it is data center operates it is super challenging right as Paul said there's neoclouds who are very aggressive securing capacity like really pay top dollar but they could be gone in a year right so do you want to sell all your capacity to them then there's a quoted the Danish corporates who know that they have to get their head around it, but they can't really commit and they're slow and they need to make decisions. Do you want to wait for them? Right? Or you have US companies who like the big tech companies who are all land grabbing, but you also don't want to have only US customers in your data center, but we need them to scale the ecosystem because, you know, they allow us to put in the first order so we can grow it.

And then the Danish, they're coming as well. We we're seeing that now. But the key thing really is for me that everything that is being built is being sold. So so far there's no you know there's no sign that there's any kind of over supply but it it is true. It is it is really hard.

So to wrap up it's a very complex issue we have but there's a massive opportunity within the physical data uh and AI infrastructure piece. So thank you to to Linda for coming up to the stage. Thank you. Thank you.