The data centre boom

The US, China, and now Europe are seeing a surge in data centres as big tech firms and other businesses develop their AI models. But, will these data centres cause the dramatic rise in power consumption that many are predicting? 

In this episode, Richard goes deep into the subject with data centre experts from RISE and Greenscale to understand who the tenants of these data centres are, what challenges the tech and energy industries need to overcome together, and how data centres will continue to evolve with the exponential growth in AI. 

Presenter: Richard Sverrisson
Contributor: Alina Trabattoni - Italy Correspondent, Montel News

Guests:
Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden)
Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale

Editors: Bled Maliqi
Producer: Sarah Knowles

Transcript

Richard Sverrisson: 0:05

Hello listeners and welcome to Plugged In - the Energy News podcast from Montel, where we bring the latest news issues and changes happening in the energy sector. This week we turn our attention to a very hot topic, a European energy markets, data centers. This coming together of tech and energy is throwing up some very big questions. And the main one being will data centers, be as truly power hungry as many are predicting. I'm going to be speaking to two data center experts to really get to the heart of this topic. But first, I'm joined by our Italy correspondent Alina Trabattoni. A warm welcome to the podcast, Alina.

Alina Trabattoni - Italy Correspondent, Montel News: 0:45

Thank you, Richard.

Richard Sverrisson: 0:47

I'd like. If I may, I'd like to ask you just to illustrate to our listeners the scale of growth, the data centers that we've seen in recent years.

Alina Trabattoni - Italy Correspondent, Montel News: 0:55

Well, to be honest with you, Richard, the numbers here are genuinely staggering. Industry analysis shows that European data centers are currently consuming between nine and 10 gigawatt. Of power. So that's roughly equivalent to let's say 10 large nuclear power plants running 24 7. But here's the kicker. Some analysts including some ISIS ones, for example, project that this will triple to 35 gigawatts by 2030. So put us in electricity terms. Some 55 to 80 terawatt hours currently being consumed in 2023 alone, this consumption will potentially jump to over 150 terawatt hours by the end of the decade.

Richard Sverrisson: 1:43

Those are astonishing numbers, Alina. Why I mean. Why are we seeing such a boom at the moment? What are the key drivers behind this growth?

Alina Trabattoni - Italy Correspondent, Montel News: 1:50

A lot of people point to ai and you know, I was talking to some analysts earlier this this week asking them, is AI really the main driver? The only driver? And one of them told me, well, yes, certainly it is a major catalyst, but it's far from the whole story. And, she told me to think of AI as a, as the latest wave in a digital to Tamiami that's been building up over a decade. So the fundamental driver in the past decade has been cloud migration. European businesses have been moving their IT infrastructure from basement server rooms to professional data centers accelerated this. Trend has been accelerated massively during COVID when remote work became essential overnight. And basically every Netflix stream, every Zoom call, every cloud backup contributes to this demand. And then you also have the Internet of Things explosion and edge computing requirements. So things like smart cities, connected manufacturing, autonomous vehicles. These are all generating continuous data streams that need. Processing close to where they're generated. And the push for ultra low latencies. Also driving demand for distributed data centers across Europe. And it's also interesting and worth considering that European regulations like GDPR have also played a role. Encouraging and in some cases, obliging companies to process European data within European borders rather than shipping it across around the world to other data centers. And of course lastly, there is a sustainability angle because data centers consume evermore energy and they're also enabling digital solutions that can reduce emissions in other sectors.

Richard Sverrisson: 3:43

Alina, thank you very much for setting the scene there. So the big question remains, will data centers cause the dramatic rise in power consumption that many are predicting? I know my guests have much to say on this subject. I'm pleased to be joined by Tor Björn Minde, Head of ICE Datacenter Unit in Stockholm. And Liam Newcombe, senior Vice President of Product, Energy Strategy and Innovation at Data Center Platform GreenScale. A warm welcome to the podcast both. Tor Björn, I'd like to start with you before we get into sort of future energy consumption and various forecasts. Can you explain to our listeners what the primary functions of data centers are at the moment and how much energy do they need to power them?

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 4:27

Yeah, so the it's powering our daily life. We start in the morning when the alarm clock goes off in our phone, there is a tick in a data center somewhere, and it goes on during the day. We use it all day long at work privately. Et cetera, et cetera. It's the phones and the laptops and the, our computers that, that drive the data center, use the cloud in the background. And the most of the energy today goes to the cloud data centers Google, Microsoft Amazon or Facebook or eBay or those kind. Estimation is 500 Terawatt hours. A little bit more maybe right now in the globe. So that's, what's going on right now? It's, it must, it's a must. If we turn them off, nothing will work for us.

Richard Sverrisson: 5:17

So we have to, we, there's a must run. These data centers aren't there. And Yes, but talking, just before we go onto the next question, could you explain where, I know that there's a big concentration of data centers in the US and also in China, but where are they concentrated in Europe?

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 5:31

In Europe, we have the big cities since most of the usage today is low latency. So you need to be close to the customer. So it is Frankfurt, London, Paris, Dublin Amsterdam, those big cities. And the Nordic is of course, Oslo, Stockholm, Helsinki. But yeah, so that, that's where the most of the data center. But the more cloud usage that is not delay sensitive, you don't need that fast response. You can put them anywhere else a little bit further away and. For example, the, one of the Facebook's or METAs data centers are in Luleå, up in the northern part of Sweden for off because they can do work in the background and so on. And that, and AI will be part of ai, the training of AI model is that kind of compute that can be anywhere really.

Richard Sverrisson: 6:25

Okay. Excellent. This is very helpful. I'd like to focus now on looking at some forecasts for power consumption because of the a boom potential boom in data centers. What do you predict the increase in power will be as, as we see a lot of these, the increase in cloud services and certainly in ai Tor Björn.

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 6:42

There's a number of predictions out there from different institutions and, I usually go with the International Energy Agency. They have the most balanced view. I think that's my belief. There are some that is really off too high. I think they're they go on on. Something is like just increasing and no optimizations going on. So those predictions are off the chart? I think so. International Energy Agency says. Two times until 2030. Three times until 2035. I think that is a balanced view. It is a lot in, it's a big increase, but it's still a balanced view on it.

Richard Sverrisson: 7:20

And that's a global perspective. Global.

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 7:21

Yeah, globally. Globally. A little bit more in the US now in the beginning and China but Europe is catch will catch up some of it at least.

Richard Sverrisson: 7:30

And Liam what do you share this view or do you side with a sort of ia or, I know some TSOs have come out quite bullishly in terms of their predicted power demand rise. What's your view?

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 7:40

Yeah, so it, it varies around the world based on the market and what workloads are going to be run. So we have our underlying growth in all of those cloud services that Toon identified. So all of us using, whether we know it or not aWS Azure, et cetera. And then we have this significant new demand coming in from ai. A lot of the model training is occurring in the us. Jonathan Kuey was interviewed recently for people who dunno, he worked for LBNL, did the EPA report Congress on data centers back in 2012. He's very knowledgeable in this topic. His numbers were around 2% of the US grid. Was data center in 2020. By 2023, US numbers had that past 4% and the forecasts are to maybe 10% of the total US grid in five years. So 1% of the US grid per year, which is vastly outstrips the actual demand growth the grid has had in Europe. Our investors are working on numbers around 20% compound annual growth, but in Europe, we're typically seeing. Things that are closer to use of AI than model development, as our energy tends to be more expensive than North America or China. And as Toon said as well, the huge range in forecasts. We can't take the the transmission distribution operators connection queues because the same project is in. Those queues multiple times as people bid to obtain the land and sell it. Just like a renewables project, what you have to do is take the survival or success ratio of those projects, and that's maybe one in five or one in 10. It's also probably not wise to take NVIDIA's revenue forecast. Some of the banks have and assume that every chip Nvidia has ever made will be out in the data center and running'cause they have a finite physical and economic lifetime. I'll move on somewhere sensible in the middle. A, an approximate doubling by 2030 seems to be a sensible and physically feasible number.

Richard Sverrisson: 9:47

That makes sense. But can you just explain as well to our listeners, maybe go back to some of the basics here. Who are the primary customers of data centers and how do they differ in terms of the locations that they're after? What, why do they choose certain locations over others?

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 10:02

So again as Tor Björn pointed out, we have the typical hyperscale operators who have the real revenue close to customers. So the backend for your streaming game or for the website or application. Hosting E-Bay, things like that. They tend to be located in the domestic market where the customers are. And in particular, if you're Microsoft and you supply a lot of services to governments or large corporates, you need to be for both latency and data sovereignty reasons close to and frequently in country. Those customers. So we have those data centers that have real targets for where to be. We. Then again, Luleå in Sweden is an excellent example for Facebook. They don't have that requirement to be close to the user because it's not super latency sensitive and there's not a big, regulatory or legislative boundary for them there so they can go somewhere, which is more attractive from a power cost perspective. And then we have the very large growth that's coming up in artificial intelligence, which I would break down between people who are developing a model and actually inventing ai. So o open ai, the teams at Facebook, et cetera, who are actually creating the new models. They are the ones that we see the press articles about half gigawatt data centers for. They're typically being placed in North America, increasingly in China by the domestic growth there. And then in Europe, what we're seeing is the growth of people who are actually doing business and getting paid for ai. So real model retraining for specific things such as image recognition and processing to optimize battery electrode manufacturing. Actions like that or the final inference where you would use chat GPT or get Google search results, et cetera. And those are really a new layer on top of that old corporate colocation and corporate cloud and consumer services.

Richard Sverrisson: 12:04

Now that's a very interesting distinction on why, you know, maybe we'll come back to cost in a bit.'cause I'm sure that has a lot to do with, as you referenced those earlier as well, Liam, about the energy costs in, in North America and China versus some parts of Europe. But Tor Björn, I if we can continue on the subject of artificial intelligence. You have talked about the improvement of AI models and how they could bring down the cost of energy. Could you clarify what could, explain what you mean here?

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 12:32

Yeah first I also wanted to say just this growth in data center, the energy use it, it comes from the, there is a gap between the re request or demand for capacity and the optimization going on. Because we have optimization both in algorithm models et cetera, and in the chip sets. Nvidia, Intel set AMD chip sets are improving. The good old more slow is there they do a lot of improvements but the demand is growing. So there was some estimation from. I saw from a researcher, I said it was like 4.6. We can say, round it off a little bit. Five times demand increase, but only two times improvement increase. So you have a gap there. And that's why we need more data centers and more energy and the improvements. Then if you go to that, there are two parts, as I said the algorithms, and it seems like at least it's not very old technology. This. Remember, it's only the second year almost running it. So two and a half years of operation of chat GPT. But the old optimization that has been going on shows a 10 times increase, 10 times decrease in, in energy use in the algorithms and models. At the same time, the Nvidia chip sets shows an a, a very steep increase. The first Chatt PT model was trained in a hundred days using the latest chip set which H 100, it takes 50 days. And the latest now. Latest. There is just a few places in the world, the GB, 200 racks, it take 12 days to train the same algorithm and so on. It'll continue these improvements and the estimation is that sometimes in the future with the VR 200, we will see one and a half hour time to train the same. So we have improvements in the hardware. And improvements in the algorithms. So all this together helps a lot of course. And, but still it is a gap to the demand.

Richard Sverrisson: 14:41

Absolutely. I wanna come back on to the, demand and what kind of. Power will be needed to power these data centers. But Liam, if I can ask you for a second on, and in terms of costs because I think you recently acquired a data site in to stud in, in, in Southern Norway. Why did your company, I know you, you don't wanna reveal any confidential information here but why did you choose that location?'cause that's where the highest prices in Norway have traditionally been, or certainly over the last few years. Power price is, that is.

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 15:09

Indeed, yes. We've acquired the rights to develop. So there isn't a data center there yet. We will be building the data center and then leasing it to one or more tenants. It's relatively large, it's 300 megawatts. Five years ago, 300 megawatts would've been huge. Now it is one of many projects at that size. Norway is attractive much of the Nordics for a number of reasons. The first is baseline cost of power in Norway is good compared to that sort of middle band through Europe. The. The levelized cost of energy of the hydro, the high capacity factor, wind, the existing very good transmission networks, tho those all combined together to give Norway a very good baseline power cost. In many ways comparable to a lot of the North American market. We see the same in the Iberian Peninsula. So for a, an AI or social media. Site moving into the Nordics is very attractive also the Nordics are well progressed and quite mature in terms of the regulatory, the planning, the visibility of grid connections. So Fin Grid, for example, has the most fantastic information portal that will simply tell you what is available at what grid supply point, which takes a huge amount of the cost and guesswork. Outta the market. So yeah, there's a very small advantage from a climate perspective that's frequently over egged. But yes, it is cheap to cool things in Norway as well. So very good baseline power cost. We could have gone further north. It would've been slightly cheaper on power, but it would then have been more expensive and higher latency to get the network. Connections. And it would also have been longer travel times for whoever our tenants end up being to get to that site. So reasonable distance from an international airport is still quite significant for us.

Richard Sverrisson: 17:06

Exactly. And that's where maybe northern Sweden and Northern Norway, maybe they're are, they're at a greater disadvantage in that sense.

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 17:13

I, I think you need to be building, a significant single campus for it to be worth the end customer taking on the additional cost and difficulty of going to a remote location. And again, Luleå is an excellent example. Meta went and built at the time, a very large campus there and it made economic sense for them.

Richard Sverrisson: 17:31

Absolutely. Tor Björn, Liam mentioned Fin Grid here, and I just wanted to talk a little bit about, grid issues. It is, it's. Is the grid expanding or developing in the, at the same pace that, or in the pace that you think is as, is required for the boost in data centers that it, that is coming.

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 17:48

I'm not a grid expert but looking at it from the distance of a data center research you can see that maybe it's. They're keeping up the speed, but they're working on it. They are really trying hard and I know that they have collaborations, for example, in Sweden, I know they collaborate between the data center operators and the energy providers. So for example, the Microsoft Web deal about 24 7 hydropower. Or renewable power. So they are working on schemes to to improve the situation. And I know now that, there is a collaboration between the grids in Nordics and now the data center industry is invited to the table of that discussion. So that, that is very good. Of course Liam, Isabel Keblin. She's the one from from the Swedish data center industry that is on, in that, on that table, together with me voices and Norwegian or Danish. A data center representative. Anyway, that's good because then we the issues can be raised and discussed and the grid providers can be more prepared and work on the right things et cetera. So yeah.

Richard Sverrisson: 19:01

No, I, absolutely.'cause they must see you guys or the data centers as big potential customers as maybe traditional heavy industry is declining somewhat a across Europe, then the data centers is gonna be a big potential customer.

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 19:14

Yeah, there's a big site in Bohleng where there was a paper mill very prepared for a big customer. And of course all things are prepared. And the idea wa I think was that the, one of the battery factories of North Wal was going to be there. But as you, that it didn't go well. This one is open. And I think one of the, hopefully one of the European gigafactories AI gigafactories will be placed there. Because it's a perfect site for that type of model training for European customers and European organizations

Richard Sverrisson: 19:50

and all the grid connections are there. Yes. Presumably. Yes. Yes. Most importantly. Liam, I'd just like to turn to you in sort of like in terms of procuring power, and, and a little bit about how, the long term versus spot, et cetera. Now some big tech can obviously afford PPAs. They have a very healthy balance sheet, so that's not a problem in probably in signing PPAs. But is this kind of, you know, an issue for the smaller customers who of, smaller data centers, who don't, who aren't big tech?

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 20:16

So I, I think that is quite well connected to the previous question data centers. Now that we're a significant part of the load, I think the industry is going to have to move to participation. With the energy industry we're no longer just a customer. So we, we are very actively looking to co-locate our data center assets with either a significant grid asset or a renewable energy generation or storage or heat recovery site. And Tor Björn mentioned Vattenfall earlier. They've done a lot of that Stockholm data parks, for example. If we do that and collaborate effectively with an energy industry partner, I think we're in a position to give those medium scale colocation providers enterprise customers and the neo cloud AI providers who don't have the financial firepower of the hyperscalers access to much more intelligent. Energy products. So we are very actively working with partners on both of our sites to understand how we can package a smart energy offering to the data center tenant without requiring that tenant to become a commodities trader or energy expert in their own right. One thing that has continued to strike me as I've learned. The very little I know about the energy markets is that we use many of the same words, but that the two markets really don't understand each other very well. And to a large part, there is a translation issue that needs to go on initially.

Richard Sverrisson: 21:54

So there's miscommunication between the two sectors, would you say?

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 21:57

I think there's a very real opportunity for people to step into that bridge. And we can make the large data center loads rather than something that is a constraint and an issue for A TSO into a real opportunity for a renewables developer or somebody else in the supply business. Between us, we should be able to make that a net positive for the TSO as well.

Richard Sverrisson: 22:24

So it's win, win in a way,

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 22:26

if we do it right? Yes.

Richard Sverrisson: 22:28

Yeah. Yeah. And just a question on the tenants potential tenants. I presume that it's clean energy, they want renewable energy. They're not looking for, for, gas fired, generation or diesel backup generators or something.

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 22:42

So yes, we still have backup generation except where we're asked to put that on the network by the TSO to back up their power. We use that as little as possible. It is emergency backup. There's an increasing trend. So air grid, for example now require data centers to put their backup generation effectively into their, balancing market so it can be called on in emergency. So in, in terms of the power procurement, we see a rush to power largely in North America. So a lot of people in states like Texas are building gas turbine fired islanded. Data centers to get to power in the fewest months possible, because right now we have a land grab going on in ai. In Europe we see a lot less of that. And yes, you're quite right. Most of the large operators are buying at least annually matched guarantee of origin. To cover their load. People like Google actually have a dashboard which will show you to what extent they match even down to hourly between their procurement and their consumption. And I think shaped products are going to become more popular. The question is, at what rate the price premium for an hourly matched shaped product will develop in the market? And that very much depends on the tenant.

Richard Sverrisson: 24:00

I think that's a very interesting discussion as well Liam because some of the big tech companies, they want hourly matching and others are very happy with the annual system that that as it's working now. So that, that would be quite an interesting situation to see how that develops in terms of the guarantees of origin, the how that evolves. But Tor Björn, can I ask you in a sense. You've talked about North America quite a lot and also mentioned China, and obviously the development is continuing at pace in these countries or continents. Is it does Europe risk falling behind?

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 24:31

Yes, there is a high risk and the big companies in Europe needs to step up. I think. But as we have seen the Euro European Commission at least have put forward a a scheme gigafactories trying to support the European industry to develop and we'll see. Maybe that can help that the next generation, at least of big scale AI data centers, or a couple of them are in Europe and support the European industry. So that's, that's what we have to see. We of course have the AI factories and that currently are default. We have one in Sweden. There is a number of them around Europe. So government backed, and we also see some movements. The big like Erickson, the berry owners swear they are now building one in in, in Sweden a big cluster for training, algorithm training models. So yeah, it's happening. But of course they're global companies so they will go where? The power is cheap and so on. Exactly as well. In us I think many people there they believe Ericsson is a, is an American company, it's a, they're so big in America as well so yeah, it's an American company. Oh, yes. Swedish. What?

Richard Sverrisson: 25:46

Yes, I remember I, I used to have an Ericsson mobile phone all, many decades ago. Yeah. Yeah. So I think, I think that's an important point Tor Björn. But, and Liam do you think the current regulatory framework is fit for purpose? I see. The commission or, is looking to provide, some kind of an industrial power price at a certain level to. Port industry that's obviously happening for traditional large industry in Europe. What kind of changes would you like to see made in the reg regulatory framework to, to support data centers and the growth of data centers within Europe?

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 26:19

So one thing that's quite inconsistent as we go around Europe is the extent to which you can self consume or private wire to a generation asset. That is very much country by country. At the moment, and I think it, it would actually be very useful if that was made more consistent and if we were given a clear roadmap to at what point we expect each of the countries to standardize onto what whatever the commission or Entso-E decide is a target there. I think if we look around the market, we see plenty of operators in the energy space trying to collaborate more directly. With data centers in some cases. So we, we have Ipetrola in Spain and Azura has just set up their new venture Tillium, a friend of mine is working at where energy companies are actually building data centers as their owned load to connect to their renewable assets. I think there is a real risk of large industry players being able to take advantage of the slow moving regulation there. And that may significantly impede the sort of rate of innovation. And other players. Then we have country by country assorted structures for taxation around energy. Data centers, as you said, are an energy intensive industry from the perspective of the data center operator. Because typically the party operating the data center is not receiving the margin from the IT services. Delivered from it. What, whilst it may look like it's easy for data centers to pay large power bills. It's not necessarily the case. But on. On the jobs and employment front, I think it's also quite important for us to look at what is really feasible e economically and the model development. So creating the new models that is concentrating in a couple of markets at the moment. We. We absolutely should not underestimate the number of smart people in China and the rate at which they can innovate. I think deep seek was a slightly surprising lesson to much of the market. But equally, America has no shortage of very smart people researching. If Europe can't catch up on the model development perspective, I think the next question we ask is, okay, but model dev development is investment in trying to own a model. What about the next stage where we actually turn it into a business? And I think this is where Europe is much better, is as a service business economy. Do we have do we want to support companies such as polar on who have. Spun out of Imperial College. The, I mentioned they do image processing for optimizing things like battery factories. Do we actually want to concentrate on people who are taking those AI models and turning them into economically valuable businesses? So real world things with real customers paying real money for. And that might perhaps be an industry supporting approach for Europe to take that's more feasible for us.

Richard Sverrisson: 29:28

I think that's a very good point indeed. On that note I'm gonna, I'm gonna have the end of the discussion there, but thank you very much for your insightful views and opening us to the complexities, but also the importance of data centers. So Liam and Tor Björn, thank you very much indeed for being guests on the Plugged in podcast.

Tor Björn Minde - Head of ICE Datacenter Unit at RISE (Research Institutes of Sweden): 29:44

Thank you for having us.

Liam Newcombe - Senior Vice President of Product, Energy Strategy and Innovation, Greenscale: 29:46

Thank you.

Richard Sverrisson: 29:47

It's been an excellent and insightful discussion. I hope you agree, listeners, and thank you for tuning in to this episode. Our podcast episodes are released every Friday. For the latest news from Montel, please visit montelnews.com and you can follow us on LinkedIn, Bluesky, and other social media channels. See you next time.