Critical Load: The race to power data centres

Alright, great. So we're about to go inside. It's you can just about hear it already, a little bit of a low hum, but it would walk through. Oh wow, you get that straight away. That is quite a noise as you walk in here. That's, I wouldn't quite say deafening, but it's going to be very difficult to have a conversation in here. It's coming from all the raps all around us. We're looking at a room. It's full, it's just corridor upon corridor whole racks, lessing cold aisles, hot aisles. That's the sound of a datacenter running. It is rack upon rack of servers and networking equipment and, crucially, the cooling fans which have to run to keep everything from overheating. Before stepping inside, I'd been warned that noise levels within datacenters can reach 90 decibels, comparable to a vacuum cleaner or motorbike. I didn't have my trusty decibel meter with me, but I would agree that even if we didn't get close to those levels when I visited, it is still far too loud to conduct an interview. The datacenter I visited is in London's Docklands and is operated by digital realty. We'll hear from their CIO Greg Wright later in the episode. Recorded on Zoom, not in a datacenter, I'm relieved to say. To discuss the importance of power in datacenter site selection, as we dive into the importance of location, connection, and keeping the neighbours happy. I'm James Limaker with PEI Group and the Infrastructure Investor Podcast, and welcome to the second installment of our four-part mini-series, Critical Load, inside the datacenter boom. Oh, and by the way, we'll be rolling out this series over the coming weeks, but if you enjoy this episode and don't want to have to wait for the next one, the whole series is available now if you follow the link in the episode description and fill out our registration form. It only takes a minute, that's full access right now with no waiting, just follow the link. But now, on with the episode. There's a mantra in real estate that the three most important factors determining a properties value are location, location, location. For datacenters, we're going to hear all about the importance of access to power, but location matters hugely because it directly affects performance, resilience, and cost. The further a datacenter sits from its users and applications, the longer it takes to respond. And while lower latency has always mattered, the rise of AI inference has made it more important than ever. But it becomes totally irrelevant how closer datacenter is to potential customers if the building doesn't have power. Tim Collier, who leads on datacenters for professional services company Turner and Townsend in the UK, notes that access to power is typically the highest priority for developers. So there's not really a single solution to the power challenge and reality. I think particularly across Europe, the preferred solution is still to have a grid connection and backup on site. So that is favored in most markets. And I think what we're seeing and what we're advising is sort of leading developers to make sure that they focus on what they can control in that process. So that is really about understanding the system you're operating within, how to navigate it, to ensure you can get reliable timelines when power can be delivered and set your project up accordingly around that. I think the other thing to recognise in that, it is a specialist area. So clients are increasingly getting specialist help in the power space to help them navigate that. And that can have a tangible sort of benefit to delivering. So we are increasingly seeing that and almost it is a project in its own right for the development of a datacenter. I think the other things that you would have seen in the press and we are seeing increasingly is people are looking at on-site solutions as well, generation solutions. And that's a pretty exciting and emerging space in reality. There are examples of that already, but I'd still say the main source of power for most datacenters is grid connections. But I think there will be, you know, the level of investment in the sector, the need for power. I think there will be some quite exciting developments over the years ahead around those on-site solutions. You know, you look at things like SMRs and hydrogen solutions. So yeah, I'm pretty sure we'll be seeing more and more of that over the years ahead. You'll remember Alexey Teplikin, managing director and head of capital markets for Blue Al Digital Infrastructure from our previous episode. When I caught up with him at Infrastructure Investment Network's Global Summit in Berlin, I asked him, "Which factors go into deciding what makes a good location for a datacenter?" Again, it's power. Yeah, so at the level of abstraction, the way we think about it is it's the combination of land, power and people. And why is it those three things? Power for obvious reasons that computer needs power to switch on. Land because datacenters benefit from an economies of scale concept. If you're able to build a large datacenter, it is cheaper and more efficient than building 10 smaller datacenters. So land is important to be able to put this large datacenter in a place. However, you also need to think about the latency of the product that you're offering. If you are leasing the datacenter to a user and that user provides a product that people are using, then it's very important what the latency is, how close it is to that city. You cannot be too close, where the land becomes too expensive, you cannot be too far, where the land can be very cheap, but it's too far away. Imagine you open up your email app and it takes 30 seconds to load, you're going to go to a different email app. So that latency is very important and then the people ultimately is critical because they're the people who are using data. So the first thing Teblik and Listed there was power. We're going to come back to those other two, land and people, but let's focus on power for now. He described this as being crucial for obvious reasons. But some of this is also less obvious than you might think. Alan Pritchard, from technology consultancy far point, is quick to note that the datacenter is so much more than just a building. When I put to him the same question I put to Teplicin, what makes a location attractive for a datacenter? He agreed that the answer is power, but he also added that the type of power is key. Yeah, I think the big one, which is front and center and everyone's conversation for the obvious reason is power. So datacenter is in particular large AI datacenter needs considerable amounts of power and that power needs to be reliable. I think the thing that people don't quite understand is, yeah, we need power, but it's not just annual power. It needs to be reliable because the datacenter runs 24 by 7 by 365. It cannot stop. If it does stop, it's incredibly expensive and restarting it again is is also incredibly expensive. So not only is power needed, but it's a right type of power. It has to be a reliable source and normally data sensors will have multiple sources of power. So people think, well, we'll just build a gas power station beside them or we'll sit them beside a nuclear power station. That's great, but these power generating assets, they don't run 24 by 7 by 365. They have downtime, they have maintenance windows and need centers. Don't really have that. So when you're building a datacenter, you know, it's reliability, consistency, reliability, diversity of supply that goes for power. That also goes for connectivity. They have to be connected. They have to be connected incredibly well to multiple operators, to multiple other datacenters. So if one fails, they just carries on as if nothing had ever happened. And I guess they're going to last of the three major criteria that are super important for a datacenter is land. And that land has to have the right planning permission, the right permitting. It has to have security. The environmental aspects of that lands are particularly important. I've seen some datacenter projects that investors got really, really nervous about proximity to water. Particularly flood risks, and I know it's quite easy to calculate or you can calculate the flood risk of our particular asset and how close it is to a body of water or surface flooding. But nothing gets investors more worried, particularly if they're investing in a very important data. That said, water is very important for datacenters. The digital reality facility at which we started this episode used a chilled water cooling system. The firm operates another datacenter around the corner from it, which is one a green datacenter of a year reward. In large part, because of its use of a dock and river cooling water system, drawing in water from a river tent. Efforts to minimise environmental impact and water use are increasingly important for datacenter developers, and certainly factor into power considerations. But sometimes just getting power, any power, can also trump those efforts. An April report from technology news site Wired found just 11 datacenter campuses around the US between them have the potential to create more greenhouse gases than Morocco. The datacenters, built for the likes of open AI, Meta, Microsoft and XAI, have the potential to emit more than 129 million tons of greenhouse gases per year. This is largely behind the Meta power, built to bypass the grid. In 2024, XAI made headline. by setting up gas turbines to its data center campus, Colossus 1, in Memphis, Tennessee. This year, an XAI affiliate was given a permit to use gas turbines for a second campus, in South Haven, Mississippi. With widespread community opposition, the NAACP has filed suit against XAI. Why its calculations estimate the gas turbines of these campuses could generate more than 6.4 million tons of CO2 equivalents at each site per year, which would be enough energy to power one and a half million homes. We're going to look at sustainability more later in the podcast, but we're running ahead of ourselves. Let's get back to choosing for right location. For a long time, data centers have clustered in the US in places such as Ashburn, North Virginia, Dallas, Fort Worth in Texas, or in Europe in the so-called flap D locations of Frankfurt, London, Amsterdam, Paris, and Dublin. Major Asian cities such as Singapore, Tokyo, Hong Kong, Sydney and Seoul have also been dominant. But that is starting to change. Here's Collier again. I would say constraints in primary markets are certainly driving increased interest in secondary or new markets for sure, but it's not a blanket approach. You know, it's very much on the case by case basis, and I think that there's probably two key things to look at really is one is geography and the type of compute you're using or need. So typically, you would see cloud data centers are typically still need to be located near to their users. That's a constraint around location, whereas AI facilities particularly training are a bit more location agnostic, so you don't have the same latency issues. So you can look wider afield, and the other one is speed, right? So how quickly you need something at the time and you need it, and that may also dictate where you build. So I suppose I'd also say that secondary markets and new markets can bring additional risks as well. So there's benefits to them, but you may not have, for example, as a mature supply chain in a newer market, as you would elsewhere, which can start to introduce delivery risks. So I suppose my summary would be, it's not necessarily a shift from primary markets, the secondary markets, it's just more of a broadening of the landscape and people looking at different markets depending on what their specific needs are. I said earlier with here from Digital Realty's CIO Greg Wright. He's been at the company for seven years, after having been an investment banker working with the company for double that time. Digital Realty has more than 300 data centers globally, including the one in London where we started this episode. There are six continents, more than 30 countries, more than 50 cities, and I've been listed on the New York Stock Exchange for more than 20 years. In his years at the firm, he has seen the hunt for power change, how attractive different locations are seen to be. Well, look, I think when you look at the locations again, we're finding is these markets right now, the data center markets are increasingly constrained. I mean, what I would say we're finding right now is power availability, land, zoning. These things are all tightening at once, nivism if you will, right? So you got these core tier one markets I'm talking about, they're very crowded, power's hard to get there. There's not a lot of availability. If you look across the globe at major markets, whether you're talking about an ask-burned in Atlanta, a Dallas, a Chicago, Northern California, whatever it may be, Frankfurt, London, Amsterdam, Singapore, Paris, any of those markets, right? They're incredibly tight. So these things are all tightening and those core markets would also are crowded with very little availability. Now what we're starting to have is a little bit of a barbell market dynamic, if you will. So we're starting to see as some of these customers that need power, especially for these AI training use cases, are being forced to go to secondary markets or even what I would call even more tertiary markets to get power. EQT Parnayian Vesselic has also seen a broadening of viable locations. The old flap-deeb dominance in Europe has certainly weakened, for example, not least because it's hard to source adequate power. So I think much of the growth is now happening in Scandinavia where there is much more power and not even sustainable power, which is a big advantage and still a big focus for most hyper-scalers and for us. I think that eventually can be a lot of growth in, you know, power markets that are good for solar, power, Spain in particular, Portugal, but it's limited. I mean, if you have any power in the flap markets, you're going to be able to sell it, but it's just not that much available. Maybe Paris is the one exception day and there's no clear obviously power in France to some extent available and being freed up for data centers. In the US, KKR's Kathleen Lawler notes that Tier 1 markets remain popular. Just as flap-deeb hasn't gone away in Europe, but once again, the search for power is bringing new markets into play. It's interesting. I do think here in the US, the hyper-scalers still want to be where they want to be in terms of Tier 1 markets. We certainly do see and hear a lot about other marketplaces where people are building really big projects that I think is driven by power and energy availability, but that doesn't mean that they're not also still building in these Tier 1 markets. We see that through our data center portfolio companies, but we also see it through our power and utility portfolio companies where electricity demand is skyrocketing in those areas because data centers are coming online, and it may be a hyper-scaler self-building a data center rather than a developer, for example. We think we have a pretty good pulse and insight into where these are getting built, and certainly I think the hyper-scalers still, they want to be where they want to be, and there's spillover that's going to other markets as well. In Asia-Pacific, the picture is more mixed. For region is, of course, both large and diverse, so one size does not fit all. Dev Gupta runs the private funds business for a data center section of capital land. The Singapore headquartered firm has some hundred billion dollars in funds under management. Gupta's markets are typically Europe and Asia-Pacific, including Australia. He too has seen a significant focus on secondary markets. We used to also have our own version of Flappedi in a pack typically used to be Sydney, Hong Kong, Singapore, Tokyo, but you see that the Native growth in those markets versus another market has been very different over the last three, four years, and if you see the potential capacity build up over the next 10 years, it's knocked those markets, it's other markets. So clearly, the likes of a Johor, which you would say, as an adjacency to Singapore, but has almost two and a half, three gigawatts of undeveloped or under construction capacity, Melbourne as an option to Sydney, Osaka as an option to Tokyo, Mumbai has become itself a three, four gigawatt market, and now we're seeing Southeast Asian markets, Thailand, other parts of Malaysia also growing rapidly. So whenever there is a lesser of a latency required workload, more a high density workload, you would see that the secondary markets do well. It's not just a power issue, it's also a construction cost for megawatt issue. Southeast Asia, India will definitely be able to compete very, very significantly to an australia or a Japan where construction costs are almost half of the developed markets. So that is also leading a lot of workloads to these geographies. Furthermore, data centers aren't just being built in different cities, but also in different parts of the same cities. You typically saw soil, let's just take that example, gas on where all the internet exchanges were traditionally and where all the data centers came up, obviously no longer there's any power or land available, but you can go 25, 30 kilometers south to an Ansan or a Hanam and find power in a land. So it's still Metro Seoul, it still services all the workloads that you would expect off that area, but it's just a different secondary market. So does this mean new locales are in the ascendancy and the old ones are on the way out? Here's Andrew Thomas, senior managing director, a real assets manager, Stone Peak, which has $88 billion in assets under management. As you might suspect, he suggests it's too simplistic to say there was a definitive changing of regard. Secondary markets are definitely emerging as more attractive than they have been in the past, that is for sure, and it is all about fast tracking the time in which the largest consumers of data center capacity, the hyperscalers, are able to bring capacity online. You can do that a lot quicker in secondary and tertiary markets than you can in tier one markets. I think that Northern Virginia is going to continue to be the largest data center market in the world for the foreseeable future. It is constrained by the amount of incremental power capacity that you can get to it, obviously. But what we've seen is that area has just continued. to expand and what we consider to be the broader North Virginia market has continued to grow. When you think about just data gravity that you have with all of the deployments that all the largest customers in the world have made in that market, as long as we can bring power to that market, I think there's going to continue to be extremely strong absorption there. We've been dancing around power as a bottleneck all episode, and there's a simple reason. Data center energy demands are controversial. We've heard a few times now from KKR's Kathleen Lawler. She leads her firm's power and utilities infrastructure investing. In a research piece recently, she said the ultimate limiting factor for building data centers isn't capital, it's power. It's become the most important piece of a puzzle. You know, three years ago, it wasn't. Building data centers was all about location and having the right land positions, and that began to change as I think data centers reached this tipping point in certain locations where they had exceeded the existing capacity of the electrical grid at those locations, and all of a sudden you could not build another data center without having access to power in some shape or form, which is usually connectivity to the grid. Not only is power limited, it's closely guarded. Private markets bare mouthed Brookfield asset management, with over $1 trillion assets on the management, was forced to accept defeat early this year when its compass data centers business had to pull out of a planned Northern Virginia digital gateway project after a zoning ruling went against them. Blackstones, QTS Realty Trust, also involved in that project, announced it would appeal against the zoning ruling, but in July, it too terminated its involvement. The decisions of First Brookfield and later Blackstone to pull the plug followed years of local opposition, including significant concerns about energy use. The digital gateway project in its entirety should have generated up to $25 billion of investment and when built, taken up over 2,000 acres, with up to 37 buildings providing more than 20 million square feet of data center space. This is far from an isolated incident. Ireland's grid operator effectively enforced a data center moratorium in 2021 because of power constraints. Singapore paused developments around the same time for similar reasons, while in the US, up to half of planned data center development this year is thought to be delayed or cancelled. While working on this podcast, private credit firm Lakeland Equity Group's $1.6 billion proposed data center in Cleveland, Ohio, was rejected by the city. The facility would have required 150 megawatts, comparable to 100,000 homes, and the energy demands are understood to have been a significant factor in that decision. Two months before that decision, in March, Senator Bernie Sanders and Representative Alexandria Acacia Cortez proposed legislation aimed at pausing the construction of new AI data centers. It has been put to mean more than once that this is something of a stunt, which is never going to lead to legislation and can therefore be dismissed. But if it is a stunt, it's done because there's an audience wishing for it, and that sentiment shouldn't be ignored. The sense is that all this power is going to data centers and there will be nothing left for everyone else. But as is so often the case, the truth is more complicated than that. He's lower again. It's definitely a huge consideration, and I think it's really important for hyperscalers, for data center developers to pay their way, to pay their fair share of the build out that's required to meet the demand that they're creating. If I just take a big step back, the way your utility bill in the US works is there's a certain amount of built infrastructure and fuel costs to generate and transport the electricity to you as a customer, whether you're a data center or a manufacturing facility or a home. We have a more or less socialized cost system here in the US for electricity, and so the way your utility bills calculated is how much cost is there, divide that by the number of kilowatt hours that are being consumed, and that that's the utility bill, and then it's allocated certain ways. But I think something that gets lost sometimes in the conversation is the data centers are contributing to the denominator by consuming a lot of electrons, and so they pay a lot because they consume a lot. And for me, that's actually a huge opportunity to ask these high credit quality customers who are investing big dollars to help us upgrade our electricity system and help pay for the upgrades to that electricity system while making sure we're protecting existing residential customers, for example, from increasing cost risk. And I think a lot of states and utility commissions and utilities are really paying attention to that and doing the right things. You know, they're implementing rules and tariffs to make sure that data centers are paying their fair share that they're going to be around for the long term. Of course, there's still a lot to figure out, but I do think people are paying attention to this issue. For example, one of our investments at KKR is on the transmission side in Ohio and Indiana. Both of those states have passed special tariffs that apply to large loads like data centers to make sure that they are truly paying their way. And we think that makes a lot of sense. And as we build data centers, that's something that we're thinking about is making sure we're making the right commitments to support what needs to be built out. In our next episode, we'll dig in to why these concerns may be overblown, and what developers are doing to solve the power problem and earn their social license to operate. But that's all we have time for today. All episodes of this series are available on infrastructureinvestor.com and all the usual podcast platforms. This episode was edited and produced by Julie Begar and Eric Fish and had additional reporting by Edward Cleaver. I'm James Linnecker. Thank you for listening. And don't forget, if you enjoyed the episode and don't want to have to wait for the next one, the whole series is available now. Just follow the link in the episode description.