The AI revolution: How infrastructure is adapting

Welcome to our Private Markets 2030 special mini-series, part of PEI Group's wider editorial initiative exploring how private markets are evolving as we move through the second half of the decade. Drawing on insights from PEI's editorial teams, alongside leading fund managers and investors, the series takes a closer look at the forces reshaping the industry, from shifting investor priorities to advances in technology and the changing geopolitical landscape. During the course of the series, we're unpacking how managers can adapt to these dynamics, attract capital and deliver performance in an increasingly complex market. In this second episode of the series, reported by Ben Peyton and sponsored by Brookfield, we're delving into the artificial intelligence revolution, we'll explore the opportunities and risks for managers delivering the supporting infrastructure around AI, and we'll take a deep dive into how investors are partnering with governments, hyperscalers and AI developers as they look to get ahead in the AI race. Two examinees themes, we spoke to Stuart Uppsan, co-president of Brookfield's Infrastructure Group, and Udhay Mathialligan, a managing partner in the Infrastructure Group and CEO of Brookfield's Global Data Centre Practice. Stuart began by discussing why the growth of AI is such a big deal for infrastructure managers. Whilst we're in the middle of what is clearly a huge amount of AI hype, we're actually only at the very early stages of AI technology, AI products, the developments and how that's going to evolve over time. And so, again, that leads us to think that the demand is going to grow materially, and also to enable that, you're going to see efficiencies improve materially as well, both in terms of the chips themselves and the amount of compute that you can output per input of electricity, and the other parts of the system, including electricity generation itself, will need to improve in efficiency and in the models. And so, all of that means, as far as we can tell, endless demand in terms of those improvements and use cases and improvements in the technology, and we expect costs will continue to decline, and that will again increase demand, as we've seen with other technologies in the past. Investment in data centres has played an increasingly important role in the growth of infrastructure as an asset class in recent years. Even before the rise of AI, data centres were proliferating around the world, but with AI relying on more powerful chips known as graphics processing units or GPUs, the technical complexity of operating a data centre has increased significantly. Ute Matthialligan says this means more advanced infrastructure is in high demand for thousands of GPUs to talk to each other. They need to be very, very close to each other to avoid latency. And what this means in practical terms is these chips are located in service that are housed and tracks that are literally stacked one on top of the other, and because it's really powerful chips, they produce heat. So this heat, at the end of the day, needs to be cooled down. So what this means is the new age data set is that accommodate AI need to basically have the ability to cool really hot racks of chips, and that involves a whole bunch of technical infrastructure being created around electricity, distribution and cooling. He adds that the build out of data centres that support AI will require huge technical and financial resources. We are talking about tens of billions of dollars worth of equipment going in here so they need to be secured and operate in such a way that they're not damaged. So it is a very technical capability that's required. It's not just a building with some computer infrastructure. So it's shifting to new infrastructure being built, purpose built for these sort of high power ratings, and at really large scale. So you were talking campuses these days, so big campus, multiple buildings, and deep partnerships required with the supply chain without which you can't create these things. So I think it's a step change, both in terms of technical capability that's required, but also the capital that needs to be deployed. Of course, one of the key parts of the AI story is the massive amount of power that will be required for AI data centres to operate. The International Energy Agency projects that global data centre power consumption were more than double by 2030. Upsen says that supplying power for the AI revolution represents a major opportunity for infrastructure funds. It is data centres, which is essentially large warehouse type buildings that have cooling and internet connectivity infrastructure. Inside them they house essentially computers that have GPUs in them to run the AI applications and they need power either locally or from an external source. That really is everything that's required to deliver AI infrastructure. And so if you think about that and break it down, clearly power is a huge opportunity to invest into it's a big bottleneck around the world. And so that's something we at Brookfield are very focused on and we've been in for a very long time in terms of renewable power as well as nuclear power and in some cases gas related power infrastructure. Then you've got the transmission networks to bring that power to site, which is another huge utility investment opportunity. The data centres themselves which today spent a lot of time investing into and then the newer opportunity is opportunities around financing their chips and they compute themselves. So all of those big opportunities in a primary investment sense. Although most consider the US to be leading the way in the AI race. Mathaeligan argues that the development of AI capacity is very much a global trend. Look, it's very healthy globally. The US clearly out in front. This is a good reason why you think about the big tech companies are based in the US, whether they're cloud players or social media players. And when you look at the frontier model companies, whether it's an open AI or an anthropic. So there's a lot of foundation training activity taking place in the US. And these activities need very, very large data centers. Right. So you're getting gigawatt style campuses being developed to support this, but it's not just limited to the US. I think it's started in the US and what we're seeing is a couple of interesting things happening even within the US beyond the models you're finding there's a lot of training happening with data at an enterprise level or within the cloud environment. So there's a lot of capacity being added. And if you look outside the US and I'll start with Europe. Europe has got some national champions around this space and they have a history of government support industry development. So there is no question there's going to be AI capacity being developed where sovereign capability meets some of these national champions. He tells us that Brookfield has entered partnerships with several governments to develop sovereign capabilities for AI and to invest in the energy infrastructure required to support these capabilities. So we're finding particular AI sovereign capabilities becoming more and more important. So we are in a number of conversations across multiple markets in partnering with governments to create sovereign AI capability. Some specific examples would be strategic arrangements we've announced with the government of France. We have similar arrangements in Sweden where we're partnering with the government as they are specifying what they need for their longer term needs in being able to provide particularly around core infrastructure of data centers and power solutions that support the sovereign capabilities. And these conversations are not limited to just those markets. We've got a number of those happening across markets in Asia Pacific and in North America. So I think sovereign capability both at a central or federal level and even specific states is becoming more and more important. Meanwhile, Ups and Highlights are major tech companies are working at a frenetic pace to deliver AI capacity. I think the very large tech players as well as sovereigns have recognised AI as essential technology for the go-forward future of their businesses. I think in the case of the large technology companies it's really existential to their future competitiveness and their place in the market and they recognise that and that's why they are really leaning in hard to want to build out the capability and make sure they are a leader in the space going forward. He adds up the sheer scale of the investments in AI capable data centers means that even the largest hyper-scalers need to form partnerships to help accelerate the build out. You'll hear one gig of what training centers thrown around as is almost standard now in the U.S. A one gig what training center is $50 billion worth of capital excluding the power. That is very large-scale capital. So why are their partnerships getting formed? It's because these are very, very hard things to do and very few people can do that you need to have access to very, very large amounts of capital. You need to have development and operating expertise and in the power sector, in the data sector and in the ability to pull together that capital in terms of financing structures. Very few people have those things and even less have them in a geographically diverse nature. And so as a result, speed to market is the most important thing in terms of competition in this AI race. And so parties that can deliver all of those sort of things have access to power, have access to land sites can pull the capital together quickly and develop to a high standard. Those partnerships are extremely valuable to these tech players and these sovereigns and that's why you're seeing these partnerships for. We have a partnership with Nvidia where they have made a significant investment into our fund strategy and committed to delivering their technology expertise into our investments and partnering with us on various investments that we're looking to make with sovereigns. And so that is hugely strategic and beneficial to us when we're going out to make these investments. Partnerships are also being formed with hyperscalers looking for renewable power sources, says Martha Alligan. Power is a really major constraint in terms of building data centers at scale. And we partner very closely with the hyperscalers in delivering not just data to the solutions, but we have a very large renewable energy business. So we have at least, you know, we've partnered with Microsoft to deliver up to 10 gigawatts of new capacity in markets in Europe and in North America. And then this is green power for new data center capability. Then we've also got a another partnership where we're providing three gigawatts of renewable power to Google for their data center capacity. So we have a multifaceted approach of addressing this market opportunity around data centers, not just with data set at capacity, but also with renewable power. This brings us back to the question of how data centers can be powered. Many major tech companies keen to minimize their carbon footprints have been working to draw on renewable power sources as their data center capacity has increased in recent years. Upsen says that while solar and wind are key parts of the power solution, data center operators will also need to consider other options. I think what you're seeing is really a move to an all of the above power solution approach. And that means that, you know, not only do we want batteries and renewable power, but we also need co located behind the meter solutions, wherever we can with the data center sites. And in most countries in the world admittedly not Australia, you know, nuclear is now a really big focus and in our energy transition business, we own Westinghouse, which is the foremost nuclear company in the world. It is looking to build many new nuclear sites in the US and other places in Europe to support this growing energy demand. Indeed, multiple companies are looking to develop small modular reactors, which are poised to enter into commercial use in the US and Europe by the early 2030s. Meanwhile, some tech companies are also looking to draw on conventional nuclear reactors. Microsoft and Google have both recently announced plans to restart mothballs nuclear power stations in the US to help power data center clusters. I think many of them have now recognized that nuclear and base load power from nuclear is essential to enabling that to happen and nuclear is obviously also a clean form of power. And so that's a key part to solving that puzzle. We in conjunction with working with the US are moving, you know, new nuclear development forward as fast as possible, Westinghouse has existing highly established mature nuclear technology, so it can be rolled out relatively quickly with the support of the US government. And there is a program now being worked on to build eight new plants in the US. Ups and also notes that while AI is driving power demand, which is increasing at a significant pace in the US for the first time in decades, AI technologies can also help manage demand for power. I think in the longer term, it is really spurning a huge wave of new large scale energy generation development that we have not seen the likes of in recent history. That I think wouldn't have happened if we were just trying to do the transition and I think in the fullness of time it's going to be huge benefit. In addition, I think, you know, AI itself, the technology is the flywheel that is going to move technological advancements forward at a much faster pace than we've ever seen before. I think that will end up applying to the efficiency of our energy generation solutions. It will also apply to the efficiency of AI chips themselves, all of those things will feed on themselves to help solve this challenge. And I do think it will be addressed and sort of normalize itself as we go through this build out. But we'll all the hype around AI proved to be overblown. Concerns of the bubble have increased over the last few months with several major AI players posting disappointing financial results. Ups and says, however, that even the bursting of a bubble will not detract from the fundamental need for AI infrastructure. You know, when we think about the dot com bubble in the internet era, we both had a bubble, a big stock market crash as a result. And the most useful piece of technology emerge that ever emerged at that point in time, which is clearly highly, highly valuable and still very valuable today and really the bedrock of one of the key bed rocks of what's enabling AI to be what it is. And so that is to say, because something may or may not be in a bubble does not mean that that is not a valuable thing. It just means that people, you know, get overexcited and particularly equity markets kind of, you know, get ahead of where values truly are. Mathaeligan believes managers that remain disciplined will be well positioned to navigate the AI growth journey. If you look back at the dot com here, a lot of infrastructure got created in the form of submarine cable and fiber and these things, there's a tendency for, you know, during a bubble for demand projections to get ahead and then, you know, stuff gets built, but it gets used. So the question is, again, from an investor perspective, how knowledgeable are you and how disciplined you can be of need to be to get through this phase. I think as long as the infrastructure that's being created is going to be useful in the future and we've paid for it in the appropriate way and we've got trade protections around it, I think it's going to be okay. And up some shares a lesson from the dot com bubble that can help guide managers thinking in the AI era, the dot com internet bill where it is a great example where the situations that got into trouble was where the infrastructure was built effectively on spec on the expectation that there was going to be this demand coming and, you know, too much got built at the same time, too early or even in the wrong place. By and large, all of that infrastructure is now valuable and useful today and the fullness of time, but obviously it can be damaging if you spend a lot of money building something well ahead of its usage. And so that is entirely possible in the AI boom as well. And, you know, again, at Brookfield, what we focus on is building for real demand where we have a real creditworthy customer who's going to sign a long term contract for that usage and then we build in partnership with them. And so, you know, that's very different than going out and building on spec and whilst others may go out and build on spec, some of those parties may get fortunate and create something very valuable that they can sell at a huge premium and others may find that they built in the wrong place or too early and don't have a customer. Given that technologies around AI are developing at a rapid pace, one concern is that some data center infrastructure might quickly become obsolete. Yet Mathialigan says infrastructure managers can minimize the risk being left with the digital equivalent of stranded assets. We're not investing at the high risk end of technical change. What we're doing is we're doing a whole bunch of very important piece of critical infrastructure that supports that technology being used. So, I think what we think about data center designs instead of fresh set every 10 years inside and parts for cooling and electricity distribution, but the core doesn't change too much. We've got more than 150 data set as across 15 plus countries. So, we've got, you know, very good experience around things that have lasted decades that are still, you know, very, very useful. And then it comes back to location, it comes back to the coming together of critical infrastructure of, you know, fiber power space. Ups and ads that most of the supporting infrastructure around data centers can be repurposed if needed. We have power generation, we have power transmission, we have the physical land and building site of the data center, then we have the compute inside the data center. Everything up to the compute doesn't really have much of a technology of solicence risk. You know, even older generation that may be less efficient than newer generation is still very likely to be required for its useful life transmission is very hard to replace. And so, again, I think that that is solid, you know, well-located, powered large-scale data center sites. Again, I think we're going to need more of them in the future, not less of them. So, again, I don't think there's much of a risk around that. The real issue is around the compute inside and the compute inside essentially is like any computer. You know, not many people are using a computer that's more than five years old if they're doing high-end tasks. As Epson mentioned at the beginning of our conversation, we're still at an early stage of the AI journey. Already in the infrastructure space, AI tools are allowing for greater efficiency, especially around automated processes. And in the longer term, Epson expects advanced robotics known as embodied AI to play a greater role in the industry. I think in the fullness of time where AI moves to is AI becomes embodied and you end up having, you know, robotic labor that's utilizing AI that can actually carry up manual tasks in the management process as well. So, you know, we're starting to see the early signs of that in specific tasks where it's very effective, but I think, you know, if you fast forward ten years, I think that will be a pretty dramatic shift. You're going to see just more and more of those sort of situations where, you know, one person is able to achieve much more because of the leverage given to them by AI. That will increase over time both in the software form where AI agents are going to, you know, be able to do more and more autonomously with less instruction and then eventually embodied AI where we will have, you know, robotic AI agents who will be able to do the same sort of things in the physical world. Eventually, I think you are going to be able to see both of those forms of AI able to do, you know, essentially any task that a human can do that is repetitive and structured. He notes the Brookfield recently formed a partnership with Silicon Valley Company Figure AI, which is developing humanoid robots. We investigated the market, we spent time with lots of different players in the industry, we ended up forming the view that Figure AI, which is a California based company, is the best place to have that chat you be team moment first, we think they have the best technology hardware wise, they have an amazing AI model that is, you know, driving the humanoid robots. And really the gap right now is data and so we formed a partnership with them where we made not only made a significant investment, but we partnered to enable them to collect data across certain of our assets where they can utilize that data to advance their AI models. And then in addition to that, there's all sorts of other aspects where you know, we will get early access to the robots, we'll get to pilot them in certain settings in our businesses and then we can be there to work with them to support their compute and infrastructure needs going forward. Looking into the future, Mathaya Alleghen says the possibilities for infrastructure investors to support the AI rollout appear almost limitless. Data has become such an integral part of everything we do like at a human level and beyond the human level in terms of machines and so it's so intrinsically linked with economic growth and activity, I don't see an end any time at all as long as this activity, there's going to be more data. And you know, the rate at which data is getting produced and needing to be moved and things done with it is like college is doubling every three years, right, like so then you need infrastructure, digital infrastructure to handle all that growth and data. And you know, that manifests itself in the form of data centers, of course, very topical right in the center of a lot of news, but there's also fiber infrastructure and other other things that are needed to support itself. From my perspective, I think there's going to be a very, very long runway of opportunity for digital infrastructure and one not even opportunity actually there's a necessity to produce these things. And so like there's no end in sight per say, but I think the way the growth will pop up and you know, turn up for different investors and different operators will be different I think over time. So there's there's going to be a very, very long, you know, runway of investment opportunity for all things digital, and you know, you can break that down into data centers fiber and new things are popping up along the way. So I think asset investment opportunities are going to grow both in scale, but also in terms of its compute infrastructure almost is another layer of service that becomes investable. Ups and echoes this point, but reminders of the key role of old fashioned power infrastructure in allowing the AI revolution to accelerate. I think it's impossible right now to estimate the scale of the opportunity. I think it's the largest investment opportunity in nominal dollars that's ever been seen. I think the investment opportunity for AI infrastructure is absolutely enormous. We estimate across the next decade, we think this spend is going to be an excess of $7 trillion across power data centers, compute and kind of adjacencies to that infrastructure. But the point is it's a huge, huge, huge investment opportunity. I think the limitation right now and it's not a limitation. It's kind of a hand break a little bit that will just moderate the pace of development, which is probably not a bad thing is largely going to be power related. The power struggle that ups and highlights will help decide the winners and the losers in the race to develop AI capabilities over the next few years. We're likely to see this race for dominance in AI unleash unprecedented technological innovation that could reshape industries ranging from robotics to automotive health care to finance. It's a race that could make or break the world's largest companies offering dizzying rewards to those who master new possibilities. And the AI race could have a profound effect on geopolitics helping to determine the balance of global power for the rest of the century. Few people today can predict with certainty how AI will develop. But as investors prepare to plow hundreds of billions of dollars into AI supporting infrastructure over the next few years, there can be little doubt that AI will be one of the most significant opportunities that infrastructure managers will ever encounter. Thanks again to Stuart Upsen and Uday Mathialligan from Brookfield for joining us. For more on Private Market 2030, explore our coverage across PI group titles including InfrastructureInvestor.com, PrivateEquityInternational.com, PrivateDebtInvestor.com and PerryNews.com. This episode was reported by Ben Payton, edited by Eric Fish and produced by me, Evie Russman. Thank you for listening. [MUSIC]