Speaker 1Hi, I'm Bruno Alves, Editor-in-Chief of Infrastructure Investor, and welcome to the Infrastructure Investor podcast. In today's episode, I sit down with Sikandar Rashid, Global Head of AI Infrastructure at Brookfield Asset Management. Sikandar's job title is newly created and comes as Brookfield launches its own dedicated AI infrastructure strategy. Unsurprisingly, we spend a lot of time talking about how that strategy will work and why Brookfield decided to create it. A highlight of our conversation is how Brookfield intends to bring down the cost of capital for compute and whether those investments will check the right infrastructure investment boxes. We also touch on the growing investment opportunity in stabilized data centers, digital sovereignty, and much more. Hi, Sikandar. Welcome to the podcast. Hey, Bruno. Great to be here. Thanks for giving us the time. So I wanted to start with your newest job title at Brookfield, which is Global Head of Artificial Intelligence Infrastructure. Why did Brookfield decide it needed to create such a job title?
Speaker 2Yeah, look, Bruno, Brookfield has been investing in infrastructure assets that formulate the backbone of the economy for multiple decades. Today, we are already one of the largest digital infrastructure investors. We are one of the largest developers globally with over $150 billion in asset center management in many relevant sectors to artificial intelligence. And these are data centers, of course, where we have operating assets or operating slash contracted assets, pipeline of over 5 gigawatts. On the telecom side, we own 300,000 towers. We're the largest private tower owner and operator in the world. And of course, on the renewable side, we have the world's largest renewable, we are the world's largest renewable investor with over $200 billion. 70 gigawatts of capacity, 50 gigawatt operational, 220 gigawatt of pipeline, and then 140 billion of AUM. And I would say Brookfield is unique in being a top tier global partner to the world's large technology firms in both data centers, as well as clean energy that powers these data centers or AI factories. And like all asset classes, infrastructure sector has evolved from railroads, ports and pipelines to towers and data centers. And this will continue to evolve in the coming years. And I would say the last thing I would say is AI is the next evolution of this asset class with the underlying infrastructure expected to play a critical role in this next industrial revolution, which will be powered by artificial intelligence. And this technology will be transformational as well as capital intensive, similar, Bruno, to all the other general purpose technologies the world has seen in the last 250 years. So with all of that in mind, the huge addressable market, as coupled with our existing operating capabilities, just supports the view that we need a new standalone strategy focused on artificial intelligence infrastructure.
Speaker 1Yeah, that makes sense. Very recently, during your recent earnings call, Brookfield Asset Management President Conor Teske spoke about precisely the creation of that dedicated AI infrastructure strategy. And so maybe for the benefit of our listeners, what are you thinking about in terms of what would fit into such a strategy? How do you envision it working? Just give us a flavor of the nuts and bolts.
Speaker 2Our view is for the world to get to artificial general intelligence, the total economic productivity gains can be as high as $10 trillion of GDP growth per annum. And for context, that is two German economies today. Obviously, it's massive for us. But if you ask us to get there, total capital required will be anywhere between $7 to $10 trillion. And this capital will be spent over the next 10 years, is our estimate, and will be spent across four major categories. First is, of course, data centers and asset class. Everyone listening to this understands quite well, it will require two out of the $7 trillion of capex per estimates behind the meter power will require, or power call it, will require half a trillion dollars. And compute infrastructure, or the kit that goes inside of a data center, will require up to $3 trillion of capital. Without this compute, there is no intelligence. And lastly, there's another $1 trillion of capital that will be spent across the AI value chain on different types of initiatives, including data center, you know, dedicated data center connectivity, on-shoring of manufacturing. Facilities, for example, fabrication sites for semiconductor manufacturing, and other capital partnerships with a host of other stakeholders. So, and I think we believe, coming back to your point, a dedicated AI strategy that is needed to deliver on this large-scale capital expectations. And that is honestly really, really exciting. We've got very exciting conversations ongoing with some of the largest companies in the world, and a very exciting pipeline to support our operating capabilities. And that gives us the confidence to launch a dedicated strategy.
Speaker 1And just to be, you know, just to be clear, really, you're thinking those four verticals, let's call it that, which you just referred to, you would have a strategy, and that strategy would be free, so to speak, to address CAPEX needs across that spectrum, right? That's kind of how you're thinking about it.
Speaker 2Yeah, that's right. That's right. And then, obviously, Bruno, with the expectation that this hard infrastructure has to be funded, and it has to be funded in ways that check all the boxes for Brookfield as it relates to infrastructure investing. So, long-term cash flows or contracted cash flows, high margin, you know, acquiring high margin businesses that are highly scalable, businesses with high barriers to entry, inflation linkage, et cetera. So, yes, we will be looking to invest across the entire AI value chain, all those four categories I outlined, but with a focus on generating infrastructure. Yeah, and I think
Speaker 1within that, and it's something you touched on, you know, in your earnings call and in this accompanying white paper you published, but maybe an area that people are less familiar there with is actually what you guys call infrastructure in the box, so the compute you were just referring to, and you actually talk a little bit about in your white paper about the infrastructure-like characteristics of GPU as a service. I kind of wanted to pick your brain there and invite you. I'd love to expand a bit more there.
Speaker 2Yeah, look, it's a really good question, and I know we touched on this in our white paper. The reason why it's an important area of focus for us, Bruno, is this is where almost 40% of the capex in the next 10 years will be spent, and that capex, it's obviously huge amounts of capital that ought to be spent, but the cost of capital for that portion of the AI value chain is very high, coming back to my comment earlier, for the world to get to AGI, which is ultimately the pursuit, it's not formation of chatbots, it is the pursuit of artificial general intelligence. We need the cost of compute to come down more significantly, so that's part of our thinking. What we're looking to do is, coming back to our strategy, we know what a data center is, we know what a power plant is, the audience, the investors are quite familiar with that asset class, but the vision here is, our vision as a firm is, the world needs not only seven trillion dollars of capital, majority of that capital has a very high cost of capital today, and we are stitching together programs, or bringing together different stakeholder balance sheets, call it, to reduce that cost of capital to effectuate Jevons' paradox on the capital side, not only on the cost of technology side. That obviously, as it relates to the GPUs, similar logic. Today, that cost of capital is very high because the contractual frameworks may not be commensurate or appropriate for an infrastructure cost of capital. So, it's evolving, it is early days, but we're in discussions with different stakeholders around not only making this expensive compute available to them, but making it available to them over the longer term, and structuring it in a way that allows the consumer to reduce their overall cost of compute, and I think it's a very exciting area which will evolve relatively quickly in the coming months and years.
Speaker 1Yeah. And to be fair, one of the things, I think you've just touched on, it's super important, the contractual frameworks behind, and like you've mentioned, what you want to do with your strategy still has to be AI infrastructure and TIC infrastructure characteristics. And to be fair, in your white paper, you talk about four to five year contractual lengths for this GPU as a service, and you are calling it infra-like, but I am thinking about how you see this evolving, because we're now at a point where, because of what's happening with power, you could argue that the contractual basis of data centers has actually improved. If you've got power, you can get a longer term contract. Four to five is probably on the shorter side of what people would want in an infrastructure context. How are you thinking about it?
Speaker 2Yeah. No, look, Bruno, you raise a really, really good point. Look, I think it would be hard to argue that the contractual tenor for these GPU deals can get longer. And the reason for that is, at least usually the way I think about this, is the contractual tenor for an asset, should ideas be commensurate with the useful life of that asset. So on the data center side, as you pointed out, the contract tenors I've seen over the last 10 years have gone from like five years to actually for retail call, it was always three years at best. So from three to five to seven to 15 today, I've seen 20-year contracts with renewal provisions. So the contract tenors are becoming more commensurate with the useful life of that asset. As it relates to the chips or the GPUs, it's hard to have that expectation because the useful life of the chip today is, you know, at best five years. And that's part of the reason why at least our focus has been on five-year contracts. It's possible as technology advances, the useful life may be shorter. Don't know the precise answer, if I'm honest. But what's more important for us and for our strategy is to focus on a return on and off of capital over that initial contract term, which is commensurate, as I said, with the useful life of the underlying asset. And that means as long as you get your money back and a very strong return on that capital without having to take technological risk or any renewal risk with the same customer, that's actually a very good outcome. And when you couple that with obviously longer data leases on power and data centers, the blended vault for often integrated. Offering is blended vault and contracted moik is very, very interesting.
Speaker 1That makes sense. And I just, since we touched on Jevons paradox and obviously you've partly answered what I'm about to ask you next, but when deep sea came on and the specter of overbuild was raised, obviously Jevons came up and people explain, you know, why overbuilding that sense perhaps isn't so much of an issue, but what I kind of wanted to ask you though. Is, you know, I think we all agree that AI as a technology is, is a game changer, which is perhaps a little bit different from saying that everybody is going to make money out of it. Right. And every corporate that engages with it is going to have a profitable business model. And so I think when you speak about overbuild, are you concerned that, you know, there is a risk there that a lot of these data centers are getting built, but not everybody is going to make money out of it. And so somebody is left holding the bag, so to speak with some of these assets.
Speaker 2Yeah no, look, Bruno, in the past, the last general purpose technology, it was the internet and obviously it led to a huge investment boom, $800 billion of capex or capital invested in the ground or in fiber networks and we all know the story, there were quite a few casualties. So clearly that this concept of overbuild, people do, investors do start to get a lot of flashbacks and obviously that, that makes it a very relevant point. are starting to sponsor AI factories. So a few examples are You know, you would have heard the European Commission president launched the Invest AI initiative back in February in Paris. And this initiative aims to mobilize up to $200 billion in public-private investment in AI. I think they'll end up building five gigafactories across Europe, Canada. Some of the developed Asian countries are all focusing on similar initiatives to get the infrastructure up and running. And I think it's really critical that governments support it. A, to ensure the AI stack is built onshore and also for AI stack is available for the local AI ecosystems. And lastly, to ensure that they're not left behind in this AI race. And the last thing I would say is a lot of the governments are willing to anchor some of these developments to help reduce the cost of capital for this construction, which is also very important. And this is where our announcements come into. Into play, we've signed, we have plans to invest 20 billion euros in France alone. France is a nuclear power hub of Europe. It is strategically located. It wants to export intelligence to the rest of Europe rather than just exporting electrons. And it's got a leadership that has the vision to compete with the U.S. and China in that regard. And now we're seeing Germany and obviously the European Union, U.K., Canada, all embarking on similar initiatives. And that's really exciting.
Speaker 1We hear a lot more these days about, well, the term stabilized data centers. I know you've recently concluded a transaction in this space. Lots of people talking about it. Very interested in hearing from you more about the characteristics of so-called stabilized data center and what kind of de-risking needs to be in place for a more core-minded investor to feel comfortable here. What kind of boxes do you feel need checking in these investments?
Speaker 2Yeah, no, look. Bruno, we did complete a transaction recently. It was a great outcome for the platform and also a great outcome for some of our incoming partners. And the reason for that is, just to give you some context, we carved out 244 megawatts from our data for platform and basically set up a new stable core, which found many suitors. And the reason for that is to answer your question. It's in addition. It's in addition to ensuring the cash flow profile of this stable core akin to high-quality core infrastructure assets. In addition to that, these assets are also located in some of the highest sought-after availability zones in Milan, Madrid, and Paris. So data for its history, the data for is our hyperscale data center platform here in Europe. It's one of the largest, if not the largest, private hyperscale data center platform. It's the largest part of the world. And data for effectively commenced, I would say, the formation of these availability zones that, you know, over time become a hub for many hyperscalers. And these campuses we have carved out were effectively the very first availability zones set up in these three major markets. And why am I telling you this and why this is an important consideration for investors is because location matters, right? Real estate, it's location. Location, location. And for these sites, going forward, locations that can house both cloud workloads as well as AI training and inference workloads will be extremely well positioned for the next several years, if not decades, from a recontracting standpoint. So I guess that was critical. That's a critical component of a good stable core. And from, you know, other infrastructure characteristics, we had that particular, in that particular instance. We. We have some of the best technology firms as our long-term contracts. It has a 12-year plus weighted average contract profile. It has fixed escalators, in some cases inflation linked. And it's a very high margin business. Relatively new. So, you know, maintenance gap expend in the next, you know, 10 to 15 years can be estimated with a high degree of confidence. And those are all the reasons why that particular transaction. We had a very successful outcome on this particular transaction. I've heard of other stable cores that have not gone as well. And in summary, Bruno, I would say it's the quality of the cash flows as well as the quality of the locations that investors are ultimately interested in.
Speaker 1And this is maybe a bit of a hypothetical follow up, but I'm also thinking because the sector is newish, there's technology evolving. There's a power question, et cetera. So, you know, in the case that let's say some of these assets, for whatever reason, more power is needed, for example. What is the cost allocation there? Like, let's say, do these more core minded investors get a call one day and get asked to chip in more, so to speak, to cover whatever need emerges? How does that work?
Speaker 2Yeah, look, I think so. The way power supply to these sites today is is obviously fixed. We know, for example, at a particular company. If you look at this, let's say, you know, we have 250 megawatts available with additional power coming online in the next two years of 150 megawatts, so that's 400 megawatts. I think from an underwriting standpoint, it is important to just underwrite in this particular example, 400 megawatts of total compute capacity or power capacity and not necessarily make any assumptions around incremental power supply, especially in a in a stabilized de-risk. Obviously, this is different for a growth oriented business plan. But for I think your question is more focused on on the stable cause. I think as long as investors are not pricing in need for incremental power to contract out the facility in the next contracting cycle, they're not necessarily taking on any more risk. Obviously, as if customers usually think about, you know, at least in minor activities. I think that's how we price our assets. Final question, you know, at times, it really seems like the entire spectrum of private capital is trying or playing the infrastructure theme these days.
Speaker 1I think that's especially true for infrastructure, our asset class, also real estate. So how are you thinking about the hybrid, somewhat hybrid nature of these data centers? And where do you see them fitting? Are they more infra? Are they more real estate? How does broadband access help? I mean, I think that's an interesting question. And I think that's a really good question. I think that's a really good question. And that, again, with Sikandar Rashid, Global Head of AI Infrastructure at Brookfield Asset Management. We're going to be writing and talking about AI infrastructure plenty in the coming months, but if you want to do it in person with your peers, join us at the Infrastructure Investor America Forum in New York on the 4th and 5th of November. We have a dedicated panel on AI investment strategies and a whole lot more, so be sure to check the agenda. Also, to hear more of our episodes, head over to infrastructureinvestor.com forward slash podcast, or you can search and subscribe to the Infrastructure Investor Podcast wherever you like to listen.
Speaker 3Registration is now open for the Infrastructure Investor Global Summit 2027, the definitive gathering for infrastructure capital, taking place February 22nd to 2028. at Station Berlin. Join more than 1,200 limited partners, 400 speakers, and 3,500 industry leaders from more than 50 countries, all converging in one location for four days of access, insight, and deal-making. Where capital, opportunity, and strategy converge, secure your place today. Early bird registration is live now at peievents.com. That's peievents.com.
PodcastTranscript AI