Can Infrastructure Keep Pace With AI?

Good afternoon. Thank you so much for staying with us for this conversation. I know that you just heard plenty about artificial intelligence, right? I mean, and just ourselves. Who here was able to resist the temptation of just using ChatGPT or Gemini or something in the past week? Right. So are we talking. Yes, we're talking enough about artificial intelligence to demand the build out. Are we talking enough about the physical infrastructure that's needed in order to keep up with what's happening with artificial intelligence, computing capacity, and what's going to be needed in the future? So this is a perfect combination of people, because we can talk about everything in that really chain that's needed in the infrastructure from the build out data centers, energy, capital, talent, everything. So, um, let's get started with going right to the point, the bottlenecks right now. Um, Abhishek, why don't you start us off? Because you're also a big data centre operator worldwide with 150 facilities, over 20 countries, and the biggest operator here in India. Where are you seeing the friction?

Well, first of all, uh, great to be here. Uh, to your specific question, let me just step back in terms of sort of where we are in this whole cycle. I think we are barely scratching the surface on the demand side of the equation. Right. So if you look at sort of token consumption today, you know, we've got 30 million software developers and sort of the fraction of it is where it's mostly concentrated, along with some, you know, research kind of work. We've got a billion knowledge workers. We're barely scratching the surface of that. We've got about 2.5 billion frontline workers. So nurses, you know, uh, factory workers, etc. And that will be all a net positive in terms of AI. And then we haven't even talked about the consumer yet, right, in terms of what consumers will do and new business models, etc. So just to frame it in terms of where we see the demand equation, I think it's there's a secular, durable trend of what's the art of the possible here.

Now, the biggest bottleneck ends up being the data on a capacity that you can stand up. Right. And and the long pole in the tent, of course, is energy. But beyond energy, it's actually a system issue because as a data center operator, you've got to bring energy locally. If it's off the grid, you know, the grid capacity as an example, in the United States has stayed flat for the last three decades. China has doubled every decade, um, in the last three decades. Right. Um, so grid capacity is an issue when you step into a data center site and we lose a lot of energy in doing an AC to DC conversion in the, in the conversion, um, and then the mechanical and electrical equipment that goes into the data centers, uh, typically runs for us. We are one of the biggest in the world, and it still runs about 42 weeks. Uh, lead time. Right. So there's bottlenecks everywhere. And then the the the biggest bottleneck nobody talks about is the hidden bottleneck is the talent. So if you think about, uh, electricians, skill technicians, design, construction. I'm an electrician at a compensation today in the United States of about $200,000, it's more than a lot of white collar job, right? Right. So there's so players who can bring that entire ecosystem and orchestrate that whole thing is where I think the the the recipe lies. Capital is not an issue. Uh, and I that's I.

Couldn't agree more. Very simply, uh, capital is not an issue and demand is not an issue. The the bottleneck is who can pull these projects, uh, out of the ground? Well, perhaps the the biggest narrative we see around the world that is somewhat detached from what we're seeing on the ground is somehow that the demand for AI, for compute, for capacity, for the energy, the supply chain that supports AI is the thought that this demand is in some way fragile. You know, we hear about a more efficient model, and people worry about the demand picture, or we hear about one data center being released and people worry about the demand picture. The demand so overwhelms supply at this point that there is going to be a supply demand imbalance through the end of this decade, and probably far beyond that. And really what we're seeing around the world with that really enduring demand profile in the scale increasing is it's not who has capital. It's exactly what was said. It's who can pair very large scale capital with very large scale operating and industry specific expertise to actually pull these projects out of the ground.

But of course we are talking about the demand. Today's frontier models per run take around half $1 billion they used. And if you were to run a three year old model, today would take 4 or $5 million. Mhm. That's how rapidly they're scaling. And the innovation needs to continue. Ethan I was just chatting. We are probably not even in the kindergarten today where the demand is going. And this is he was talking of knowledge workers. And before physical eye takes shape and we're very embryonic and physically I think demand is going to be stratospheric from where we are today. So how are you seeing that affect Hillman, for example, because you're basically building the plumbing of the.

Yeah. Look, I think this is. But it's as Connor said, keep building. Demand is not an issue. There will be stumbles and stumbles will happen because there will be pockets and locations and specifics where people may build ahead of demand. It has been in all infrastructure projects. Was discussing that somebody if you build an eight lane highway to a village with 50 people, that will not get the usage. Those kind of construction, those kind of investments will stumble. But if you're sensible about where you invest it, the demand is so far ahead of supply for the next multiple decades to come.

I want to ask you about how you differentiate that. But before I do, that sounds weird because Sandra keeps talking about the demand. And of course you are providing AA implementation to your enterprise clients as well. What are you seeing in terms of how people are really adopting artificial intelligence?

Yeah, I think it's going one direction. And in the exact same way that users are getting more efficient in using the most appropriate, uh, model for their use case. The infrastructure is getting more efficient as well. You spoke about how the grid is not going to catch up with the amount of generation that needs to come online and the transmission. We're seeing far greater buildout and acceleration of behind the meter solutions to address energy demand needs. And I infrastructure demand needs because the market's not going to wait for the grid. Um, so the infrastructure buildout will become more fit for purpose, the exact same way the model usage will become more fit for purpose, in an ironic way. By the way, I is helping with that as well. You now have demand response mechanisms that are being implemented in utilities that actually managed, not just the I work workloads in the local data centers, but also the local demand in the communities. Right. So especially when you're doing inferencing because remember training follows power inferencing follows consumption the users. So inferencing has to be close to urban areas as an example. Oh sorry. Inferencing. Yeah. Yeah. So we are seeing lots of innovations now across the whole stack. Right? So whether it's off the grid fuel cells, I think we'll see nuclear as well come through. Yeah.

If I could jump in because you in the last sentence, you still have my favorite line, which is the the energy demand environment is so significant right now that it's going to require an any and all or all of the above type solution. Every form of generation has demand right now, but we see four forms of generation capturing an outsized portion of the go for demand growth because they each have a characteristic a right to win in something. The market needs fuel cells because they're the fastest to deploy, renewables, because they're the cheapest batteries, because they stabilize an increasingly volatile electricity grid, and nuclear because it provides the baseload energy security, everything's going to grow. But we think those four grow the most. I think by 2030, our forecast suggests, or at least a third of the dataset on capacity will have off the grid on site generation.

Mhm. A lot of our US work is behind the beta. Predominant part is pretty much behind the meter. And one other thing. We are seeing the demand coming. And then I'll come back to that energy thing is truly at the edge. And this is just beginning. But I can tell you in the last 1:45 quarters, the demand at the edge, two quarters of good didn't exist at all. And within the last two quarters, we just seeing orders coming in to build one megawatt, two megawatt edge data centers. And we never thought we thought those days were over. But I think this is a lot of the discussions focus on data centers. I think the biggest discussion and biggest challenges energy and energy infrastructure, whether it's transformers or the grid. In most cases, I think going forward, excluding China. You have to discount the grid because you should not be relying on the grid almost anywhere in the world. India is slightly unique because it does have a national grid. But places like Europe, places. You know, UK just announced a big investment in the grid for us. Germany has a generation issue. They just don't generate enough electricity. France does have enough, but its grid is old. So every country. I think you have to forget the grid as the existing grid as being part of. So the alternatives.

Do we have enough right now? I mean, you listed all of those, but for example, even nuclear, there's a lot of mixed sentiment about what that takes in a lot of countries. How do you overcome that?

I think your question. Um, uh, we I'm sure we all get some version of the question. Will there be enough? Um, uh. What's enough? Um, there certainly will not be enough supply to meet demand. Uh, I think that is that is clear. I mean, we've got roughly, what, 200GW of capacity today. If it triple it to that tomorrow by magic, we'd use all of it. And we were talking with Conor just in the green room about where the risk actually lies. Right. If you, for example, have over capacity, if there are for any reason, stranded assets. I mean, we have seen the.com bubble, for example, and you are like, no, there is enough demand out there. But yeah, so somebody has to pay at the end of the day.

So there's there's two things. I think you're absolutely right. There will be stumbles that we all get up here, and we see demand so significantly outweighs supply. A very simple response to that would be, let's just throw as much money at it, because it's all going to be productive. That's not true. I'll tell you from the Brookfield perspective, we see this immense investment opportunity as a situation where we can be incredibly selective, focusing on only the best assets in the best markets, with the best contracts and the best corporate credit counterparties. And we can still deploy as much capital as we would like and in very attractive risk adjusted return opportunities. But you highlighted an important point. One thing that sometimes is often underappreciated about this generational I build out is this is like every other infrastructure build out over time except one it's larger scale and two, it's happening faster. The other thing that is very, very different about it is the vast majority of the infrastructure that is being built is being built against very long term take or pay contracts with the best corporate credit counterparties in the world, which, if they're the best corporate credit counterparties in the world today, it probably means they're the greatest corporate credit counterparties in the history of time. The biggest backstop of this infrastructure buildout, or the major hyperscalers in both the United States and in Asia, and in places where organic, hyperscale or demand isn't driving the infrastructure buildout, we're actually seeing the sovereigns step in and backstop so different than some of the everyone compares to the fiber in the.com bubble. Much more of this built out is being backed by long term take or pay contracts with the greatest corporate credit counterparties in the world.

One more difference, if I may. Um, the data sort of capacity was already in a supply constrained situation pre ChatGPT moment. Because remember we've been building data center, been growing capacity been growing at 15% plus for the last decade. You know streaming payments public cloud etc.. So now you layer in AI infrastructure on top of that just more than doubles. Right. So that that challenge becomes more acute.

Let me ask Sanjeev, is that also what you're seeing when we were talking earlier about differentiating between the projects that work and don't write contract structures to backers.

But let me go back to what Connor was saying. We were around when the fiber buildout was happening in the US we were supplying to, but we were providing and generating a project, managing some of the largest fiber build outs. The fiber build out was happening expecting the demand to be there. I would have discussions with the guys building Quest Building level three, who are all my peers. I still know many of them was. The basic thesis was let's lay the fiber. Demand will happen. Very different now. The demand is there, not being met. Let's get the data center capacity out there. The biggest advantages one is the. Absolutely. An economy at the point is the counterparty has to be right. Where the challenges are coming is not as much even on the cost, but time to deliver. And that's where supply chain is broken. Mhm. He's saying 42 months to get the Transformers, 42 weeks to get the Transformers. He's lucky because he hasn't. You know it's not. Yeah. And he's also not talking off the phone. He's talking averages. But there are projects where it takes a year and a half to get this stuff done. That project's behind schedule. If it's behind schedule, it's over cost. And he's not satisfying the customer demand. And that's the challenge. Supply chain. Simultaneous with that. And I think one of the points is we have at least taken one single global supply chain and split it into two parts. One is the US centric. The other is a China self.

I was going to say, how does geopolitics add to this whole friction and imbalance in the world right now?

Totally. You have two supply chains. Mhm. There are not just some but you have this is war is ongoing. You have crossed over. That's another set of issues before we even get to the world problems. You have to handle. A supply chain that was built over 6 or 7 decades. Break it apart and say I have chips that I can only use for Chinese hyperscalers. Chips I can only use for the other hyperscalers. Every piece of the supply chain is different.

Go ahead. I want to end this a little in a second.