Gigawatts & Megadeals: The Sheer Scale of AI Infrastructure

Cherry : Welcome to Silicon Valley Byte Size, an update on the tech sector from the Allianz Technology Trust. I'm Cherry Reynard, and with me today is Mike Seidenberg, Fund Manager on the Trust. Welcome, Mike.

Mike: Hi, how are you?

Cherry: So, Mike, in our last conversation, you talked about looking at the first and second derivatives of the AI theme, so beyond just the chip makers to companies like Amphenol, which are making data centre connectors and that sort of thing. I wonder if you can paint a picture of the scale of this infrastructure build-out.

Mike: Yeah, it's a great question, and I think the scale is really hard to digest, so I'll just try to put it in some terms that I think most of the listeners will understand. So a gigawatt data centre is roughly the amount of power for 750,000 to a million homes. So that gives you a sense of what a gigawatt represents in terms that you and I can think about.

And you hear companies like Meta, whereby they're talking about building multiple gigawatt data centres. And each of these, a gigawatt is approximately $40 billion to $45 billion per gigawatt. So when you see these numbers from a CapEx perspective, it is truly daunting, and that should tell the listeners, and it tells me as an investor, this is not a passing fad. This is a real secular theme of which we've seen the likes of previously in the cloud, software as a service, etc. So, I find the scale to be one where, as an investor, it really encourages me to think about every single aspect of that data centre and what can we do to potentially monetize it.

And one of the areas we started to look at is the power supply is a huge bottleneck. We have looked at some of the various kind of alternative energy companies or companies that are supplying different forms of energy to the data centre. But I mean, it's cooling, it's power. It's not just the chips that you hear about that are on the front page of the FT. There are just multiple facets.

And it's fascinating in that because it's such a top priority for so many companies, the amount of money that is being spent is truly daunting to me. I mean, when someone throws out $250 billion, I mean, I think that I was looking at the cost of the Crossrail, which is now the Elizabeth line as I was riding on it this morning because I absolutely love that line. I was looking at the overall cost and then I was thinking about it relative to what companies are spending on data centres. And it just shows you that the enthusiasm around artificial intelligence is really real and the spend is real. So we're spending a lot of time trying to figure out how to monetize it.

Cherry: And I mean, you talk about the electricity requirements there and you hear that just one of these data centres is almost like a small city in terms of its energy needs.

Mike: I mean, not even a small city, right? I mean, if you think about it, if you build a three gigawatt data centre, and by the way, most of these data centres are being built in areas that have proximity to power. So they're not building them in the Silicon Valley or San Francisco, they're building them in the middle of Texas, West Texas, where they have access to oil and power and some of those things, but if you think about it, so let's use 700,000 as our example, so if you build a three gigawatt data centre and that's whatever, 2.1 million households, I mean, that just goes to show you just a tremendous build out.

Cherry: And so what does that look like on the ground? I mean, are you seeing cranes and construction sites and all these kind of things emerging?

Mike: So with respect to the data centres, which are mainly being built, not in high density urban areas like San Francisco, London, etc., where you're seeing data centre build out and some of the members of the teams have been out there, are in places like West Texas, where you have access to power and energy sources that tend to be cheaper, land is cheaper, power is cheaper.

What you are beginning to see in the likes San Francisco is you're reading about lease ups. And when I say lease ups, I mean buildings being leased back up. And San Francisco really went through a difficult time during post COVID and during COVID. And now you're starting to hear about these big AI companies taking down large chunks of space in major metropolitan areas, which means that they're hiring people, which means that there's a whole trickle down effect for the economy of those local economies as these businesses need to build out teams and there are hundreds/thousands of employees if you start thinking about a company like OpenAI.

So it's exciting. I mean, it's one of those times where in my career where you can really feel it and you can feel it just based upon job activity, you're less so about your earlier question about do I see lots of cranes building data centres, just not in the right location. I'm sure if you went to certain places you could, but it's exciting. I mean, it really is.

Cherry: And you mentioned some of the kind of other plumbing companies that are benefiting from this. I wonder if you can talk about some of them. I mean, you mentioned heating and cooling.

Mike: Here again, we tend to focus on technology. So I'm less likely to buy a concrete company or I wouldn't buy a concrete company just because I think that there's going to be a lot of use of concrete for building data centres, which by the way, use the shells of these are enormous. I mean, it's just, it's a scale.

It really is a scale business, but we're looking at a variety of different companies that really go inside that data centre. And they could be something, they could be something along the lines of backups, backup power supply software, right? A niche business that basically handles the power, the backup power that these data centres need and they are software solutions in that. So we're really trying to dig in there and really understand who the various companies and various beneficiaries are. At the same time, remembering that our focus is, our job and our mantra is to focus on technology investing.

Cherry: Yeah, and so with a huge trend like this, you obviously have lots of companies in the zone and benefiting from it. I wonder how you sort of exercise discernment in that environment. I mean, what separates a good business from a great business?

Mike: Yeah, that's a great question. I mean, a lot of what separates a good business from a great business is ultimately, it's how do they execute? How good are the management teams that are running these particular businesses? And if you've heard me talk previously, in technology, a lot of times, the number one or two player takes disproportional share. I was just reading, there was a tidbit I read last week from Arthur Patterson at Excel, which he's one of the grandfathers of the Silicon Valley from a venture capitalists perspective. He was just talking about how companies and technology tend to dominate cycles. And it's why we have always been focused on the number one or two players in a given subsector.

And to be that player really revolves around execution. Execution is a function of management, being able to take the solution and really propagate the culture through, whether it's sales, marketing, or customer service, all the things that really matter. So our North Star is always to really invest in what we think are excellent companies without being what I'd call attracted to, because your question really alludes to it, really being attracted to a number three or four player because, oh, well, we like this space. It's an interesting space. Shouldn't we own all four of the competitors in that space? And we really try to remember to be disciplined because I will tell you over a cycle, the number one or two players will inevitably win.

Cherry: Is it ever possible to get as excited about more of a widget maker, a connector manufacturer or that kind of thing, than some very kind of sexy AI software company, or is it a case of having a balance in a portfolio because they're kind of doing different things?

Mike: I mean, ultimately it really revolves around risk reward. I covered the video game sector earlier in my career and it was really fun, right? I mean, you know, interesting IP, sports games, but, ultimately as a portfolio manager, my job is to empower the team to bring us good risk reward ideas. And yeah, I mean, is it more fun to go visit a company that's making some incredible AI for healthcare? Probably, but I can tell you that widget company whose total addressable market may have gone up by 4X, I mean, that's really exciting to dive in, figure out the business, figure out where it's going, where are they going to earn three years out. I mean, that's just as exciting, and sure not every business we invest in is sexy or cool, but ultimately, I can get just as excited about a business that, we'll continue the example around connectors, I can get just as excited about that as I can somebody making whatever special effects software for movies for an example.

Cherry: How far into this infrastructure build-out cycle do you think we are?

Mike: I mean, it's early days. It really is. And it's not just me saying it. If you listen to the likes of Larry Ellison talking about it in the case of Oracle, or the CEO of Microsoft, they are constrained, right? And they've talked about being constrained. And that constraint is just getting worse.

You can't get the power and you can't get these things spun up. They're talking about 2026/ 2030 deliveries on a lot of this stuff. So, I mean, there are long lead times. This isn't just something you go down to the Main Street hardware store and buy a bunch of stuff. I mean, they’re powers, power is of a bottleneck. There's all kinds of bottlenecks needed that are occurring that really preclude people from getting these things up and running quickly. And they're moving as fast as they can, right? I mean, there's no, like companies, why companies are spending so much and they're so aggressive about their spending is the opportunity. But I mean, it takes a while, it really does.

Cherry: How much of that is meeting sort of demand today and how much of it is kind of anticipating future demand?

Mike: I think for a lot of the companies providing the services today, they are constrained on the amount of service they can provide to their potential customers. Over time, that'll even out because more and more, there'll be more competition, there'll be more things that will come online.

But today, I really think it's not that there isn’t demand, it's can we supply that demand?

Cherry: Just finally, are there any infrastructure needs that are coming next that aren't on investors' radars yet? I mean, you mentioned some of those bottlenecks.

Mike: It's a good question. There's just so much happening in technology and we haven't even talked about the likes of what's happening with respect to drones and stuff like that. One of the things I remind myself is our most finite resource is our time. And I want to make sure we use it wisely. So, this to me is the biggest thing that I've come across and the team's come across in a long time.

So I'm not really spending much time thinking about what else right now just because our plates are pretty full. But inevitably there will be, and that's the great thing about the job and the great thing about the sector is you just go through these cycles where, I often say that difficult problems are solved by these really innovative companies and that creates an opportunity.

Cherry: Absolutely. Okay, great. We will wrap up there. Thank you so much, Mike.

Mike: Great talking to you.

Cherry: And thank you to our listeners for tuning in. You can find out more about the Trust at our website, allianzetechnologytrust.com. Until next time.