Investing In AI: Data Centers, Power & Global Infrastructure

[00:00:00] Speaker 1: I'm Sean Klimczak, Blackstone's Global Head of Infrastructure. And I'm Nadeem Megji, Global Co-Head of Real Estate at Blackstone. Nadeem and I are incredibly excited to speak with you today about data centers and power, the picks and shovels of the AI revolution. As the biggest owner of data centers and utilities, we're really happy to be able to share our perspectives here today. [00:00:24] Speaker 2: So here we go. We kind of all acknowledge that there are trillions of dollars of capital that are being invested in data centers, in GPUs, in power, and just open the Wall Street Journal right now and there's going to be at least two headlines about this subject. And our hope today is to try to clarify what's really happening, try to separate a bit of the myth from the reality, share our perspectives, but to start by sharing some of our favorite factoids about what's happening in AI. And I'll go first. The cost of inference today is 99% lower than it was only two years ago, and yet we're seeing 37 times more data center absorption today than five years ago. What that tells you is that even as these models and chips are getting more efficient, demand is growing even faster than those efficiency gains. As the power and infrastructure guy, [00:01:17] Speaker 1: I love the next stat that a one minute AI video can require as much as 10,000 times more power than a Google search. That's the equivalent of charging your iPhone 119 times. And then just zooming out a little bit more broadly, AI, the electrification of vehicles and heating and re-industrialization are driving an expected 40% increase in US power demand over the next decade. That's relative to the last 25 years where power demand was flat. [00:01:50] Speaker 2: So the next one I love is a real estate investor. 60% of all AI funding today, venture funding, is happening in one city, San Francisco. That makes us very bullish on San Francisco's real estate market. We just bought the four seasons in San Francisco, and we recommend that all of you stay there when [00:02:07] Speaker 1: your next visit. Good Blackstone ad. And then next for me, AI, as we all know, is not just a US phenomena. This is a global phenomena, but the data center build out globally has been greatly lagging, that of the US. To put that in some context, Chicago, the fifth largest data center market in the US, has as many data centers today as all of India. We're all fairly sure that a few years from now, [00:02:34] Speaker 2: that will not be the case. So what does it mean for data centers, all those stats that we just shared? I want to start by talking about demand. You know, a couple years ago, the AI labs were generating less less than a billion dollars in total revenues. Today, the number is 30 billion, expected to double next year. And yet, as you heard from others, we're just at the beginning. Enterprise adoption is still in its incipient sort of stages. Five percent of work hours today are aided by AI. From a consumer perspective. You know, we all use ChatGPT for research, to ask questions, to draft emails. We're going to go to a world soon where AI agents will help us with everything, help us with booking travel, with online shopping, even driving our cars. And, you know, one factoid that I love is that this year, Waymo surpassed Lyft in market share in San Francisco. And we think that's going to happen in other markets here over time. And then, of course, we had a great panel on robotics. And one thing I can tell you, I don't know which of those founders wins, but robots will use a lot of compute, more than any of the existing applications. And that need for compute will lead to needs for data centers and power. Now, in terms of data centers, we are the largest owner, as Sean said, in the world. We've done this with the leadership and partnership of our portfolio companies here in the room, QTS, Air Trunk, Lumina. And what you see on the screen is that the hyperscalers continue to accelerate their spend, projected to invest $2 trillion of CapEx over the next five years in their digital infrastructure. And on the right, what you see is the forecast for data center demand. And each curve represents a revision to the forecast. Today's forecast is 60% higher than the forecast from only a year ago. And when we talk to our customers, we're seeing it. QTS has seen a doubling of our leasing pipeline just in the last quarter, Air Trunk seeing similar trends. Our customers are telling us that they're constrained by compute. That is the thing that is slowing them down. Now, anytime something like this happens, the question is, is there a bubble forming? And of course, there's always the risk of that in moments like this, of excess in certain places, of a misallocation of capital, as John said yesterday. But by and large, what we're seeing today is CapEx that is rational in the context of the scale of this opportunity. To put some perspective and provide some context around that, we thought it'd be helpful to frame what's happening in AI with what happened with the railroads in the late 1800s with the dot-com boom in the late 1990s. On the left, what you see is that we were investing 4.4% of GDP on average in railroads for decades. And it peaked at 9% of GDP in the highest CapEx year, just in railroads. Today, in technology CapEx, we're spending around 2% of GDP. On the right side of the screen, what you see is that in both the railroad boom and the dot-com boom, companies were overextended. They were investing 2 to 4 times revenues in the form of CapEx. Today, the hyperscalers are investing 25% of revenues in CapEx. Again, there will be exceptions. Some people might make mistakes. But by and large, as we think about where this is heading, we think CapEx is heading higher, that the technology is real and is going to have massive applications that we haven't even scratched the surface of. And that makes us constructive both in the US, but also around the world. And I'll pass it to Sean to talk about the rest of the world. [00:06:26] Speaker 1: That's great. Thanks, Nadim. Well, maybe what we can do is zoom out for a moment and just take a little bit more of a global context. Let's take a look at data centers in the US relative to the rest of the world. As you'll see from this page, there are roughly three times as many data centers in the US as are in either Europe or Asia today. And importantly, we in the US are building more data centers this year than exist in either Europe or in Asia. So double clicking on Europe, you'll see 65% fewer data centers, 80% lower data center build out per capita than the US. APAC is even further behind the US. Roughly 70% fewer data centers and 95% fewer data centers per capita. So a lot for the rest of the world to catch up. As you think about just usage, data centers in the US are about 60% of the data centers are being used today for AI. In APAC and in Europe, it's only 10%. In APAC, we've yet to even build our first training data center. So very, very early stage. And then one last final fact of the US relative to the world. The largest data center in the US is roughly 10 times the size of the largest data center anywhere else in the world. So as we take a look at the world, as I said before, we don't expect AI to just be a US phenomena. We do expect Europe to catch up and pretty soon power demand more broadly. We expect just over the next six years, roughly a 28% increase in US power demand. That will not be spread like peanut butter across the country. You'll have a place like Arizona with 60% power demand growth. You'll have a place like Northern Virginia with 85% power demand growth. And a place like Northern Indiana, they just signed a contract that more than doubled the amount of power in that region with a single contract. So put very simply, what we think is, is that access to power and the time to power will certainly be shaping where you place your factories, where we place our data data centers. And just to use a fact, Northern Virginia, the time to power for a data center is now up to seven years versus just one year, a handful of years ago. So power will certainly be shaping the future. Let's double click on data centers. Only 4% of US power today is being used by data centers, contrary to much of what you see in the press. That is expected to grow by 4x over the next six years to about 14%. But a majority of power demand in the US is in fact actually expected to include reindustrialization and electrification. So in terms of reindustrialization, $9 trillion of new factories announced in the United States just in the past handful of months. Put some of that in the context, a new semiconductor facility represents roughly 100,000 homes worth of power. On electric vehicles, Goldman Sachs estimates that by 2040, roughly half of all new vehicles sold in the United States will be EVs. Every new EV that gets added to your house adds about 37% incremental power. Then we've got roughly 40% of our states in the US that are actively trying to encourage the shift from natural gas to electric heating. Every house that flips from natural gas to electric increases power demand by about 50%. So a number of different factors, AI, the electrification of vehicles and heating and reindustrialization driving that roughly 28% increase over the next five or so years and 40 plus percent demand growth over the next decade. We believe that will require about $2.4 trillion of capital in the broader utility and power sector. So another spot that we're all going to be focused upon. And so if I just zoom out again and just say as we take a look at how AI and reindustrialization will be affecting our society, we do think that will meaningfully impact where many of you are thinking about placing your factories and your people as we go forward. So with that, on behalf of Nadeem and myself, thank you for the opportunity and we look forward to continuing the discussion about two topics that we're quite passionate about. Thank you. Thank you. Thank you.