The Outlook for Data Center Power Demand as AI Token Use Grows

**Brian Singer** (0:05)
Put in perspective, what is AI doing from a power demand perspective? It's adding the power equivalent of the whole country of Japan, which is the number five power consuming country. So this is really significant.

**Allison Nathan** (0:18)
The last time we sat down to talk about AI and power, the debate was whether the grid could handle it. That question has now been answered in basically the least comforting way possible, partly unevenly and expensively.
Welcome back to AI Exchanges. I'm Allison Nathan and I'm here together with George Lee, the co-head of the Goldman Sachs Global Institute. Together we're co-hosting a series of episodes exploring the rise of AI and everything it could mean for companies, investors, and economies.
George, great to see you again. Great to be here. So George, let's first set the table for people who have followed this series. When we first covered this topic, the shorthand was basically that AI would drive a step change in data center power demand after nearly two decades of essentially flat electricity load. But since then, demand forecasts have gone up again, even as we've discussed many times, models have become more efficient. The binding constraint has migrated away from megawatts and towards what I think of as all the T's. We have turbines, transformers, transmission and trades people. So new constraints entering the picture and the politics have arrived. So we have regulators, rate payers and neighbors now really having a seat at the table and having a lot to say about this topic. So before we bring in the experts George, what's the single biggest thing that's changed in your mind on this topic?

**George Lee** (1:43)
I think one thing that kind of syncs with my expectations is the demand cycle continues and what's fascinating is to watch as you described, the supply chain gets stretched all the way back from models, but to all of the physical affordances, really almost all the way back to the base elements. So this is one of the most complex, intricate and global supply chains in the world. It would be fascinating to talk with our guests today about how that's playing out, where the real bottlenecks are, and whether we can navigate those bottlenecks and the political issues you described.

**Allison Nathan** (2:18)
Yes. So we have brought in what we think of as exactly the right people to talk about this and help us untangle it all. Joining us again are my colleagues in Goldman Sachs Research, Brian Singer and Carly Davenport. Brian is the global head of GS Sustain, and he thinks about the innovation cycle and the sustainability trade-offs of all of this. Carly is our America's utility team senior analyst. She puts the numbers on just how much power we're going to need and where it's all going to come from. So, Brian, Carly, welcome to the program.

**Carly Davenport** (2:46)
Thanks for having us. Thank you.

**George Lee** (2:48)
Great to have you both.

**Allison Nathan** (2:49)
Carly, let's start with you. You and your team recently raised your overall US power demand outlook. What prompted that change and where does it leave us in terms of power demand?

**Carly Davenport** (3:00)
Yeah. So we recently raised our forecast up to a three and a half percent CAGR through 2030 And that was up from our prior forecast of about 3.2 percent CAGR through 2030 And we continue to be convicted in the demand cycle as George was just mentioning. As we think about where we sit today in power demand growth, year-to-date we're up over 4 percent. And so we feel like we're already on our road to maintaining that 3 percent CAGR through 2030 Now, what drove that forecast change? The main thing was a positive revision in our data center power demand outlook. We use a data set from 451 research, and we saw a couple of things change. One is a lot of new projects were added to the queue as we continue to see focused development in this area.
And that has taken our data center power demand in 2030 up to about 108 gigawatts up from about 83 gigawatts before. So there's new projects being added. That's point number one. Point number two is we're also seeing increased utilization of the existing data centers that we have. So we've seen vacancy rates on data centers in the US coming down across all of the major data center markets. They're sitting today around one to two percent. Over the last few years, they were ranging from about two to seven percent. So we're seeing increased utilization and occupancy of existing infrastructure in addition to the new projects that are getting added, and we forecast about a three percent vacancy rate in 2030 on a US average basis.