Power Play: Investing in Energy Infrastructure for the AI Era

PODCEPT, the podcast from Deutsche Bank Research, with interviews on current economic and financial topics. Listen as economists and analysts from Deutsche Bank present their views. Welcome, everyone.

You're listening to another episode of PODCEPT, where we discuss some of the best ideas coming out of Deutsche Bank Research. I'm Laura Daniels, and I'm the Deputy Associate Director of Research here in New York for our equity division. So we've been hearing a lot lately from investors on AI and energy usage.

So I'm here with some people today who are U.S.-based experts who can speak to this topic. Joining me are Ross Seymour, who's our lead analyst on the semiconductor sector, Brett Ryan, our senior U.S. economist, James Mulholland from our U.S. autos and space tech team, Nicole DeBlaise, our lead on multi-industry and electrical equipment, and Karin Blanchard, our expert in clean tech. So I can just start by setting the scene here.

This conversation is coming off the heels of a report that we wrote along some of the same themes that we can dig into a bit deeper today. That covered AI-driven energy demand, its magnitude, and where that's headed in the coming years. We also took a look at how businesses are responding and how those decisions are affecting the broader economy and the future of this critical infrastructure.

So let me just start with a couple of data points to put everything in perspective here that stood out to me while we were researching this report. So one of those is that the U.S. generated last year 4.43 terawatt hours of energy. And if that sounds like a big number, it is in fact record-breaking according to the U.S.

Energy Information Administration that put out this data and has been tracking it since 1949. That's also record-breaking for the last three years in a row. Maybe not surprisingly, the report lists data centers as one of the drivers of that.

And then another data point that stood out to me is that in the next four years, it's expected that global energy demand from AI data centers is going to quadruple. And that figure comes from the International Energy Agency. So let me start by turning to Ross here.

Ross, again, you cover semiconductors for us. And I should say also you're based in the West Coast, so that gives you a front-row view to what's going on in Silicon Valley. And you've been tracking what major cloud companies, so these are companies like Google and Amazon and other names, have been investing in AI infrastructure over the last decade or so.

So can you tell us a little bit about what you're finding and what does this mean for how big AI development is getting? Sure, happy to. So the numbers have been exploding from the cloud service providers or CSPs.

Last year, the company spent $465 billion in CapEx. That is up from just three years prior in 2022. That number was $161 billion.

So from $161 billion in 2022, a 42% kegger to $465 billion last year. Despite many fears of this potentially slowing going forward, this year, 2026, consensus estimates from these companies have that growth rising to $762 billion. So up nearly another $300 billion or 64% year-over-year.

Overall analysts do fear perpetually that the out-year will slow. In this instance, 2027, consensus estimates are up for CapEx growth from that group of 10% year-over-year, so slowing from the 64% to the 10%. But we note that these initial estimates for the out-years have been consistently increased over the past couple of years. 2026 is a perfect example of this.

Like I said, it's supposed to grow 64% this year. At the beginning of this year, that number was 33%. So just in the first quarter of this year alone, the CapEx growth has risen from a 33% rate to 64%.

So people will still question whether the out-year will continue to grow, but we do note that there's been a consistent upward bias to that number over the past two years. Yeah. So it sounds big.

In other words, I heard you use the term exploding. Can you, I mean, you've been looking at this for this space for a while, and I just wanted to ask you, like, are you completely inured to these numbers or are they eye-popping even to you as well? Is there a way that you could put that into some perspective for us?

Yeah. After over 25 years of covering the semiconductor space, one of the biggest beneficiaries of the CapEx from these cloud service providers, I've never seen anything of this magnitude. The good news is just recently, CEOs from NVIDIA and Marvell publicly stated that they saw no signs of slowing demand.

So it seems like the direction is still positive. And what it means to the semiconductor companies has been quite staggering. NVIDIA is publicly guided for its AI revenue to reach an aggregate of a trillion dollars from 2025 through 2027.

And yes, that's a T, one trillion. Broadcom, another company that plays in the AI space, has publicly guided for its AI revenue to rise to over $100 billion next year alone. And that's up from only, quote unquote, $20 billion in the last fiscal year.

So these numbers are quite massive, even when you bring it down to the semiconductor corporate level. Yeah, that's pretty amazing. So I think, Brett, you also have been looking at these numbers that Ross has been putting out.

So I'd like to turn to you as well. You've been looking at them in the context of the macro environment and the broader economy to try to understand what this all means. And I know you'd been talking a bit even before the war in Iran escalated, even before people were talking every night on the evening news about pressure at the gas pumps on energy prices and what this could mean.

I think you use the term energy tax is essentially what this could translate to. So could you talk a little bit about that? What's that mean?

And when are people going to start to feel this? Are we already feeling it? Or I'm almost afraid to ask, is it is there more coming?

Yeah, sure. So the energy energy prices will be top of mind for for voters come this November for the midterm elections. And at the moment, with oil at around one hundred hundred dollars per barrel, the implied increase in the energy tax would be about one hundred and fifteen billion.

So for every one penny increase in gasoline prices, it's about one point one five billion of additional spending for consumers on energy. And everybody needs to not everybody, but many people need to drive their cars to work. And so it's not an option to not do so.

So in the short term, basically, you can't consumers have a hard time adjusting their energy patterns. But you know, that's to the backdrop of tax cuts that with additional refunds on twenty twenty five would total about one hundred and eighty billion benefits for for U.S. consumers this year. So at one hundred dollar oil and the one dollar increase in gasoline prices have sustained over the course of the year, you know, the tax benefit, the tax bill from last year is able to kind of handle that that increase as you get above that, though, that it starts to eat into it.

And by with one hundred and fifty dollar oil basically wipes out the tax benefit and then some. And so that's that's one way of framing it. It's not that, you know, you have to look at it in the context of what fiscal policy is doing already in terms of benefits to consumers and then how much that increase in energy prices could potentially eat into that benefit or even overwhelm it at one hundred and fifty billion, one hundred and fifty dollar oil.

And so, you know, this is coming after a period of kind of stable demand growth in electricity. It grew at about, you know, zero point seven percent annualized rate between 2015 and 2025. And the estimates are in the low single digits, I believe, going forward based on this demand.

So you know how that what the implications are for for electricity prices, it's kind of different for each each region in the U.S. But suffice to say that, you know, the combination of what's going on now and, you know, oil is globally traded as well as LNG, you know, that's going to have upward pressure on households electricity prices as well. Most of the energy demand in terms of the increase from data centers is expected to be met in the near term by gas fired plants, power plants.

With Qatar, Qatari supply offline that makes, you know, it's more expensive all over the world. So, you know, it's unless the U.S. imposes some sort of export restrictions, which is unlikely at the moment. So, you know, the bottom line is that there's going to be increasing pressures on consumers.

And from a capex perspective, as Ross discussed, you know, we see it in our it's it factors into our macro GDP forecast. So when we saw that, you know, these hyperscalers were surprising to the upside of 50 billion or so apiece on their capex guidance, it was kind of eye opening and suggests upside risk to our to our GDP growth forecast. So as Ross pointed out, I mean, I just looked at the first the top four, you know, Amazon, Microsoft, Alphabet and Meta and their capex assumptions, as Ross said, it's like it's about it's 63 percent year over year growth for these guys.

And just from those four alone, that's two hundred and fifty billion. And that's that translates directly into what we see in computers and peripheral equipment in capex. And last year's growth there was about 71 percent.

So, you know, we expected a slowdown. We got very little slowdown in what they have now. Can they deliver on that in the time period with potential supply chain disruptions in the offing?

That's a different story. Thanks. Yeah.

So there's definitely a lot of broad implications there. And I guess something that's on everyone's mind, of course, going forward is inflation. So can you talk a little bit more to that?

For some of the inflationary impacts, if you're looking at those at all and some of the policy implications coming out of the Fed? Sure. So the the energy problem is is even more acute for for central bankers in the sense that, you know, traditionally Fed policymakers would tend to look through temporary inflationary impacts of an energy shock.

However, the backdrop to the latest one is that is one in which inflation is persistently exceeded the Fed's 2 percent target for the better part of the past five years. And so, you know, this raises the risk that inflation expectations could become unanchored. And you know, the bottom line is that the energy supply outlook is critical for nearly every asset class from interest rates to stocks.

And so in that view, in that sense, the Fed rate cycle and where we are on that and rate outlook is very much in flux right now, given the latest energy energy shock. Yeah. Thanks, Brett.

And I think, you know, you were talking about expected slowdown, but I know I've heard both you and Ross say, you know, sometimes we expect a slowdown, but often we see those numbers keep going up anyways. So those numbers can be kind of conservative on that front. So definitely there are a lot of impact on prices, on the economy, on policy, a lot of reverberating effects.

That's very big picture. So let me zoom in a little bit. I think we should start to talk a bit about the business level and what we're seeing there.

So one of the things, again, in the report, things that we saw businesses doing, we saw businesses that were solving for near term constraints, and we saw businesses also that were talking about longer term grid build out and scale. So we'll talk about both of those today. As we're talking about them, some of the things that we came across with these companies and what they're doing is that some are, you know, using time tested proven methods and others are being a little bit more innovative and adaptive.

And so that's where I want to bring you in, James. You're looking at a good example of a company that is operating on that side of the equation solving for immediate constraint and taking a little bit more of an adaptive approach than what they had been doing in the past. So I guess you know where I'm going with this, but one way to phrase the question is what's somebody from the autos team doing here in a conversation on AI and energy?

Yeah, no. So what BorgWarner is doing is definitely quite unique. So Borg is what we would call a tier one auto supplier.

So basically they're providing the most complete parts either into the final assembly of an engine or into the vehicle. So what they did is they recently actually announced that they've partnered with a company that builds out AI data centers for clients to produce something called a turbo cell, which is what they describe as a modular power generation solution. Each of these units are going to have three turbines in them and they're going to produce about a megawatt of energy.

Interestingly, they will be able to be used for what's called primary power generation, bridge power, and also backup power generation. Borg is uniquely situated in this individual instance because they have such a large portfolio of IP that they can draw on. So in essence, they were able to go to their parts bin and pull the turbochargers, the air management, and the other pieces.

These are components that have been field tested for years on vehicles and so their shelf life and their abilities are very well understood. Now they're piecing them together for a different type of combustion engine and based on what's been shared by them and if demand is there, which it seems like it is based on our conversations today, they're going to be looking to produce about 2,000 of these a year. And so put that another way, that's about two gigawatts a year of much needed extra power generation.

And so I heard you say, you know, backup power and primary power, which is pretty easy to picture. Could you also talk about what a bridge power is in this case? Yeah, of course.

So yeah, you're totally correct. Most people understand that prime is, that's initial power generation before it's connected to the grid or backup is really when you need to fall back on it if the grid goes down or if other generation options run into a bit of a hiccup. The way to think about bridge, and this is really the most interesting component, is as data center clusters get a task, they can cause large surges in power, which either strains the grid or it strains the stationary power generation that's already on site.

Now bridge systems essentially help to make sure that a data center gets what it needs by supplementing the grid in those times of higher demand. In other words, as the cluster spools up, many of these or all of these generators will click into high gear and provide that extra bit of juice that's required to really maintain a nice even load. The modularity side of it means that they can scale that up or down to fit the exact needs of what the data center is looking for.

And in reality, they can be rolled up in the back of an 18-wheeler. And if it works as outlined, they're wired up to generate power in unison. And again, if it does work, given the long lead time from traditional power players to deliver their own solutions, there's going to be demand for something like this.

Nice. Yeah. And it sounds like that's pretty much what's differentiated from these particular products.

They seem a little bit innovative in that degree compared to some of the other ones that we more traditionally see around, right? Yeah. I think that's fair.

So traditionally, bridge power solutions were in many ways a cluster of interdependent but separate systems. And what makes, again, this turbo cell unique is that under the hood, you have those three turbine generators, but you also have the electronics and the backup power, excuse me, the backup battery systems that go into it. And all of those things can essentially be just wired up in unison very easily as opposed to having to custom build it for an individual data center.

Nice. So I think with that, we're going to shift to talking about companies that are solving for scale as well and look at those that are building out grid power and building up this infrastructure. When we think about those, we typically think about, Nicole, the kind of stocks that you look at.

So these companies, you cover the multi-industry and electrical equipment companies. Since these are kind of the typical ones that you think of in terms of energy, they are also a pretty good bellwether for what we're talking about. So can you tell me what you're seeing in your space?

Is it business as usual or are things ramping up with this kind of AI buildout that, as Ross talked about, was exploding? Yeah, sure. So I think definitely not business as usual.

I mean, if I go back, I've been covering this space for about 20 years, and this used to be a pretty boring 2% to 4% GDP-plus kind of growth sector. And I think with the advent of what's happening with data center and AI investment, we now have this extremely attractive secular growth trend that's impacting both the data center vertical as well as the power and utility vertical within our space. So we have several companies that have much higher exposure to these industries.

And for those, they've seen absolutely—I'll steal Ross's word again—explosive growth. To give an example, in the fourth quarter, most companies reported much stronger-than-expected data center order growth. A company called Vertiv, which makes both thermal and electrical infrastructure for data centers, reported almost 300% year-on-year order growth, which was much higher than expectations.

And you saw dollar orders essentially double sequentially. So absolutely seeing extremely explosive growth on the data center infrastructure side. And then I think, with respect to power, we covered GE Vernova, which is one of three large global manufacturers of gas turbines for not the bridge power, but the actual base load power.

Gas turbines are the workhorse of base load power in the U.S. and in many countries worldwide. And GE Vernova's backlog is now out to 2029, and in many cases, they're taking orders into the 2030s. So, you know, some examples there, like, absolutely non-business as usual, and the growth has been really significant and continues to surprise to the upside.

Yeah, thanks for putting that in perspective. So I'd also like to talk to Corinne. You cover also some companies that are looking at this broader grid scale-up, particularly in renewables.

And I think it's hard to overlook renewables when we think about just the level of capacity that Nicole, Ross, and others have been talking about when it comes to the AI era. So, Corinne, can you tell us a little bit – you've talked especially on solar and batteries – and how is this technology solving some of the problems that we're talking about here in terms of energy scale-up? Sure.

And I would probably start with a similar comment that Nicole is saying for our space. The AI slash data center slash robotics is actually bringing, like, you know, new AI into the space, and our investor can be looking at those companies. But especially for solar and renewables, they are among the fastest to deploy sources of new generation.

And when you pair this with energy storage – or, you know, people refer to that as BESS – you provide, like, a more firm supply profile that really matches, like, the 24-7 load shape of AI-driven data center demand. So that's creating an opportunity for companies throughout the entire supply chain and sector for cleantech. Why this is particularly well-suited for solar energy – it's really, I would say, three parts here.

The first one is fast scalability. So you are going to match relatively easily that fast pace of growth in demand from AI. The second one would be a low-to-zero marginal cost, and that's very important in the current environment.

And the third one is the favorable SCOE profile, so levelized cost of energy for solar. So this trend, again, benefits the cleantech company, and we see, like, it's not just one part of the supply chain. It's really ranging from downstream energy producers, such as Clearway or Enlight, for example, to upstream equipment suppliers.

So we're talking about companies like, next, Power, Array, SolarEdge. So it's really kind of a benefit to everyone, actually, what we're seeing. And maybe one other fact that I wanted to bring here, or interesting fact, is we're already actually increasingly seeing direct off-techs.

And it's done through, like, power purchase agreements, so PPA, with large technology players. So that includes Amazon, Google, to just name a few. And again, this is kind of a full circle happening here, where it supports the downstream producer and in turn results in a build-out of development projects that support the equipment suppliers here.

Yeah. Thanks. And so it's obvious how some of this technology can obviously help to fill some of these gaps in its state now, but do you have visibility on how this technology is evolving?

Do you have some expectations for maybe the next couple of years out, or even farther on the horizon, that might help in this space here? Yeah. I would say, probably from an energy supply standpoint, look, I think we could see, like, a higher efficiency, especially for solar panels.

So I think this is definitely a goal of the industry since years, but I think that will be achieved through kind of, you know, new technologies. So we're seeing, like, some of the players using perovskite, such as First Solar, for example, versus the traditional polycrystalline panel. This is really just, I would say, like a pure technology generation view and, like, advantage.

I think maybe a little bit less technological development, but would be to see, like, increased battery technology slash chemistry that we can see across the grid. And again, it's not something that we're going to see evolve over the next three to six months, obviously, in probably, you know, medium term, so a few years, or maybe three to five years. But I think as we see more energy storage capacity and chemistry emerging, I think that's really going to be, like, a key for that shift, and they are key load-shifting entities.

And I think as we see increasing duration and storage duration across the energy storage space, I think that's really would support this. Maybe one last, and I think we see some conversation around small modular reactors getting some attention. So again, nothing in the next six months, but I would say probably a topic to keep in mind for the next, you know, past 2030.

Yeah, there's definitely a lot to talk about and a lot to pay attention to in this space right now, and I think in the years to come. So thank you for this. Thanks.

I think I'm going to wrap it up here because we really have covered a lot of ground. So thanks to everybody for chipping into the conversation. You know, a lot with AI, certainly rapid developments, things that are evolving quickly.

I feel like I'm getting whiplash these days with everything that's going on in the tech space. Plus, when you layer on all of the geopolitical volatility, that's keeping this a little bit interesting, so to speak, to put it that way. Thanks also to everyone who's been listening.

If you'd like to check out the report that sparked this conversation, it's called PowerPlay Investing in Energy Infrastructure for the AI Era. You can find it on our Deutsche Bank Research Institute page, or there's also a client-facing version on our portal. You can also reach out to your Deutsche Bank representative.

If you'd like more information on anything we discussed here today. This has been Podsept from Deutsche Bank. Thanks again, and we'll catch you on the next Podsept.

Podsept, the podcast from Deutsche Bank Research. This podcast has been produced by Deutsche Bank and may contain research as defined in Method 2. The information discussed is believed to be reliable and has been obtained from public sources believed to be reliable, although Deutsche Bank makes no representation as to its accuracy or completeness.

Opinions, estimates, and projections discussed constitute the current judgment of the speaker at the time of recording. They do not necessarily reflect the opinions of Deutsche Bank and are subject to change without notice. For further important information, please visit research.db.com.