AI Infrastructure (Part 2): Powering the AI Buildout

00:21, Chris

Welcome to William Blair Thinking Presents, it’s November 2nd, 2026.

And today we're picking up where we left off in our recent conversation with Tim Mulrooney on the midterms and data centers. For part two, Richard de Chazal, William Blair's macro analyst, is back as host. And then we're joined by Jed Dosheimer. He's group head of energy and power technologies research.

They'll dig deeper into the energy and infrastructure side of the story and what it could mean for the AI build out and investment.

So, Richard, I'll turn it over to you.

00:50, Richard

Thanks, Chris. Thanks, Jed, for joining me today to discuss part two of our recent call with Tim I had last week on the midterms and the impact on data centers. And that was the basis of the note which the two of you guys put out in the week before, earlier that week.

So, I think the genesis, though, of your note seems to have been that investors are looking at the midterms potentially as a kind of referendum on data centers in their states. They're worried about the impact of these data centers on their electricity bills, water bills, and energy consumption.

So, are you seeing these midterms as sort of an important test of political resistance? And what do you think that impact could have on demand? Or are we going to start to see projects being delayed and investments starting to slow down?

So that's one aspect I'd like to talk about, but I'd also like to maybe get into, as well, discuss the bipartisan American Affordability and Jobs Act, or the BAAJA, which is a proposed federal permitting and energy infrastructure bill.

Let's just kick off with the first bit, though, on data centers and the midterms as an issue. Is it just because voters are getting their utility bills in the mail and those are being jacked up? Is that really the concern? Or are they concerned about, yet again, looking at a big boom where we're going to be socializing the costs and privatizing the gains, just like we did in the global financial crisis with the investment banks?

Or maybe that's a little bit premature, but I guess what I'm really asking is, what do you think is going on? I mean, is it really a voter backlash on data centers? And is this something that's going to continue to snowball if we do see anti-data-center-type candidates winning in the elections?

03:13, Jed

Yeah. Okay. So you got a lot there. Let's try and pick it apart, sort of one by one.

First, I think it's important to acknowledge how we got here. And I think you touched on a couple of these points. And I want to acknowledge that the fears that are driving the NIMBYism are real. And I think this is where big tech really did a poor job. So I'd give them an F in terms of thinking that people are just going to take this.

And I think that you have a growing mistrust in the country in terms of government and technology. And so when you kind of frame it as, so what's the pitch been?

So yes, we wrote about pain at the plug, I think a year and a half ago, or at least a year ago. Data centers are not what's driving up electricity pricing in the U.S. today. Will it be what drives it up in the future? That is the load growth, and very much could be. But if you look at what's been packaged, it's your children's future is going to be replaced by compute and that you're going to have to pay for that and finance that. And, oh, by the way, your water systems are going to be polluted.

Like, that is an incredibly powerful message that has resonated deeply with the distrust that, I'll say, has grown in the U.S. And I don't have empirical evidence to show that, so that's kind of my M-plus-one. But I think, from a world perspective, maybe one of the consequences of social media is the ability for these ideas to spread more rapidly.

So I think that's what you're seeing. I don't think midterms actually address this at all. I think midterms are going to do something very different that I'll touch on.

I want to acknowledge that those fears are very real. And, oh, by the way, then you have some insiders come out and say that there's a 10% to 20% species risk from this technology, which contributes to that. And I have no idea whether or not - those percentages seem like they were kind of plucked from the air. But, you know, if you have a non-zero chance, to me, this feels eerily similar to when I was a child growing up in the early '80s.  We had to hide under desks around this idea of nuclear war. And so that fear was very palpable.

And to me, I think the fear around data centers and AI seems very existential. I think it's important to kind of determine what's real by using data and facts and sort of what's hyperbole and driving things.

My take is that, kind of separating out what midterms actually give us insight into, I think midterms are really an insight into 2028, and a potential power shift, and what that could mean.

If I look at what a Democrat might want versus what a Republican might want, I think for Republicans, probably a consequence that we'll see play out in this election, the midterm election, is you want to reduce inflation, you want to increase buying power, and give people confidence that the future is better. I think that they're largely failing at that, whether it's their fault or not.

And I think that if I'm a Democrat, I need to probably do some of the same, but I want to push more of the problems onto Republicans going into 2028.

So I think midterms are probably going to be a litmus test on whether or not things can get done in the next two years in the U.S. And as you and I talked about at a separate event, my view is that if we look at growth – so there's two ways of dealing with this.

If I want to lose weight, there's two ways that I can address this. I can eat less or I can work out more. And the eating less is less attractive, although it has a bigger impact on my stomach.

And so I kind of look at it from a political posturing perspective. We can either grow our way through our problems or we can make cuts in terms of dealing with some of what's causing inflation.

Well, it's going to be the deficit. And so how do I get those back in check? I can either grow my way through that or I can cut my way through that.

And from my position, the cuts don't seem like they're palatable. So growth probably becomes a more attractive engine, and therefore AI becomes the greatest growth engine post this election to try and get there.

So I'll pause there. I know there's a lot of stuff that we can dig into and kind of take in the direction that you want.

So I look at it as midterms are going to be a data point versus a finish line that we're crossing. And quite frankly, you mentioned the BAAJA bill. We put a note out on this. I think that is much more consequential than whatever the election results will be.

08:43, Richard

Right. So on the election results, basically a lot of rhetoric. It's a bit early for anything too fundamental to come out of it, but it's sort of a way marker towards 2028, when it could actually be more consequential?

09:01, Jed

Correct. And I think all of this comes back, as I like to joke with you, I think understanding the role of energy as a substrate for our economy is the most important thing.

So I kind of always bring it back to energy policy. And what does that mean? That's why I said I think the BAAJA bill is consequential, because if you can't build large projects, then it's going to be really tough to get growth.

09:29, Richard

Okay. So let's go there quickly.

So, what, you know, the BAAJA bill, I call it the "BAAJA," so the BAAJA bill, maybe just quickly, in plain English, what is it?

09:41, Jed

It's permitting reform.

So if we look at the current administration pulling permits on already permitted projects, the pushback is, "Hey, you're changing the rules of the game after we've decided on those rules."

And I agree with this. I don't think this is a partisan statement. You have to have a set of rules that you can build around, or you're not going to have any projects, long-tailed projects, that can be built.

So the Democrats want to get solar and wind, that this administration has pulled permits on reinstated, and change that so that neither party can choose to pull permits once a project has already been greenlit.

I think that's a good thing for Democrats and Republicans because, by the way, if you have a shift of power and you have a nuclear or a gas project that might be seen as more right-leaning, you don't want the other side pulling that, or we're just going to be in this ping-pong game back and forth every four or eight years.

The other thing that's really important is that NGOs can't use lawfare or challenge projects that they're not part of in perpetuity, which is one of the reasons that, if you look at large projects, China in particular is very good and efficient at getting those projects stood up and built, particularly around energy infrastructure.

And the U.S. is particularly bad at that. And so removing that ability to sue indefinitely on a project, I also think, is a positive.

So this is a bipartisan bill that was brought forth by four different senators: Senator Mike Lee on the energy subcommittee, Senator Sheldon Whitehouse, Senator Heinrich out of New Mexico, and Senator Capito in West Virginia.

So it is truly bipartisan. I think the challenge is going to be, on the midterms, if one side gets an overwhelming victory, how will the negotiations go? This has not gone to the floor for a vote.

So something we're tracking, which we think is consequential and meaningful is a future signal or indicator in terms of can we actually build these projects.

12:08, Richard

Right. So if there's a Democratic sweep, do you think there would be significant amendments to this bill coming through or, you know, can kind of sort of push through as is?

12:19, Jed

My guess is, so our take is that if it can't get passed in this Congress, which I think there's like 15 days left in that which ends in January, so voting days.  I think the likelihood of it kind of falling apart is high.

So if either side feels... you mentioned a Democratic sweep. I think if a Democratic sweep that positions Democrats as having more power, they ask for more.

Or if the GOP feels that they didn't lose as badly as predicted in the polls, either of those scenarios on either tail I think are bad for a bipartisan bill getting passed.

13:11, Richard

Okay, let's change course a bit to sort of the energy and productivity side of things, because I think you had some interesting commentary and thoughts there.

And your argument is basically that energy’s availability is a constraint on growth, while AI’s economic upside depends on productivity.

So if that's the case, maybe I'm wrong there, you can correct me. But, you know, what do you think are the power and productivity assumptions that need to hold true for all of this AI investment to really be justified?

13:46, Jed

I think one of the greatest shortcomings around our understanding of the economic models is that energy is infinitely substitutable, or that because capital is only created through the transformation of that energy into goods or services, which is determined by productivity. I think you and I agree on that.

And debt is just a future lien on energy. And that's why, as I look at the amount of government debt in the U.S., to me, the common denominator to all of the macro problems is the same, which is building out the energy systems. And if we look at electricity, that's the filet mignon of energy. It's a highly reformed value of that energy. So it's been upconverted to that highest form.

So, therefore, if we take a transitive property, we could say that building out our power and electrical infrastructure is the only way to reduce debt, or therefore grow GDP or productivity.

I kind of use oil as an example, where a barrel of oil energetically costs about $50 to take out of the ground in the Permian. Neal Dingmann is our expert here, but I think he agrees with that.

The price is the market demand, supply and demand, which today is around $100 per barrel. But the value is what you can create with that. And so there's 2.8 labor years in every barrel of oil. So if we use $50,000 as the median value of income in the U.S., something that costs $50 is contributing $150,000 of value into our economic system. And that's why we've had the Industrial Revolution and the harvesting of this energy-dense form of productivity.

AI just takes that to another level. If I look at these, these are all just log scale, where I'm now taking compute. Or what is compute? Compute is really decisions.

So I'm taking decisions and I'm looking at the floating logic operations that could fit within a barrel. And I'm getting, and I'm using the barrel to keep the metaphor consistent, but all I'm doing is just getting a 10x increase or 100x increase in terms of value creation each time that I look at that.

And so if I look at how do I increase productivity, by the way, in both oil we're removing the negative consequences in terms of exhaust streams or potential pollution, etc. And if I look at AI, I'm going to do the same. But if I'm looking from a purely economic perspective and ignoring any negative externalities associated with that, we continue along that curve, which means that the productivity or growth engine can largely be reduced.

Even though AI, I think, is only around 3% or 4% of GDP, it's 90-plus percent of GDP growth. So it presents really the only major growth engine that we have available to us to grow our way through some of the challenges that the world faces.

Which then comes back to energy. Energy is your gating factor. If we look at data centers, we're 100 GW short by 2030. That's our entire nuclear fleet we have to rebuild, which can't be done within that time frame.

So your gating factor becomes the amount of electrons that can be produced and transmitted on the grid, or even a behind-the-meter, off-grid solution.

Productivity, you have a second derivative, maybe it's a third derivative, of the funding of all of this, right?

And so if we're looking at, and I'll defer to you, you're an expert here, but in terms of credit spreads, credit default swaps, on the ability to fund, what's happening with yields, the pressure that that's placing in order to actually stand up the infrastructure.

And this is why I come back to the BAAJA bill and why it's so important. Because if the true cost to actually stand up infrastructure and deliver that power is going to be far greater than, say, another nation's ability to do that, then that's going to cost more. That's going to put more upward pressure. And you have this self-fulfilling feedback loop that we seem to be caught in the middle of right now in terms of where risk is shifting. Is it shifting to the infrastructure? Is it now shifting to the debt or the financial system to fund this? Is it shifting to compute? And it's been a little bit of a game of whack-a-mole to try and figure out where that's going and where the opportunity and alpha exist for our clients.

18:51, Richard

Yeah. I mean, I think that's right. I think what you're seeing is clients, and particularly in the debt markets, credit spreads and CDS spreads, those are all sort of widening.

It does feel like investors are questioning the timing of all of this, in the sense that when are you going to get this productivity relative to the expenses that we're taking on today for that investment? Are you seeing anything in terms of that timing, or what are you seeing that would help you, give you some a better picture of how that is developing?

19:26, Jed

Well, there's two things here.

I think one of the things that's important is, yeah, I've been doing this job in some context for about 30 years, just under 30 years. So I've gone through a few different cycles. And I think that, you know, a lived experience, we tend to, as humans, prioritize over something that I might read historically, for example.

And so within that time frame, there's been largely a rules-based order that we've been operating in. That's determined where, you know, when you tend to hear an analyst say, "This time is different," you kind of roll your eyes. And that's usually, you know, a lot of people say, "Oh, that's the kiss of death."

And so I think at a base level, it's important to look at what is different this time around or is this is just another cycle. And, you know, your job focuses on a lot. From a macro perspective, you get to look at these different cycles. So I'll defer to you on some of this.

But one of the things that I've noticed is, you know, if we look at post-World War II and the order that came, which, by the way, energy determined the outcome of World War II, and if we look at those rules, we've largely operated in the same system with a couple of deviations.

Maybe we call coming off of a gold standard in 1973 as sort of a major change to a fiat based model. But if we look at really the last 30 or 40 years, we've operated within a system with kind of guardrails associated with that. And now, to me at least, it looks like that move toward a path of globalization has been ruptured, that the need for a multipolar economic system seems to be the divergence or the pathway that we're taking.

I'm not saying that's good or saying it's bad. I'm not placing any of those views on that. I'm just looking at it as is.

And if that is the case, then if the previous operating system, the code, was on a truly globalized system, and now we're moving to a multipolar system, well, that code is going to shift. And so I look at what does that mean?

In particular, this is kind of the basis of our "Pain at the Plug." The level of reshoring that has to occur, I feel like COVID was the first shock that really showcased the fragility associated with global supply chains. And I think that AI is really the second shock that's kind of showing that fragility associated with the power systems. And the consequence of that is a level of reshoring to shore up supply chains in the case of moving to a multipolar economic system.

Of course, globalization and specialization will be the most efficient and be the way that we can... so if we look at it from a "how do we get the most out of our natural resources that we have available to us," a truly globalized system would actually be best and support the highest level of population, etc.

Moving to a multipolar system, a more nationalistic system, is going to constrain that. You're going to move higher on the resilience scale and lower on the efficiency scale. And that's going to require a level of buildout.

And to me, that's where inflation is going to be sticky. Because if I look at the level of energy intensity per dollar of GDP, that has to go up and not come down.

And therefore, the only other option would be how much efficiency can affect that. But efficiency, in my view, reduces the resistivity of current flow. It actually doesn't reduce the amount of current.

So if we look at that all together, I think there's some real nuances here that suggest that we're going to be in this...

I kind of come back to where I started with. The common denominator of all of this is building out the power infrastructure. That is the only way through any of the major problems that we have.

23:52, Richard

Turning to your stock coverage, what do you think, you know, what we're seeing today? I mean, I guess maybe not so much the midterms, but BAAJA. Who are the sort of beneficiaries there, and maybe who or what are going to struggle a bit? And has any of this been reflected in current valuations?

24:17, Jed

Sure. The obvious natural beneficiary was natural gas.

So I should also take a step back. Building out your power systems, we overbuilt on the peak or surge, which you can do really well with renewables, and kind of underinvested in baseload.

And so there's a need for more baseload right now because a data center is going to operate on a 24/7 power need. That could change a little bit as we go from training to more inference in terms of that role. But generally, you want more baseload.

So if we think about what can handle baseload, I'm going to have natural gas, coal, nuclear, whether it's a light-water or a small modular, geothermal, hydroelectric.

Those tend to be in... You can couple using a decent amount of batteries with renewables. And the issue is the intermittent issues that are inherent to renewables.

So that's kind of the cohort that we're looking at from a generation perspective.

And I think early on, we just saw the basket of valuations across the board kind of expand on all of these.

And I think what's since happened is, particularly around debt and the cost of this, natural gas and the turbine suppliers have been the ones that have remained with higher valuations that have expanded over the past couple of years.

Although we do have clients that are questioning how sustainable this level of valuation expansion is. And then in nuclear, you have the same. But I think, more recently, we've really seen that risk-off trade that's occurred and valuations contract quite a bit.

There are still some names we like, and we've taken an approach of choosing the leaders in each one of the categories. Now, I don't cover the electrical switchgear space, with the exception of GE Vernova, that has exposure there and is one of the leaders in that space. But you're also equally seeing that cohort that's benefited from multiple expansion there.

So I think you are seeing that. The question is, are we in for a contraction? Is this going to be temporary? And what is the durability associated with the earnings power or EBITDA generally speaking at each different level of the supply chain?

I think that's the big question that's being wrestled with by large institutions in the marketplace right now.

27:14, Richard

Do you want to venture a reply to that, what your thoughts are or how that's going to play out?

27:18, Jed

Sure. So this is why I brought up, we started with BAAJA, you know, to kind of come back around.

I think that if you can't get projects stood up, then that's going to be a problem. And therefore valuations won't be sustainable, or the durability will be undermined.

Conversely, being able to...

So who benefits? In my opinion, the beneficiaries are the American public, the suppliers throughout the supply chain of this infrastructure, and AI companies in general. I would think consultants and lawyers would probably be the consequence of that bill going through. And if I look at, so to me that's the most important signal associated with the ability for whether or not... I'll use a U.S. baseball analogy. What inning are we in? Are we in the sixth inning? Are we in the fourth inning, third inning, or the ninth inning? And I think there are different investors that are betting on what the time frame is.

My take is that getting that bill passed moves back several innings, right? So we go from maybe the seventh-inning stretch back to the fourth inning in the game because you're going to lower the cost and you're going to be able to stand up timing within a time frame that's going to have a greater likelihood of filling that gap in terms of the power requirements and getting the productivity out of these LLMs. And I think that's a positive.

29:03, Richard

Definitely, yeah. You get a bit more clarity there going forward.

29:05, Jed

And I think, if it was probably not... you know, I don't... I think we're getting at: you don't have conflict during a period of abundance, only during scarcity. So I do think it's really important. I think this touches on some very core themes. And I think this is critical, not just on how can my portfolio outperform, but this touches on democracy. This touches on a lot of big topics. And that's why I think it's really important to understand energy policy and what the first, second, third, and fourth derivatives are to all of this.

29:54, Richard

Okay. Well, so we're running out of time. But just following along from that, maybe just a final question.

What are the sort of signposts you're looking at here for whether this is playing out as expected, or we're starting to see too much political resistance, or not enough?

And in terms of the AI demand, is it beginning to slow down a bit, or is it, you know, Jevons paradox still full-on driving this thing? How are you thinking about that?

30:28, Jed

Well, I think Jevons is fundamental. So, and for the listeners who don't understand that, as we move to greater efficiency, we won't actually use less. The absolute amount of, in this case, energy will be greater. And William Jevons came up with that in, I think, the late 1800s, or mid-1800s, around the development of coal. And it's held strong. That's why energy density is so important. And never before have we gone to a diffuse system and seen growth.

So I think that's an important underpinning, just to frame out and kind of understand what's possible and what's not possible, which means that I don't see a way of saving our way through some of these challenges. Which doesn't mean that energy efficiency is not important. It clearly is. In the U.S., the compute technology, are what's allowing the U.S. to remain in the race vis-à-vis China, which has more raw power but lower conversion efficiency.

I don't think AI can be... that genie is out of the bottle. And I don't think that can be put back in. So the question then becomes, is it ramping? Is it slowing?

I think when we look around, there is a little bit of game theory playing out, and musical chairs, where not every LLM, or hyperscaler, is going to win. That's not my area of expertise, so I defer to Jason Ader and Arjun, who focus on that. And maybe Sebastian. My area is to try and come back and look at what are the infrastructure and power needs.

And to me, I kind of come back to that 100-gigawatt shortage. That is a major shortage that's hard to comprehend, you know, how to fill all of that.

And you're not going to be able to... if NVIDIA comes out with some new chip that's 100 times more efficient than the last one, my view, using Jevons, is that that will only increase the use models. It won't actually save the total amount. That's kind of the point of that.

So I feel pretty comfortable that the needs are there.

Alternatively, I look at AI as the only real growth engine to deal with the fiscal debt that we carry, and I say that collectively for the West, and if that's your growth engine and power is kind of the gating factor, then the need for that infrastructure and the power is the logical conclusion of where that bottleneck and that you're going to require that.

Could we see... and this is probably more in your domain... what's the risk of recession? Do we go to a stagflation-type situation where you get maybe an aberration of pause, or do you get a resumption of growth post-midterms?

I kind of come back and say, well, if you do get a little diversion off the highway that we're on, you still have to come back onto the highway. There's no other path where this all of a sudden changes and just goes away.

And therefore, I feel comfortable that the durability associated with the next 10, 20 years... if the last 20 years was about virtualization, this is about returning to physical.

34:25, Richard

Excellent. So definitely a lot to digest. There is always, speaking with you, Jed, so I enjoyed that, and I think our listeners will as well. So thank you very much for joining us today.

34:38, Jed

Thank you.

34:38, Chris

All right. Thanks, Richard and Jed.

If you haven't already, be sure to check out our previous conversation with Tim Mulrooney for more on the political backdrop surrounding data centers in the midterms.

Thanks for listening to William Blair Thinking Presents.