At a Glance

• Utilities have met AI’s growing energy demand (so far) largely through technology and process improvements, particularly in transmission projects.
• Construction timelines for new gas plants are increasing from about four years to more than six years as turbine demand strains the supply chain.
• Utilities and tech companies are forming creative partnerships and exploring new operating models as both sides rethink how to build, invest, and scale.

From their business models to their risk appetites, technology companies and utilities are so dissimilar that partnering with each other is like trying to combine oil and water. And yet, as the artificial intelligence boom’s massive energy requirements pressure electrical grids, these two industries’ marriage of necessity has become one of the most fascinating and consequential business developments of the past two years.

What follows are highlights from a discussion between leaders from Bain & Company’s Technology, Energy, and Strategy practices. In the year since we gathered leaders for a similar discussion on this topic, much of the landscape has changed. Generative AI user numbers continue to grow rapidly. To win the AI race, tech companies are investing hundreds of billions of dollars in product innovations and data center infrastructure. Meanwhile, geopolitical tensions and escalating uncertainty around tariffs have significant implications for tech and utility companies’ supply chains and capital projects.

This latest conversation explores the tension between data center energy demand and finite energy resources, and what it all means for the future of tech companies, utilities, and the businesses and consumers that rely on them.

Read excerpts from a transcript below.

Dunigan O'Keeffe: To a certain degree, if I was reflecting on where my mind was coming out of last year's conversation, it was like, “Oh my gosh, we're about to hit a wall, and all of this planned spending was going to come to a halt.” We haven't hit a wall. Are we closer to that wall? Are we seeing real hard trade-offs that are either slowing down data center projects or having downstream implications for other sectors [and] other users of energy? We haven’t heard of this bottleneck becoming a true bottleneck yet, so I’m curious what’s enabled us over the past 12 months to keep going.

Grant Dougans: You know, utilities are innovative. And they move quickly. And I think that utilities have been moving mountains to find ways to connect all of this load.

And so, whether that be new types of generation sources or [the] distribution side of generation sources, in many cases, perhaps some of those plant retirements that were contemplated might get delayed some to keep the lights on, if you will. So, we haven't seen it at least this year, but you know, it's sort of taking it every year or two at a time in terms of making sure we have enough capacity along the way.

But yeah, Matt, I invite your comments from the customer side.

Matthew Crupi: Yeah, I mean, I think the good news from an efficiency perspective is that those big players are being very, very creative around what are all the things that they can do. So, you're seeing advances in things like liquid cooling. You're seeing software innovation, and the way that you actually write the code to use the chip more efficiently, and that that can actually drive out 10%, 20%, 30%, 40%, 50% efficiency gains. You're seeing energy-efficient chips being launched that combine CPU and GPU technologies.

And so, I think everyone is recognizing that in the short term, we have what we have. But in the long term, recognizing some of the other constraints that we see down the pike, we have to be thoughtful around both the infrastructure that we're using, the silicon, and then the cooling technologies that we use—and that those can help us make up for some of the bottlenecks we might see in other places, like long-term power generation or other things.

You know, if we really hit that point where supply and demand are out of kilter, you may need to see new construction to be able to take advantage of some of these technology-side efficiency improvements that we’re seeing.

You’re seeing software innovation, and the way that you actually write the code to use the chip more efficiently, and that that can actually drive out 10%, 20%, 30%, 40%, 50% efficiency gains.

Dunigan: What about on the energy provision side or the transmission side? What sort of innovations are we seeing coming down the pike or [what sort of] plan for there to meet this opportunity and set of challenges?

David Armstrong: There's only a discrete set of tools that the utility industry has available to them. We've got natural gas, we've got nuclear. We talked a lot about those last year. And I think the near-term investments into natural gas are real, and you're seeing the supply chain start to respond to the demand. That's both costs going up and the length of time it takes to build a turbine and get one delivered to you going from roughly four years to build an unplanned gas plant to more like six or seven. So, I think that those tools are consistent.

On the transmission side, though, you are seeing some advanced conductoring, new tower designs start to get deployed, and those are, especially in a world where it's difficult to expand rights of way or get new rights of way, making the best use of an existing footprint.

Those technologies are real. I think you're seeing more adoption that maybe a couple of years ago a fairly risk-averse industry was not yet willing to start to put investments down into, they are actually starting to use those, if only because necessity is the mother of innovation.

The length of time it takes to build a turbine and get one delivered [is] going from roughly four years to build an unplanned gas plant to more like six or seven.

Grant: I think some of this, too, is good, old-fashioned critical path work. Transmission has been a pretty quiet, I would say, part of the industry for quite some time (for the last decade-plus), kind of more compliance oriented. We didn't have a lot of load growth for most of the last decade, and so, all of a sudden, that part of most utilities has gone from sort of a quieter space to the central part to enable all this load growth. And so, they're really scrutinizing everything in terms of, hey, when I think about all the steps that, if it's a three- or four-year timeline to build a transmission line, you might only be in construction for nine months of that four years. It might be permitting and engineering, all that kind of stuff. So, it might be a bit more risky, but there are probably things I can do to really start to kind of neck that down to get that done in 24 months, for example.

Rethinking the utility business model

Hannes Grassegger: Maybe as we talk provision, also—we saw this already a year ago, right—it's not going to be, maybe, the classic model of how utilities have thought about building out the grid. And it's not as utility driven, so it's going to require coalitions of players, utilities, EPCs [engineering, procurement, and construction firms], like Grant just said, joined in with what the utility has done historically.

David: Yeah, on that point, I think the evolving business model of a utility is increasingly a service integrator between developers, EPCs, customers, and the grid—and really figuring out how all of those pieces fit together to get the customer speed to service. That looks very different than a decade ago, and that sort of service integrator role is something that I think they're struggling with. What is the right operating model to develop that? What are the right skill sets that I need to staff that new operating model? And how do I make sure that, ultimately, they don't want to be the bottleneck to what is a major national security concern? And that's requiring them to evolve many of their practices—not just their technology but how they operate.

It’s not going to be the classic model of how utilities have thought about building out the grid. And it’s not as utility driven, so it’s going to require coalitions.

Matthew: That sparks maybe a related thought, which is, I think one of the things that we're seeing from the user side is exploring repurposing of existing facilities that have great access to power infrastructure, availability of industrial-scale water access. They have the zoning and environmental permits in place so that they don't have to go through that process.

You're seeing a bunch of older industrial sites that people are saying, “What's the best use of this now that there's a data center opportunity or demand here? Because I have a lot of things in place, and I can go fast.” And for some of it, it's a total repurposing of that spot. For others, it's an incremental revenue stream on top of what's already there because they have excess capacity or everything else.

People are getting creative because they're sort of seeing that there either are constraints in place or there will be constraints in place. And therefore, [they’re] trying to think through what are all the levers that I have? And I think it’s opening up a new opportunity which fits in really nicely with that sort of service integrator kind of role that you talk about, which is you’re trying to find and broker new opportunities that may not be as obvious vs. just get the power online.

Tech companies developing their own power sources

Dunigan: It raises to me a question about what is the future competitive landscape here? Like, to what degree will some tech companies or a tech company decide that they actually want to get in energy provision directly themselves?

Matthew: I think what I would say is a lot of the big tech firms are evaluating control, and so they're looking at new sites where they can build from scratch using all of the latest techniques to be as efficient as possible to include their own energy provision.

So, either repurposing of existing sites where that's already there—but they can build next to it—or, longer down the pike, looking at small nuclear reactors as a kind of new source. And you see there's a bunch of different designs here, and the regulatory piece hasn't a hundred percent caught up (i.e., not everything is approved that may come online eventually). And so, there's a lot of time to value before all that comes to be part of the calculus here.

But definitely, people are looking at securing their own energy supplies, and they're doing it in creative ways, and not necessarily going through the traditional routes to do that.

Grant: I think it’s a really interesting space. There’s a lot going on there. As companies try to start doing more and more self-developed power, you will get more and more into the kind of stuff that makes utilities slow, which is like building things, permitting things, engineering things, [and they] will become more like the kind of things that tech companies need to do. And I think the question is, “Is there a reason to think a tech company can build a dozen square miles of solar more quickly than a utility can?”

David: It's can they do it, and do they want to do it? Somebody quoted earlier, $300 billion of capex spend amongst the hyperscalers this year alone. These are traditionally asset-light businesses. And so my question is, if you are already deploying that much capital in the data center infrastructure themselves, how much additional capex do you have the appetite to deploy in energy, which is not really your core business?

In many ways, yeah, you can put a gas plant next to a data center and colocate it. But plants go down for maintenance. And so, if you're not grid-connected, you've got to have enough resiliency built in to have several units. We have seen a few examples of clients that have built quasi-colocated, behind-the-meter plants, but they still have grid connection just for that reason.

Those kinds of questions are new and novel, and I guess I'd ask if that's something there's appetite for. I mean, if necessity is the mother of innovation and this is a major constraint, I see no reason why they wouldn’t. But it certainly seems to buck parts of the traditional business model.

Is there a reason to think a tech company can build a dozen square miles of solar more quickly than a utility can?

Grant: I think that what’s being tested is to what extent can nonutility players do those things more quickly? Is it really just utilities are risk averse or is it actually, there are genuinely structural challenges associated with building this much power this quickly? In which case, what is the role of consortiums? Right? I think that's what we'll have to see, and maybe that’s for the 2026 edition of this conversation.

Risks, investment models, and lessons from telcos

David: I want to go back to something that Hannes said earlier, which is the VC [venture capital] model. There are no two business models that I think are as dissimilar as the utility business model, which has asymmetric upside. There is no value to a regulated utility in pursuing a speculative investment because they only make a regulated return on that investment.

On the inverse, the tech industry is built on making speculative investment and having asymmetric upside if that investment pays off. And so, I think you know what really needs to happen to solve this is the balance sheets and risk appetite of the tech companies need to fund, in the short term, the massive amounts of investment that are required.

And then, if it is true at the end state that tech companies don't want to operate these for the long term, utilities are wonderful operators of these assets and stewards for the public good. There is a challenge, but I think actually an opportunity as well, in what is really like an oil and water marriage of necessity in these past couple of years.

Dunigan: There may be an analog there if we go back a decade or so to tech companies and telco. There was a lot of existing telco infrastructure which was really required to enable us all to adopt the Internet and get the data fed to where [it needed to go], but it was going very slowly until the iPhone hit.

And then there was this aha moment of, “Oh my gosh, this is an order-of-magnitude different sort of data.” It triggered new commercial models for telcos. We went from having minute plans to plans which are fundamentally grounded in data. The telcos themselves ended up doing a whole bunch of investment in last-mile connectivity specifically because that's what really made the difference.

There's a lot of competition in there, but what's interesting is, quietly behind the scenes, many of these large tech companies have built their own global connectivity. Now, mainly, it's wholesale connectivity; it’s not all the way to the users. But quietly, they've skimmed off the infrastructure that was most essential to de-bottleneck and most valuable for their business model. And there's a lot of reasons why utilities and telcos are very, very different, but I wonder if there's something about unbundling this a little bit and saying, “Well, what are the pieces in the value chain?”

And I agree it's not going to be turbines.

Where [it's], "No, I think there is a new capability here, energy provision capability of some sort for data centers, that actually it's going to shift to the companies running the data centers because it's going to be so essential."

Or if I own eight data centers in some part of the country, am I going to think about connecting them on my own? Because there's some advantages to that as opposed to only doing it on the grid or something.

But it's just curious to me that it seems unlikely, with all of this capital, that there's not some boundary shifting at some micro level of some sorts because we'll actually realize, “Oh yeah, that's better for somebody with all the risk appetite to take, and we're not going to be allowed or really want to go there.” I’m curious if that triggers any other reflections from folks.

It seems unlikely, with all of this capital, that there's not some boundary shifting.

David: One thing I think the utilities are asking of the tech companies is, we build to the peak, and the vast majority of the time, what we've built to serve that peak sits unutilized on our system. And so the question becomes, what would it look like? We talk about sharing of risk. What would it look like for the data centers to not operate at full capacity, particularly as you move from training loads to inference loads? Is there an opportunity?

I think recent studies suggest that even a 0.5% curtailment of the hours of data center operation could get 90 incremental gigawatts of data center load onto the system. And just by way of comparison, there's more than 30 gigawatts nationally on the system today. And so, “curtail-ability,” “interrupt-ability” is essentially treated like a resource for utility and demand response. Up until now, the phrase 24/7 power demand has been parried about as a way to say, “These data centers are different and can't be interrupted.”

I think there's, at least as we understood it, some interesting technological questions about, “Can you distribute load, move it around to respond to demand input from the grid?” But I don't know where the tech companies are at in their willingness.

Dunigan: Yeah, that's fascinating. So, effectively run the data centers really hot at night when nobody's using it, and then figure out how to take workloads down during the day when other people have demands on the network. Something like that, in theory.

David: Exactly, and it's even fewer hours than that. It's when everybody gets home to turn on their TVs and oven.

Dunigan: 5 [p.m.] to 7 p.m.

David: If you could take those two hours offline for the hottest 10 days of the year, that really drives all of this incremental value because those are the people that are as spiky as they are. And frankly, you've built your system to be able to ensure that every time you flip the switch at five o'clock, that the light turns on.

Matthew: I think intellectually it's a very interesting push, and I think it would be challenging in practice to do but could be done. I think every business and every hyperscaler will have workloads that don't need to be 24/7. If we could figure out what those are and put those all in a certain place, that could easily be the thing that gets turned off.

And then the question is, is there a way to write the software to actually move the need around so that you could always have something offline? Nobody's been forced to write it that way. It’s not necessarily the most efficient way. But if you were taking total cost of ownership in place, and you were thinking about it differently, are there ways to do that where you're pulling other parts of the program at certain times from certain places and not from others?

We know moving workloads is very, very challenging. But if you were working towards that going forward, could you be thoughtful around where you placed incremental [load] and at least do that for some portion of it? And then over time that would sort of increase, increase, increase.

Dunigan: Is there a potential silver lining, like, could we encourage so much investment in upgrading our grid and our power generation assets across the country that, in fact, there was a dividend that went back to consumers in some form or fashion? Or are we just always imagining such constraint that we'll be navigating trade-offs as we go?

I think in the case of telecom—just to use that example, I didn't know I was going to go there—but I think the proposition: It's a smaller proportion of your real wallet now than it was 20 years ago—in part, because that infrastructure was invested in to facilitate this past wave of digital innovation. And, of course, you need the right regulatory construct. There's a whole bunch of other things which you would need to have that happen. But man, it'd be awesome if there was a dividend through all of this that was going back to the consumer.

Grant: We'll have to see. I think a lot of the things we've talked about today, different trends are still playing out in real time. The question about what I would call equity issues, if we get to a point where we don't have enough power collectively and we're starting to see existing power get moved from one group of customers to another for any reason—that's going to cause equity challenges, and that will have political challenges to go with it. That's probably a challenging road to go down.

There's another world in which you say, to your point, we actually can find a win-win where we're building new investment being funded by new players that are helping to harden the grid [and] make the grid more stable.

And there's everything in between, right? And state by state, it'll be different. Politically, getting to a spot where you are seeing inequity issues of any kind is very challenging.

Negotiating amid ROI uncertainties

David: Agree, Grant. Where I spend a lot of my time is supporting clients in their active negotiations with hyperscalers. There is a willingness on the part of hyperscalers to pay their fair share. The question is, are they willing to pay more than their fair share? And the answer up until now has been, “No.” But it's a rapidly evolving space. I think where there may have been a bit of slack in the system or geographies where folks were more willing to take a flier on the next big thing that is data centers, I think you're seeing more utilities start to grapple with the fact that these are big investments for an uncertain payoff at the end.

And so, as fewer of those geographies are open to the data centers, the question becomes, “Does the balance of power in terms of who holds the leverage shift?” Because I think, Dunigan, to your point, at the end of the day, this is largely a commercial conversation between a data center and a utility. Until now, the hyperscaler can say, “Oh, we’ll pick up and move somewhere else.” Increasingly, somewhere else is having the same problem that the first party was having. I think how these commercial discussions play out over time in connection with how the regulatory landscape evolves, and then in connection, Grant, to your point, which is how quickly can we build this stuff? I think those are the three threads that we probably pick up on in May 2026.

I think you're seeing more utilities start to grapple with the fact that these are big investments for an uncertain payoff at the end.

Matthew: I was going to go back to one thing that David said around, “Are they willing to pay more than their fair share?” It's an interesting world in the data center because you're running workloads, and those workloads generate economic value. Sometimes, the cost of running the workloads is not that much, and so if you were to double the cost of doing that, you'd gladly do it because you're going to make so much money on top of it, even if that's not the most economical decision to make. Which is why people don't like to move workloads because the risk of screwing up a workload that's working perfectly well is high, right?

At the same time, you have tech companies whose valuations are driven by growth and the expectation of future growth. But at some point, there's the expectation that that growth turns into profitable growth. And so, I think we're in the stage of data center and power build-out that's, “We're in a race, and I need to capture as much market share as possible. Yes, we're thinking about the economics of all this, but we don't really have the models all built out that tell me the ROI of the $100 billion that I'm going to spend this year.”

But at some point, that's going to happen. I do think there's the ultimate question, which is economics will come into play for everybody as they're asked to monetize their investments by the street. The question is, how quickly will that happen. And what does that look like? And frankly, how are they seeing their business models actually delivering value? Is it as lucrative as they thought? Is it not? What's the competition look like?

I think there'll be a little bit of that that plays out here, which is to say, if they're still making money hand over fist, they'll pay whatever they need to get it. And as that competition gets tighter and they're actually asked to drive real profitability, they'll start to be a little bit more sane in how they're making those decisions. And they'll drive real business plan discussions around what to do and what not to do, which I think today is happening, but it's a little bit fast and loose.

David: I will say, that is exactly what terrifies the utilities executive, Matt—the idea that there are more hyperscalers chasing the demand than will actually shake out in the end. Because [utilities are] building several power plants for each of those. What happens when half of them turn out to have made the wrong bet?

Dunigan: What? Wait, wait, hold on. What does happen in that event? Is that the consumer surplus I was looking for?

David: The opposite because they're still on the utility’s rate base, and who's paying for that but the existing customers? Unless the tech company has financed and paid for all of those assets and they're sitting on the system, great—that's a reliability benefit for everybody. But if they're unrecovered, the utility will either go bankrupt or recover them through existing customers.

That is the ultimate risk question that utility executives are grappling with, which is, “This is great that we are having a grid expansion opportunity. By the same token, if this dries up and this is not as durable of demand as I thought it was, we're going to have a real problem on the back end.”

Dunigan: In a weird way, the utilities need to be very clear on picking the winners or at least being clear that they're doing business with the winners. If you're a second-tier player, you might need to pay up to get the kind of attention you need from the utilities, if people feel there are risks that you might not be around to pay the full asset that's going to have a lifetime of decades and decades, right?

David: And for a utility, paying up may not be price. It may be duration of commitment, and those that are willing to commit for longer are going to be better customers than those that are willing to pay a super high price for the next four years, but without any durability of that demand going forward.