AI's Power Problem

In this podcast, Motley Fool analysts Nick Sciple, Seth Jayson, and Tim Beyers:

• Discuss the vast sums being invested in power infrastructure, and whether current plans will be enough to meet demand.
• Cover the opportunities and complications from filling the void with existing and emerging nuclear technology.
• Play a nuclear-themed game of Faker or Breaker.

Don't wait! Be sure to get to your local bookstore and pick up a copy of David's Gardner's new book -- Rule Breaker Investing: How to Pick the Best Stocks of the Future and Build Lasting Wealth.

To catch full episodes of all The Motley Fool's free podcasts, check out our podcast center. When you're ready to invest, check out this top 10 list of stocks to buy.

A full transcript is below.

This podcast was recorded on Oct. 27, 2025.

Tim Beyers: Which AI companies are getting a power up, you're listening to Motley Fool Money. It is Monday, October 27th, and I am your host, Tim Beyers, with me, our long time fools, Nick Sciple also supports our Canadian Services, and Seth Jason has been covering Emerging Tech as long as I have, maybe longer. Both have forgotten more about power and the energy markets than I will ever know, which is why it's great to have both of you here. Today, we want to talk about a different element of the AI story, and it's all AI all the time, and it's exhausting. But we need to talk about how and where we'll get all the power needed to make sure our soon to be robot overlords are serving us well. Nick, let's start, talking about the AI power problem here, I want to see you up with this. The apparently, we ought to be adding 80 I guess it's gigawatts, I don't know if it's gigawatts or gigawatts. I can't get any of this right, but of new power generation capacity a year to keep pace with AI, as well as cloud computing, crypto, industrial demand, and electrification trends, and this is according to a consulting and technology firm, ICF. That sounds big, I mean, is this doable one? If we do it, how do we do it?

Nick Sciple: It is doable, but they say the most dangerous words in investing are this time is different. When it comes to the US power grid, the era we're entering is different than the one we've been in for the past, call it 20 years. If you look 2005-2020, energy consumption in the US grew just 0.1%, essentially flat for the better part of two decades. There increases in electricity demand that were offset by things like efficiency improvements, structural changes in the economy. You've heard folks talk about the US shifting from a manufacturing economy to a service economy. All that has reduced the load needed on the grid over the past 20 years. Here, as we enter the decade of the 2020s, that is changing the emergence of AI in 2020, data centers as a portion of the overall electricity demand for the US grid were less than 2%. Now, if you look at current estimates out there, expected by the end of the decade to be somewhere between 9 and 12% overall electricity demand. We've gone from a regime where we've been treading water, maintaining the current grid to now we've had a new sources of load coming on from AI that we're scrambling to meet that demand. That's why you're seeing basically all hands on deck, whether it's Google signing deals for hydropower, Microsoft signing deals for renewable, reactivating existing nuclear plants and what we'll talk about later, increasingly moving toward new small modular reactors all trying to address the needs that are coming to the grid from AI.

Tim Beyers: Seth, you're my guy that I rely on for the reality check here. Tell me how big the gap is between what we say we're going to do with AI and bringing, power capacity online to service all of this AI.

Seth Jason: That's a big is the question?

Tim Beyers: What's the cold cup of coffee here?

Seth Jason: That's the question mark. If you look at what is being pledged to be built, it's probably much bigger than that 80 gigs that you said. I mean, every week, Nick had the greatest line about this. Every week, these AI CEOs are playing the congratulations, you said the biggest number. I want to point out that they actually sell doom as part of the PR they do. When they say, oh, AI is an existential threat, they're really trying to tell you that's how good their AI. When they try to tell you that AI is going to use up, like all the power, they're really trying to convince you that that's necessary. Now, you've seen a huge increase in demand, and some of these data centers really do suck an enormous amount of power compared to prior ones. You're talking about even smaller ones, 100 megawatts or something, small compared to what they're promising. Using as much as entire towns used to use and so this is really incredible. Then when you look at the product we get out of it, you might because it's hard to be like, do we really need AI videos? Do we need to use you might ask yourself that question. One of the things I think we don't know and we won't know until we're in the middle of building this is how much demand there is really going to be. The reality is that electricity is really expensive, even if you cut a sweetheart deal with somebody to get it cheaper, which most of these data center companies do. But as you both know and everybody out there listening knows, AI efficiency is also a real thing. Every time a video comes out with a new era of chips, it uses a lot less energy, and then there's always software and other optimizations they're doing constantly every year. What you have is basically two curves, and they're picture them maybe as hockey sticks, maybe not, but they're exponentially growing right now. You don't know when they're going to flatten into S curves. But where they cross over, one of them is, like, electrical demand for MI, the other one is power generation that can survey. We don't actually know that and so right now they're leaning heavily on trying to get gas to do most of this because that's the easiest way to get a reliable baseload. But GE Vernova what do they have like a three year backlog, four year backlog right now in turbines? By the way, they're publicly saying, we're building capacity, but not that much capacity because we don't want to spend too much building capacity and over build capacity. We'd rather keep our fat profit margins. Thus, everybody is saying, we got to go back to nuclear.

Tim Beyers: GE Vernova ticker is GEV and when you're referencing this, just to be clear, because I'm a dummy about this, when you say gas, you're talking about gas turbines.

Seth Jason: Gas turbines, and there are other makers. GE Vernova one of the premiers, and if you're in the US, you're going to try to buy from them because you're probably going to have to pay a heck of a tariff on anything that comes from overseas.

Tim Beyers: Nick, I have heard a lot from both of you in the prep for this. There is a whole boatload of discussion about nuclear, so I'm going to give you another stat here. This is called apparently the Cascade Advanced Energy Facility. This is in Washington and it'll be built by apparently a utility called Energy Northwest.Amazon is going to fund this, it's going to deploy 12 small modular reactors. I guess these things are called SMRs. Tell me what an SMR is, should I care? Where are we on the nuclear spectrum here.

Nick Sciple: Small modular reactors, as the name might describe, are smaller reactors than your traditional large scale nuclear reactors of the past, just for contexts, one of these large reactors, probably about a gigawatt of electricity generation. These small modular reactors, you're looking at 300 megawatts and last, there's also micro reactors that are in the 10, 20, 30, 50 megawatt range, significantly smaller than these historical large nuclear plants. The idea with these small modular reactors is that you can manufacture the components at a centralized manufacturing hub and then ship those to the construction site and assemble them there that save some of having to reinvent the wheel every time you invent one of these nuclear every time you build one of these new large scale nuclear plants. One of the issues that we've seen at least with reactor construction in the US going back the past 20 years is these plants because of the bespoke nature of them and that you're spinning up the nuclear workforce to support them each time you build one tends to run a lot longer than the original estimates for deadline and also come in significantly over budget. The hope is with these small modular reactors, you can get some of those costs under control, get an order book in place and start to get some of that learning curve to bring down the costs. That's something that if you look in other markets like China, they've been able to build nuclear reactors back to back to back using the same workforce and bring down costs over time. The US hasn't had that same success, but the hope is with these new small modular reactors you'll be able to do that. Now, there's dozens of designs out there and you see more of them coming in the public markets almost every week, it seems like in the past month or so. However, the companies that look most interesting to me are these private companies. X-energy is one of those you talk about, that's the company supplying the small modular reactor. For Amazon, Amazon certainly supporting this is actually not going to be the first of a kind deployment of the X-energy reactor. The first of those is going to be with Dow Chemical in Louisiana, they have the construction permit. It is now under consideration by the Nuclear Regulatory Commission, expects to have a final decision on that by November of 2026. If they can get that moving forward, then Amazon looks like it's going to have an order book in place as we move into the 2030s to hopefully bring those costs down and start stamping these things out.

Seth Jason: We're looking here at behind the meter, we're talking about co locating. Is that one of the differences with the reactors. The idea that instead of this thing being out there on a grid or something and everybody can maybe get some of it, that some of these will be stuck, next to the factory, they might have their own battery backup or something, behind the meter, so that they have their own load provider right there. That's some of the story with this.

Nick Sciple: That's an important factor as well, you can fit these on existing sites near the data center and you get around competing with the retail electricity customer for this electricity. That's one of the challenges of attaching these large data centers to the grid is all of a sudden you're adding this additional large load that is competing with you and me trying to have hot water and keep the lights on and watch Monday night football and that thing. To the extent that these large tech companies can fund these reactors, have them behind the meter, where they actually own the power, it can potentially get around some of those conflicts.

Seth Jason: Should point out that grid hookups for data centers, the lead time on those is right now a lot longer than even getting a gas turbine. It might take you three years to get a gas turbine, it might take you five years to get a grid hookup. That's where we are right now.

Tim Beyers: I want to make sure I understand this properly. The idea here is behind the meter, meaning not on the grid. That's what we're talking about. Am I hearing you right?

Seth Jason: In some cases, that won't be the case, up in Canada, where they're adding to an existing nuclear plant with SMRs, those are presumably going to be tapped into the plant which is then feeding the grid so this is a smaller one. Presumably, you could have it behind the meter providing just for you.

Tim Beyers: Already, this is incredibly complicated. There's a lot of rules around this. It sounds like there's also if I've heard you correctly, there's a lot of potential configurations here as well. There's lots of things you can do with this, so talk to me a bit, then, and maybe, Seth, I'll go to you on this one. How much do I want to believe or should I believe that nuclear is going to play a role here in the AI build out? Because I two things I've seen, and what you just said is one of the things like it takes a really long time that's the thing. Another is that we're going to repurpose different facilities that either went dormant or didn't work. It looks like Brookfield Asset Management is doing one of those types of deals. Then we have maybe the worst thing I've ever heard in this entire setup, which is now we're going to have SPACs that are selling, I guess, the nuclear promise. One of these is a company called One Nuclear Energy, which is going to go public through SPAC. What should I believe, Seth?

Seth Jason: Well, it's good to depend from place to place, I think Nick is right. Nick, whenever I communicate with him, he's always got the soberst and the best ones to discuss, but this is really complex. You talked about that Brookfield deal, that is a site which has been sitting there empty and been getting rained on. For what is it nearly a decade? Even Brookfield has an out, if they get there, they say, woof, we can't use what's here. Apparently, they can get out of it, on the other hand, restarting palisades up on the shores of Lake Michigan is a much easier lift, and they're on the way to doing that to getting that nuclear plant restarted up there. They're working on restarting some three mile island capacity to I believe it was Microsoft who did that so it all depends. It's really specific from case to case and the issue is that like Nick said, there's dozens of these SMR companies. A lot of them have super cool ideas. If you listen to what they're all saying, almost all of it sounds really good. But it's not nearly as simple as they all make it sound, and Nick has said a bunch of times, and I agree, most of these are going nowhere. A few of the better ones may get somewhere. I'm always on the side of it's going to be more expensive and take longer than people think, but I think some of it has to happen.

Nick Sciple: The VC Summer reactivation, Brookfield. I think it's a great illustration of where we are in the cycle, where we let off talking about we were in a period for multiple decades of stagnant energy demand, and that coincided with the VC Summer and we talked about, as well, that many of these projects go over budget and take longer than expected, and that's really what you saw with the VC Summer program. The aditial expectation was that it was going to be $9 billion to complete the entire facility. After, I think it was six years, they had reached their $9 billion milestone back in 2017 and they decided to abandon the project. Here we are in 2025, where we're looking for all hands on deck to get access to any possible type of energy capacity. We can and you see Brookfield, which owns Westinghouse, which went bankrupt after the previous project was canceled coming back in to try to bring this facility online. I just reflects the power needs of these AI data centers and that we'll take any additional power that we can find, and that includes reactivating shutter nuclear plants and picking up sites that had been abandoned mid construction's really all hands on deck trying to meet the power demand coming down the line. I think today, we're looking to reactivate all the sites that we can in place, keep coal and gas plants that were set to be shuttered extended to maintain that power production. But we're going to run out of that type of easy ish to access power capacity pretty soon. As we move into the 2030s, that's when you really will need these small modular reactors and really new forms of of energy production to come online to meet the growing needs of AI and that's not without even getting into. I mean, Jensen Wang and everybody is talking about that we're going to have humanoid robots in the 2030s. We're going to have to power those things as well. This is going to be a long haul toward just growing energy demand and more and more applications that are more and more power hungry.

Tim Beyers: Well, up next, we'll talk about some fakers and maybe even a breaker or two. You're listening to Motley Fool Money.

We're back, we're going to play some faker breaker for those who don't know how this game works, it's very simple. We are going to talk about three companies that are either a faker or maybe resemble a rule breaker. A faker is a company that's growing really really fast, but unsustainably so or there's a ton of hype around this that I mean, it looks real in the moment, but it's a Mirage. Three companies, you tell me faker or breaker. Seth, I'll start with you Oklo Ticker O K LO. faker or breaker. What do you think?

Seth Jason: I got to give it faker, I'm ready to get some hate mail, I'm sure. In any of this nuclear space, I want a company that knows what they're doing has been around the block already. All of these are going to ring up the cash register the wrong way for me. This is I love that everybody's so excited. New engineers are like, hey, we're two guys MIT. We can come out and do a new flash old type of nuclear reactor, but I'll believe it when I see it.

Tim Beyers: Nick, let's move on to another one here. We're sticking in the nuclear sector. This one feels a little cheeky to me. New scale power and has decided to co op to the very clever ticker SMR, faker or breaker. Are they the leader here?

Nick Sciple: I'm going to go faker. In some ways, you could point toward new scale as a first mover in small modular reactors. The company was founded in the early 2000s, was the first one of these small modular reactor companies to get licensed by the nuclear regulatory commission in the US. It actually has some amount of revenue, unlike pretty much every other small modular reactor company on the public markets, that said it's tied to a small consulting contract in Romania. It's new scale small modular reactor design, had some partners in Utah partnered with it. However, those folks pulled out back in 2023. After, surprise, the estimates for the cost of what these facilities would actually came in significantly higher than any of its partners, we're willing to spend. That said, the stock has gone up a ton since then, again, because of the hype around small modular reactors, but I don't think new scale is significantly closer to deploying a reactor today than it was five years ago, and it's in a very very crowded field.

Tim Beyers: Seth, close out here. Nano Nuclear Energy, ticker NNE. faker, braker.

Seth Jason: Still a faker to me. It combines two my favorite things blast from the past nano. I feel like I'm back in 2005 when nanotechnology was a very hot topic. Just about every company you looked at that had the word nano in it was something you needed to scream and run away from. Like Nick said, most of these have not much in the way of revenue, if anything. Again, I'll believe it when I see a little more proof of projects.

Tim Beyers: Fair enough. I mean, what I'm hearing is that this is something that is really going to accelerate into 2030, but in 2025, maybe slow your roll just a little bit.

Nick Sciple: If I was going to talk about the designs today that I think have a chance to win, I mean, many of them aren't independently publicly traded companies, but small modular reactors under construction in North America. Today, there's the BWRX 300 design being built at the Darlington site in Canada. That is a GE Hitachi design, GE Hitachi partially owned by GE Vernova, the big gas turbine manufacturer, BWX technology, a company I've talked about in the past is one of the key suppliers for building that. That reactor getting a lot of support from the Canadian government. I think that's one design to look at. The other area I would look at, as well are designs that are working together with the US military. Last month, actually, it was earlier this month, I believe. The Army announced Project Janie, which will seek to deploy micro reactors at US military bases here in the next couple of years. There were eight companies selected as potential contributors to that, and I think those would be the companies to look at to list those out. It's Antares, nuclear, BWX technologies, General Atomics, Cairo's Power, which has partnered up with Google, Oklo which we talked about earlier with a little bit of a side eye, Radiant Industries, which actually has a reactor design under construction that will supply an Air Force base here in the coming years, Westinghouse Government Services, which is, again, that Brookfield affiliated entity that is participating in the reconstruction of the VC Summer site and X-energy, which is, again, contracted with Amazon that we talked about earlier. Most of these companies are private, but I think if you look to what the military is doing, it can give you a sign of some of these areas where we might see construction earlier than some of these I guess SPACs that have come public in recent months.

Tim Beyers: Up next, we take a look at what's coming down the pike this week. Lots of earnings, you're listening. Motley Fool Money.

Fools, it's going to be earnings week for the next few weeks. Let's quickly go around the horn here before we preview tomorrow. Seth, an earnings report you're looking forward to this week. Who's reporting, who do you got?

Seth Jason: A couple, and they're pretty different Cadence, selling the software that is used to design chips. Interested to see what their.

Tim Beyers: Ticker Ks CDNS. Yep,.

Seth Jason: But their demand looks and they've been subject of tariff fights. Nucor, I'm interested to see what steel demand is like and what profitability is or isn't given their advantaged tariff situation since they're in the US here.

Tim Beyers: Nucor, Ticker NUE. Excellent. Nick, what do you got?

Nick Sciple: I'll be looking at Dexcom, tickers DXCM. They'll release their third quarter earnings results on October 30th. Dexcom, one of the leaders in continuous glucose monitoring devices in a duopoly with Abbott Labs freestyle Lebe system. Dexcom has had a challenging last few quarters. Last year, lots of pressure from the onset GLP-1 drugs. This year, some execution missteps that they've transitioned from selling in the durable medical equipment channel to now selling primarily through the pharmacy channel, and then their CEO and chairman Kevin Sayer had to take a leave of absence last month for medical reasons. This is a time where a solid earnings report could really reset the narrative around a business that I think has lots of growth potential in the years to come.

Tim Beyers: I like it, I'm going to say Microsoft. I'm endlessly fascinated by Microsoft anyway, but I want to hear what they have to say if they're going to say anything about the OpenAI relationship. Probably, maybe not, I don't know, but I just want to hear more. Plus, I'm very curious to see how much they spend. For tomorrow, will the government shutdown cause more concern for investors? What are we learning from the latest earnings reports? Emily Flippen, Jason Hall, and Keith Spice, will walk you through on tomorrow's edition of Motley Fool Money so please stay tuned for that.

As always, people on the program may have interest in the stocks they talk about, and the Motley Fool may have formal recommendations for or against, so please don't buy or sell stocks based solely on what you hear. All personal finance content follows Motley Fool Editorial standards and is not approved by advertisers. Advertisements are sponsored content and provided for informational purposes only. See our full advertising disclosure, please check out our show notes. Seth, Nick, thanks for joining me. Our engineer today, as always, Dan Boyd and our producer Anna Chuck Lou, Forseth Jason, and Nick Sciple, I'm Tim Beyers. Thank you for listening to Motley Fool Money. We'll see you next time, Fools, move on.

Nicholas Sciple has positions in BWX Technologies and DexCom and has the following options: short January 2026 $70 puts on DexCom. Seth Jayson has positions in Amazon and Microsoft. Tim Beyers has positions in Alphabet and Amazon. The Motley Fool has positions in and recommends Abbott Laboratories, Alphabet, Amazon, BWX Technologies, Brookfield Asset Management, Cadence Design Systems, Microsoft, and Westinghouse Air Brake Technologies. The Motley Fool recommends DexCom, Ge Vernova, and NuScale Power and recommends the following options: long January 2026 $395 calls on Microsoft, long January 2027 $65 calls on DexCom, short January 2026 $405 calls on Microsoft, and short January 2027 $75 calls on DexCom. The Motley Fool has a disclosure policy.