Big Tech Goes Nuclear

In this podcast, Motley Fool analyst Asit Sharma and host Mary Long discuss the collaborations laying the groundwork for the coming “Intelligence Age.”

Then, Motley Fool analyst Sanmeet Deo and host Ricky Mulvey chew over Tesla‘s “We, Robot” event and take a look at the humanoid landscape.

To catch full episodes of all The Motley Fool’s free podcasts, check out our podcast center. To get started investing, check out our beginner’s guide to investing in stocks. A full transcript follows the video.

This video was recorded on Oct. 17, 2024.

Mary Long: We’ve got the power. You’re listening to Motley Fool Money. I’m Mary Long, joined today by Asit Sharma. Asit thank you so much for being here on this lovely Thursday afternoon.

Asit Sharma: Thank you Mary but you know as much as I like you I have a thing against people planning earworms in my ear. Just before I start a conversation and now of course I’m hearing a certain song from back in the day with your Intro. But nonetheless let’s proceed with our conversation.

Mary Long: I have to be honest, I almost I debated singing it, and then thought too much. I’ve already had that song stuck in my head all morning as I thought about that intro line. There we go. Torturing myself and you and others. Hence the Intro line Amazon just became the latest Big Tech company to ink some nuclear power deal. The biggest part of this particular deal is that Amazon’s teaming up with Dominion Energy to explore the development of small modular nuclear reactors in Virginia. If you’re thinking that you’ve heard a similar story before, it’s likely because you have. Amazon is not the first hyperscalar to go this route of teaming up with an energy utility company, Microsoft and Constellation Energy earlier last month announced that they were going to restart three Mile Island. The site of the most serious nuclear meltdown and radiation lake in US history. Alphabet announced earlier this week that it’s teaming up with the privately held Kairos energy to build seven small nuclear reactors, Nuclear and Big Tech, but we’ll just stick on nuclear. Nuclear is the common denominator in these stories but that power source is certainly not without its controversies Asit and there are a lot of other types of energy. Why nuclear?

Asit Sharma: Well Mary you and I were chatting before the show and you threw out like wind and solar. As alternatives, and that’s interesting because those come up in the conversation a lot. Why not just things like wind and solar, also non carbon forms of energy like nuclear. The reason is we’re just not going to get there fast enough. That is when I say we, I’m talking on the level of a society, but if you look at it from the hyperscalars vantage point, what they need to power data centers to keep that cost from getting out of hand, just can’t be developed quick enough if they go simply a wind route or a solar route or a combination of those so a natural alternative is nuclear energy. I’ll note that all three of the big hyperscalars have made commitments to have non carbon transitions in their energy sources. Amazon may be the fastest to get there but each of them is telling shareholders, look, we’re not going to build out data centers for AI and have to build new cold power plants to do that. We’re going to go other routes. This is why the focus is now on nuclear, as you mentioned, that’s not without its drawbacks.

Mary Long: I’m glad you mentioned timeline because the narrative around a lot of these deals is often exactly as you described, like Big Tech needs more power for AI, and they need it right now. This constellation energy Microsoft deal, they want to restart the three mile Island reactor by 2028. That’s really not that far away. Nuclear power projects do have a reputation for running very long and being very over budget. I got a follow up question that’s coming, but the first question for you is, how realistic are the timelines of these partnerships?

Asit Sharma: I don’t think they’re that realistic. Number 1 I think part of this is that we are doing the best we can, we as a society again to try all types of non carbon energy sources at the same time that we are placing unprecedented demand. If you think that the demand is not going to grow at this linear rate, I would just suggest, look at crypto mining from a few years ago. That was the first blush that we got that maybe computation could be something that places a stress on our power grids, and now we have generative AI though there will probably be something else down the road. While society is trying to solve these problems, the big giants are putting money in today. They’re putting out these timelines, number 1 to try to do what they do in their own work, which is to set something ambitious, to move with agile precision, to innovate, to iterate all this venture capitalist, quasi big tech, corporate speak we hear about just moving fast Meta famously move fast and break things. We don’t want to break nuclear, that might gate.

Mary Long: It’s important to that don’t break that.

Asit Sharma: But they want to be aggressive with the timelines. The history of nuclear power, though, is one of over extended budgets, missed deadlines, especially if you’re talking about large scale nuclear reactors, which is not really the case here, but we will talk about very soon here is also something that’s unproven and I think may not meet the stated deadlines.

Mary Long: To get to that follow up question that I promised, even at their most generous. Are these projects, these promised timelines are they even moving at the pace that Big Tech really wants them to? We say, we want the power now. Well, now isn’t actually now, and it’s probably not four years down the road. How do you square what we need now with the realistic timeline?

Asit Sharma: One way that we square it is to keep developing technologies that have nothing to do with the energy source, but are focused on reducing the power demands within the data center so innovations is in chips, innovations is in the way we cool servers, innovations in the way we build up server racks. Those are all ways that we can at the margins, help the problem along. But yesterday wouldn’t be soon enough when companies like Amazon and Microsoft, Oracle, you name it, look ahead to what these demands will look like in four to five years. I think there’s going to be a price reckoning, so someone will have to pay it because we’re going to stress the grid as it exists. With that higher consumption, higher demand will come higher prices. Who picks it up? I hope most of it is picked up by big businesses, enterprise businesses that are doing a lot of AI computation, but part of me already understands that we the consumer are going to pick that up in one way or another. It’s really to everyone’s interest in our society to try to figure out how we can innovate, not just with the power source, but every part that’s involved with training and inference of these AI workloads.

Mary Long: I want to focus on a detail of difference between these three different types of partnerships that we’ve mentioned. Microsoft’s deal with Constellation Energy focuses on reviving a currently closed but already massive plant. Amazon and Google, on the other hand they want to develop this new generation of small modular reactors. What’s the difference between those two paths?

Asit Sharma: The small modular reactor path might be more viable in the future. There are many differences between SMRs, we’ll just call them that. These are described in various terms with various acronyms, but let’s stick with this one. Many differences between this and your conventional large scale nuclear power plant. For one, the power output is smaller so it’s a smaller setup. It’s what it’s purports to be a small reactor, it is modular. By modular this means that it’s almost something you could produce in a factory. In fact it can be produced its parts in a factory setting. That’s not the case. If you look at any big nuclear reactor you might have seen driving around those ones that tower above us and emit those ominous clouds of steam, those are built on site, Mary they’re highly complex with these small modular reactors, you can assemble them, and you can assemble them as your energy needs scale up so that you can get started quicker, and this is one of the things Amazon is gunning for and other companies are gunning for. Let’s get that first bit in. Let’s get it started so we prove the concept, and then we can add on more energy. They typically are thought to be a little bit safer than large scale plants.

They have a different technology of cooling, which has to do with just the intrinsic way the water cools in the system. The other thing that I think most people will gravitate toward too is that they can be put adjacent to a data center. They don’t have to be plunked in the middle of nowhere so there not something that we typically associate with these large scale reactors out in the open, a source of concern for different community. Many fronts, they make sense. Again these two aren’t without their own drawbacks.

Mary Long: You read about this at all, and you’re going to come across some metrics that might be hard for the average person to visualize megawatt, kilowatt, gigawatt, this type of stuff. Often to underscore and illuminate the type of to illustrate how much energy AI takes up. We use comparisons. A single ChatGPT prompt consumes the same amount of energy that it takes to power a light bulb for an hour. A Google search by comparison is the equivalent of powering a light bulb for two minutes. I’ve also heard this described in terms of water. A 100 word email generated by something like ChatGPT 4 requires 519 milliliters of water, which is a little bit more than a water bottle. That helps me visualize what this energy consumption looks like. But I think for a lot of people, we’ve become so used to asking the computer a question that we take for granted, how that happens, and you’d be forgiven for not thinking about how the process that actually powers that. What does that chain look like? How does this innocent little asking of a question, typing something into Google or ChatGPT suck up so much energy and water?

Asit Sharma: Mary when I types that question into ChatGPT. We’re sending that question over to a computer on a server, and that is starting a series of calculations so those calculations are interacting with stuff that’s stored in memory. What is that thing that’s stored in memory? It’s a picture of the world. We think in terms of large language models, those are representations of different objects, different concepts, different words, statistical representations. We all understand that ChatGPT is predicting what should come next in a sequence. It has to constantly interact with that model when we send the question over. Think about a lot of computation that’s involved to pick the different parts of that model to form the response. That’s the inference part. But then on the chip level too there’s so much going on. If you picture a GPU, what you’re thinking about here is a chip that’s performing compatations, but it also has to access memory in generative AI to answer the questions. It could be going off of the chip to access memory.

There are some chips, some GPUs that have memory that’s built in three dimensional space around the chip. Picture data zinging around the chip, then going up in a stack to access a bit of memory and come back down, and then having to go to a whole cluster of other GPUs. Some GPUs now are linked together in the hundreds and in the thousands, Elon Musk has built a version of this. What I’m trying to communicate here or help listeners visualize, is that a simple question involves a lot of mathematical operations and a lot of memory because we’re relying on the computer to access its vision of the world that we’ve built by training it on billions and billions of parameters. It’s way different than what we used to do with computers, which is just to type in a request and go to something that’s already indexed, that’s static, and that’s Google Search for example, just consult this index and pull me a result. That takes so much less computation.

Mary Long: Looking back on the third quarter, utilities was the best performing sector of that period. Energy meanwhile was the worst performing sector of that period and yet here we’ve got these stories where big tech companies are pairing up with utility company, energy companies to move forward, our progression toward AI and this so called intelligence age, if you’re an investor looking for a picks and shovels play in the AI game, how do you play this? Especially considering that just over the past quarter, these two sectors that are close have also had pretty different results.

Asit Sharma: I think energy companies are interesting in so many ways because they are getting more and more requests to help solve this puzzle. EQT is a company that is basically a natural gas company, the whole pipeline of natural gas, but it has a role to play in generative AI as well, as an alternate source for energy. Suddenly that becomes an AI play. Looking at regional utility companies is so interesting. There are certain parts of the United States where some data centers are being built out. You can see this going on in many southern states, especially Virginia. Northern Virginia is like an amazing global hub for generative AI, and it is flush with data centers. If you’ve ever driven around Northern Virginia, for those of you who haven’t, then in the Pacific Northwest, you know, we have projects that are going to come online there.

I like looking at these different hot spots and seeing, which are the utility companies and energy companies that are playing in this space? Because inevitably, as you alluded to Mary, they’re already in all talks and partnerships with the Big Cloud companies and so they have this new attractive revenue source over the years. That’s one way to play it. Then keep an eye on the small companies to I think you referred to new scale Energy, which is one of the companies that’s publicly traded that plays in the space of small modular reactors. I think it’s symbol is actually that SMR. Keep an eye on those, but as you’ve mentioned so much of this hasn’t been realistic in terms of time frame. You have to be a patient investor. If you’re going to pick up some of these small companies, don’t expect the moon tomorrow, be ready for some volatility and be ready for some ups and downs as they receive contracts, constructions delayed, they get more contracts. It’s going to be a while before these companies, very few of them which have discernible revenue yet to be like free cash flow propositions, but it is fun to keep an eye on them.

Mary Long: You did a great job explaining earlier how we get from me typing a query into ChatGPT to the process behind that. We didn’t quite touch on the fundamentals of nuclear physics, but a guardian Warren Buffett tells me that I should only invest in companies that I fully understand. Do I need to have a PhD in nuclear physics to be investing in this space at all?

Asit Sharma: That’s so interesting. I think that Warren Buffett said that he only invests in companies he understands and has very humbly owned up to missing some great companies because he didn’t understand them. I think a good way that we can all have a fruitful twist on what Warren Buffett says is to only invest in the companies that you’re curious about because you can learn as you go along. If you’re curious about a company or a technology, it’s OK to invest it in it if you don’t understand it. As long as you’re willing to put in the work over time, if it becomes material to your portfolio to make sure you understand it because the last thing you want to do is to wake up with a great winner in your portfolio and not really know how it makes money, because then you won’t know what to do with it. Do I sell it? Is it going to go further from here? I don’t know what to do now because I didn’t put in the work. If you’re buying companies you won’t want to think about later, that could be counterproductive. I happen to have the Oxford Dictionary of Physics on my bookshelf. Not that I consult it, but I was thinking of this when you and I were planning the podcast, how to dust that puppy off, and maybe it’ll help me understand a little bit more about this industry. Maybe that approach is a little bit better for most of us. Not all of us can be Warren Buffett, because he can totally avoid things he doesn’t understand and still make so much money with all that capital and all that acumen.

Mary Long: Asit I already thought you are a renaissance man. You write. You invest. You have all these various hobbies. Now I’m learning that you also have a physics textbook in your library.

Asit Sharma: Let’s correct that. Now you also know that I have dusty books on my bookshelves. Appearance only gets you so far unfortunately.

Mary Long: That’s part of it. I like to think that if you have enough books, you can learn through Osmosis sometimes.

Asit Sharma: Totally.

Mary Long: Asit Sharma thanks so much for reading us and for the insight into what can be a pretty complex topic.

Asit Sharma: Thanks a lot Mary. This was a ton of fun.

Mary Long: Today is the last day to vote for Motley Fool Money as Signal’s best money and finance podcast. If you enjoy the show, and haven’t yet had a chance to cast your vote for us, you’ve got till the end of today to do so. We so appreciate you listening always and we really appreciate your vote. There will be a link to vote in today’s show notes. Reminder case you forgot you will have to share your email to prove that you are a human not a robot. Speaking of robot. Elon Musk likes to talk a lot about humanoids. But Tesla is not the only company making progress in this space. Up next Ricky Mulvey talks with full analyst Sanmeet Deo about the present and future possibilities of these not quite human robots.

Ricky Mulvey: Sanmeet his conversation was about to be very different if we recorded it just a few days ago. Last week, Tesla demoed its optimist Robots at the We Robot event, in addition to the Robot taxis and the new Tesla van. Here’s the layer is that these robots were remote controlled by humans. According to a Bloomberg report, Elon Musk told the engineers, you need to get these robots ready for prime time at the event. The engineer said we can only do that if there’s tela operation, and alas, we have this controversy where Tesla CEO Elon Musk didn’t necessarily reveal that those robots were operated in help by humans during that event. But before we get to that controversy, what was your reaction first to the demonstration of the optimist robots and then how did that change when you learned they were tele operated?

Sanmeet Deo: Honestly, I was like are you kidding me? Apparently, parts of it were remote controlled, but it was disappointing you rolling out these humanoid robots that are supposed to mimic humans, and you got someone in the back just controlling them. Wasn’t so exciting. It was a little bit disappointing, but it also limits the fear of robots taking over if a human is controlling.

Ricky Mulvey: Also there’s a video of the event where one of the optimist robots is making drinks, and this person taking the video keeps asking, are you operated by a human? While it’s pouring drinks, the optimist robot or the person behind the optimist robot. The third person becomes very difficult in these sentences when we’re referring to a tele operated robot. Anyway the robot says, I’m assisted by a human right now. It was revealed during the event, but, Elon Musk he’s a showman. He’s an innovator, and he’s also a showman, was talking about all the solutions that these robots would provide, but not necessarily where they currently were. When you look at the current state of where humanoids are, not just with the Tesla optimists, but also with Agility robotics, which is working with Amazon Boston Dynamics. What can these humanoids do right now and what can’t they do?

Sanmeet Deo: We’re getting to see a new technology grow before our eyes. While humanoids aren’t ready for prime time yet they’re developing at a rapid pace. Many of them can walk, maintain, balance they walk, albeit at a slow pace. They can perform some basic tasks, lifting, moving objects, unloading trailers, moving packages. Logistics environments with AI they’re able to understand and respond to voice commands and learn from experiences. What they can’t do is perform, tac and efficient pace. They say that if humans worked at the base of these humanoids right now, then we’d be fired. They don’t have the human flexibility and fine motor skills for precision tasks. Their understanding is limited. It’s funny because when you watch these humanoids, and as they try to pick up things you have more appreciation for your own hands and how precise they can really be when you’re trying to do very basic tasks.

Ricky Mulvey: The ability to crack an egg, for example, we’re going to see a lot of them show up first. Maybe not walking your dog or babysitting your children, but factories. You could imagine Amazon being very interested in having humanoids working in their factories. Robots that don’t need 401ks, robots that don’t need to take much of a break besides battery recharging, robots that don’t go on strike. In fact they’re working on bringing more robots to their factory with the digit robot.

Sanmeet Deo: Amazon’s journey with robotics began in 2012, when they acquired Kiva systems for $775 million and launched Amazon robotics. They’ve been using robots in their fulfillment centers to move shelves of inventory, palettes, large items, sorting and handling packages. A lot of the robots they have had prior to digit is think of larger Rombas that are these big Romba vacuums that are holding palettes and boxes. Digit is more official humanoid, bipedal humoid, which is hopefully going to improve efficiency automate repetitive tasks, and it’ll be well suited for human tasks. I could also take on some of the more dangerous tasks that humans might be taking and reduce that chance for employees to hurt themselves.

Ricky Mulvey: What is your I know you’ve looked into this space quite a bit. There’s a range of outcomes between, I think there’s a pretty clear industrial use case. Then we also have a use case of humanoids is dog walkers, lawn mowers. Heck they could even be your friend. Where’s your bullishness on humanoids lie? What do you think they’re going to be doing?

Sanmeet Deo: Well if you take a step back a lot of projections are saying that the global humanoid robot could reach anywhere from, 38 billion by 2035, which Goldman Sachs says to other estimates that are over 4 trillion by 2035. Regardless I think it’s going to be a huge market. But where do those humanoids land. What do they do? I think some of the key areas are major job needs, where we’re seeing gaps in employment when it comes to manufacturing, agriculture, elderly care. It said that we’re going to face like an 8 million plus job gap in essential manufacturing. That’s something that humanoids could easily take on as they start ramping up. Just Morgan Stanley estimates that by 2040, United States may even have 8 million working humanoid robots that would have a $357 billion impact on wages. Some of these jobs where it’s the employee safety is of concern of repetitive tasks. The digit actually one of its tasks is literally emptying the tote bags where products are in and putting them away. That is actually done by human right now and it’s repetitive, it’s boring, and I’m sure humans have better things to do than that.

Ricky Mulvey: The robots don’t get bored. I think the concern comes from in my brain is when you start matching these humanoid robots that are physically very capable, and we’ll see as they continue to develop their balance and their ability to perform these repetitive and creative tasks, and these large language models, which are able to make really good inferences. It’s that merging in between them that Tesla is working on and Boston Dynamics is working on. I think that’s where you have the greatest bull case speculation and also the greatest concern of what are these things going to be capable of when we develop a machine that is bigger, faster, stronger, and smarter than you Sanmeet.

Sanmeet Deo: Well one of the companies that you haven’t mentioned is Figure, which is a private company, and they are working also with Open AI. They’re working with BMW. Their founder has worked on some other interesting new age creations or innovations inventions, I should, I guess say, and he’s a little more tempered than Elon Musk. He’s a lot more rational. Those actually impressed me the most. They’re doing some great stuff. That’s one to look at too.

Ricky Mulvey: These start-ups for humanoid robots. We have some of the big companies like Tesla getting involved with it. Startups have raised about 1.6 billion inventure capital to develop these bots. But those are for the private investors. Those accredited investors. For the the rest of us. I’m not an accredited investor. I’m in the lowly. Is this is this space investable for me yet or is it too early?

Sanmeet Deo: It’s in terms of pure play publicly traded humanoid companies. I don’t know of many. There are you said the figures, the Gili robotics, all the private companies. Obviously, if you invest in Tesla, it gives you exposure to optimist. But then you’re getting in EVs and autonomous driving, and battery tech all in there. Hyundai Motors actually owns Boston Dynamics, which are famous for the Atlas on the spot and the robots that do all those fancy funny tricks jumping and such. One other area where where private investors could explore retail investors could explore is crowdfunding platforms Republic or Start engine, MicroVentures. It is much riskier than the publicly traded markets, as a whole another game, so you want to really do your research and really look into that. But in terms of other publicly traded investing vehicles, the thing that I’m going to look into more is the picks and shovel stuff, the things that make up those humanoids and those robotics that will power them.

Ricky Mulvey: What are the picks and shovels? What’s powering them?

Sanmeet Deo: Well, with AI you got the chip names, the semiconductor names. The standard ones and Nvidia and the likes. I don’t know specific companies yet, but I’m looking into, AI vision technologies, sensor technologies. Lots of different things that, you know, I’m going to go digging around one day and probably go into a rabbit hole of a breakdown of these humanoids. They do those breakdowns on YouTube and such, and dig into that. But that would be worth exploring.

Ricky Mulvey: For anyone thinking about the crowdfunding stuff, especially if you if you’re a newer investor I would be extraordinarily cautious of getting into any investment where you can where you don’t have liquidity, where you’re not able to take your investment in one day and pull it out the other day where you have things like lockup periods. Because liquidity is a lot like oxygen in the investing world. You don’t recognize how important it is until you really need it. There’s a lot of use cases for humanoids that some of them are scary, some of them are fairly common like lifting things in a factory. Are there any less expected use cases that you’re going to be watching as this technology develops?

Sanmeet Deo: Couple of ones. I’m actually very intrigued by elderly care. I have parents they’re getting older. I know friends that parents that are getting older. Many times they live at home alone. Their kids might be living very far away, so they have a lot of trouble doing basic stuff. That will be an interesting area where humanoids can play a part. I always say too just household tasks. I think I don’t go a day now doing dishes and laundry where I think, isn’t there a humanoid or a robot? They can do this for me because It’s pretty low risk stuff that once you train them up and get them going, they should be able to do. We need a Rosie the Robot from Jetsons.

Ricky Mulvey: I’m OK with a robot crushing a couple of plates if it means I don’t have to do dishes. That’s great. This is going to be the best out for anyone who doesn’t want to do the dishes. That’s a job for the humanoids now. Sanmeet Deo appreciate your time and your insight. Thanks for looking into this technology. We’re going to keep talking about it on Michel.

Sanmeet Deo: Thanks, Ricky.

Mary Long: 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 snow buyer sell stocks based solely on what you hear. I’m Mary Long. Thanks for listening. We’ll see you tomorrow.