Motley Fool Hidden Gems Investing - DeepSeek and AI’s Efficiency Era

Episode Date: January 27, 2025

The tech battle between China and the U.S. is heating up – one player just showed the world that Ai can be done well with a lot less. (00:14) Tim Beyers and Dylan Lewis discuss: - The Chinese app that is shaking up the AI landscape – DeepSeek – and how its leaner, less resource-intensive approach will affect artificial intelligence technology. - Why software prowess and efficiency will be the major themes for AI in 2025. - What to focus on when the affected big tech companies – Apple, Microsoft, Meta, ASML – report earnings this week. (19:15) Senior Fool Analyst David Meier, interviews John Zahurancik Fluence Energy’s President of the Americas, about the utilities side of the renewables market. Companies discussed: MSFT, NVDA, AAPL, ASML, FLNC Host: Dylan Lewis Guests: Tim Beyers, David Meier, John Zahurancik Engineers: Rick Engdahl Learn more about your ad choices. Visit megaphone.fm/adchoices

Welcome to Planet Fitness, a.k.a. Planet Strength. Planet Sweat. Planet Calm. Planet Social. Planet Me Time. And, of course, Planet Value. Whatever you're here for, we're here for it. Judgment free.

Get started at Planet Fitness for $1 enrollment and then $15 a month. $49 annual fee applies. Deal ends September 10th. Deep Seek enters the chat. Motley Fool Money starts now. I'm Dylan Lewis, and I'm joined over the airwaves by Motley Fool analyst Tim Byers. Tim, thanks for joining me on what has proven to be a tech-filled Monday. Yeah, not caffeinated, but I'm very ready to go. Yeah, some energy, some headlines may be getting your attention and bringing some of that energy. The tech battle between China and the United States continues to heat up.

Over the past week, AI app DeepSeek has topped app store charts as one of the most downloaded apps, period. And its rise is leading to some pretty serious questions about the resources and investment in AI, in particular from American publicly traded firms. We are seeing the market process that very much in real time today, Tim. Yeah. And part of the issue is the freakout that China has essentially developed an AI model for orders of magnitude less than the dominant models here in the United States. So you have open AI, anthropics clod, perplexity, all of these others were built with just, I mean, we have spent hundreds upon hundreds of billions of dollars and we are still spending hundreds and hundreds of billions of dollars building out AI models. we have presumed here in the United States, the predominant narrative is these are resource intensive. They take a lot of capital to build out. They consume unbelievable amounts of energy.

And so we have said, look, we have to invest. We have to invest big. It is here in the United States, AI has been a go big or go home story. And then China came along with DeepSeek and the Chinese entrepreneurs and introduced this and said, actually, not so fast. We can make this for a fraction of what you're saying it is. Now, I don't know if you should believe their exact number, Dylan. I think they said something like 6 million. But I don't think it matters. I don't think it matters. I think that we have now on the market a model that is objectively far more efficient than the competitors. And they're kind of, you know, they're kind of tweaking us a little bit. Or maybe I should say, DeepSeek is tweaking OpenAI because they've named

their latest model R1 as what seems to be a very cheeky hat tip to OpenAI's O1. Yeah, I think we will have to dig through a lot of these details, you know, today, this week, And I think over the coming months, because this is one of those stories that is developing very much in real time. And I think this competitor kind of came out of nowhere. And I think a lot of folks in the U.S. press are working to put it together. What we do know, based on some of the initial reporting, is researchers claim it was developed for less than $6 million. That seems low to me. And it seems like, in reality, that is a number that is being bandied about.

The real costs are probably quite a bit higher. But this is a free and open source system. From a lot of the testing that has been going on on the technical side, it seems to be at parity with some of the major models. I have seen reports that if you go into more technical things, more advanced, more industry specific things, it has some more limitations. But I think part of the concern here for a lot of the American firms is for the average use case, it is right there with a lot of the major ones that already have some substantial

market share. There's no question. I think you're spot on about that. And I think the thing I maybe object to broadly, like the freakout we're seeing is like, we're surprised? Why are we surprised? I don't understand why we're surprised, Dylan. And here's the reason why. Whenever you introduce constraints into a system, this is one of the rules of technology. When you introduce constraints into a system, you unleash the possibility of innovation. And this is what venture capitalists love. They love to invest in companies that are building something magnificent with massive amounts of constraints. That's where the biggest innovations come from. Now, let me take you to the backstory of DeepSeek.

DeepSeek comes out of China, where there's a ban. There's literally a ban on the most advanced GPUs, the stuff that has been used, and we are ordering like it's bourbon at happy hour here here in the united states right like it's it's flowing like crazy gpus are everywhere and we're just buying them hand over fist you can't do that in china you can't do that with the most advanced chips you have to work within incredibly significant constraints so what do you do in that situation you work on what you can optimize you take the limited resources you have and then you use clever software engineering to create optimizations that's what deep seek did and you know what you do get like what you said dylan you get parody and

this is kind of the arc of technology innovation you get the big bump you get the hype and then we spend spend spend we get all bought into the new idea the new new thing and we overspend overspend overspend to scale the idea and then at some point cfos look at the spending and say you think we could do a little better than that and that's when software engineers start looking at at the system and saying i bet we could make this better because we don't have as much money to spend anymore and once you do that you start getting innovations like what we're seeing with deep seek. So in many ways, Dylan, my point is that deep seek feels like it came out of nowhere, but deep seek is also something that we should expect because technology always

operates this way. I think speaking of venture capitalists, Mark Andreessen called this a Sputnik moment in AI. This idea that we are no longer looking at American supremacy in this technology. There are other countries that are able to develop this and move this along in spite of some of those constraints that you mentioned, one of the main things that has popped up a lot in the reporting on this is that the compute necessary for what is running on DeepSeq is a fraction of the compute for some of the other systems. And watching the way that the market is processing this, we are seeing big tech companies take a hit. We are seeing some of the chip companies take a hit. We're also seeing energy companies take a hit because there is this

feeling that maybe as we get a little bit more technologically advanced, as other players start coming into the space, some of the energy demands for this technology won't be as big as people have maybe originally thought. Yes. Why is that a surprise? Because those energy demands, I know I'm sounding hyperbolic here, Dylan, but didn't we all realize that like, come on, we're not going to bring 50 nuclear reactors online in the space of a couple of years? I don't deserve any credit for this insight, Dylan. I think this is obvious. And I'll tell you who was way earlier than this than I was, our colleague, Seth Jason. He was like, there is no way we are going to be building out the amount of energy infrastructure required to service all this at the level we are talking

about in the timeframe we were talking about. So then what happens? You have a constraint. Do you keep doing what you're doing and overwhelm the energy infrastructure knowing full well you can't build it out at the level that you want to in the timeframe you want to? Or do you do what the industry always does, which is find areas of efficiency to scale in a better, more economical way. That's what always happens. In a way, I feel like I'm saying, yeah, of course, this was always going to happen. You point out something important. I guess I'll maybe tee up the rest of our conversation around this. I am certain. I have not looked at the markets today yet, but I'm sure NVIDIA is taking a proper beating right now. Is that fair? Some of it's fair. Some of it's fair

because you are probably at a point where the arc of innovation in AI is going to be less hardware-driven going forward. Is it appropriate to be a little bit more skeptical of NVIDIA right now, yeah, I would say that's not an irrational response. But we're still going to use GPUs, Dylan. NVIDIA is going nowhere. Yeah. Yeah. I think what is interesting about this news is it feels a little doomy if you are looking at coverage of US companies today. But to your point about the evolution of technology and kind of the natural way that this tends to happen over time. This feels like great news if you are long-term interested in the development of AI and you want to see this technology flourish, become cheaper, become more accessible,

and have more players that are pushing it forward. Big time. There's no question. This is the kind of breakthrough that unleashes more innovation. It spins the flywheel of AI innovation would be how how i would describe it you have you have constraints now and now you have a player that has discovered a way to get past those constraints that creates a constraint now for the existing us players like well we have to figure this out we cannot be all in on just buying all the hardware and doing you know sam altman can't be like well i'm just gonna raise a trillion dollars which by the way he wanted to do trillions of dollars you know like you You can't just do that anymore.

You have a new constraint. And now that new constraint creates an opportunity for new innovations here in the U.S., but also abroad. Now it's a race. And that is super interesting. And I'll tell you, I am excited about that because let me tell you where I expect the U.S. to show up and deliver some real innovations in this space now. May I introduce you to the open source community? They will be all over this, Dylan, and I think that is great. There is so much investment in U.S. open source and just around the globe, but this is perfect for the open source community to get involved and figure out the next stage of optimizations because what we're building is a foundational layer.

A foundational layer of technology that we can use and reuse to build new things. We're going to have maybe improvements in foundational models that will lead to new types of AI-driven applications. It's a very exciting time. I don't think it's a nervous time. I guess I'm a little off kilter here, but I think it's less doom, more, are you kidding me? this, we are about to see the introduction to take Andreessen's analogy, say like, great, it's a Sputnik moment. Give me NASA. Let's go. Right? Yeah. You're excited. You want to see the launch. Yeah. Yes. I want to see, give me my Saturn V. Let's go to the moon.

I'm with you. And I do understand the concern for investors that are seeing red in portfolios, Because, you know, as we tape, I think a couple hundred billion dollars in market cap have been taken off the books looking at some of the big tech companies. And I tip my cap to the cosmic script writers on this one. Because not only did this news come out this week, but we also are going to be getting quarterly updates and management calls from ASML, Meta, Microsoft, Tesla, and Apple this week. And I am guessing that there are going to be some… All of those companies are very interested in this news. Yeah, I'm guessing there are probably going to be some analysts on their conference calls that are interested in this news. What do you expect to see? We can do a little bit of a big tech

earnings preview here with this news and what these companies are going to be talking about this week. Well, so over the last several quarters, the most important thing I've been watching is capital expenditures. I've been doing this a lot on This Week in Tech when we do the earnings reviews, Tim White and I, and every time I'm looking at, all right, what's the growth rate for CapEx. And it has been extraordinary, Dylan, these last few quarters, routinely over 50% increase in capital investment. Will that continue? It'll tell you the actual numbers will continue to be high because they'll have been before this news came out. But it'll be super interesting to hear what their CapEx guidance is going to be.

And where they're going to point resources over, let's say, the next quarter and next year, next fiscal year. Because I'll tell you, I would expect a little bit of a drawdown in the overall number, but I would expect significant R&D investment because this is – I love that we have reached this point in the AI cycle. at least this is my, let's just do a reckless prediction here because I've done this reckless prediction on this week in tech. We have reached the point in this market where it's less about hardware and it's a lot more about clever software engineering and clever software engineers. And I think that is exciting. It's an interesting dynamic because for the last six quarters, eight quarters, these companies have been applauded for those CapEx numbers. The market has said, we want to see that investment. And watching how management teams navigate this idea that maybe there is a cheaper option out there, maybe there is a route that is

not so capital intensive, and trying to strike the balance of what they've been rewarded for and what they might be rewarded for going forward is going to be fascinating. It's going to be very fascinating. I would expect that you're going to have reporting over the next several quarters from each of these companies on efficiency scale in their models, and more than that, building out interfaces into their models to try and grow ecosystems around like, okay, look at all of the applications or stuff that's being built that leverages my model. It's a real race right now. It's very early. That makes this a very exciting time. I do feel bad for in portfolios that I run, one of them has NVIDIA. It's going to take a giant beating today. That stinks. Short-term losses, I don't take that lightly, especially if you are retired and you are on a fixed income and you have a big position in NVIDIA,

I feel for you a little bit, I think, but I wouldn't panic about it either. I think this is one of those, Dylan, where we're going to figure out a lot over the next year and companies that are well-positioned today are likely to continue to be well-positioned, but how they allocate resources is probably going to be different. Let's keep that lookout for the rest of the year. I think in a recent conversation for our full members over on the premium side, you talked with Francis Schwepp, who's a venture capitalist, early stage investor, and you said 2024 was the year of the GPU. And I know it's early, and I know we are trying to process some very disruptive information, so basically in real time here. But what do you

feel like 2025 is the year of for ai it's a year of ai efficiency it is the year of ai efficiency deep seek made sure of that i said this on this week in tech and and tim and i have been talking about this a lot for several months now where we both thought that the most likely next big breakthrough would be on the software side where you would where you wouldn't want to The thing that makes AI models just such resource hogs and such energy hogs is that it can be like every word is a token. If you're going to make a prediction and you are constantly using tokens just for basic words and you never sharpen the results, then it's always going to be a hog. I go back to – let's go all the way back to Bletchley Park and Alan Turing and the creation of the first computer and the breaking of the Enigma code. What broke the Enigma code wasn't that the computer could crunch all of this unbelievable information.

What broke the Enigma code, Dylan, and kind of what's happening with DeepSeek right now is like, what if we know some words that are always used in every single code so we don't have to process those words? And then we can make the processing more efficient. That's what won World War II, efficiency. I'm sorry, I sound like Morgan Housel right now. I'm talking about World War II. But it's important. We say it time and time again that history has a lot of very good lessons for us here in the present. Tim, I am grateful that you are here to help teach me and our listeners. Thanks for joining me today. Thanks, Dylan.

Listeners, Tim mentioned This Week in Tech. That's his weekly show with Tim White over on our member live stream. You can get access to that and get two stock picks a month by joining stockadvisor, fool.com slash signup is where you can do that. All right, coming up on the show, last week, Fool analyst Seth Jason joined Mary Long to talk through some of the struggles that the rooftop solar industry is facing. Today, we're taking a look at a company that operates more on the utility side of the renewables market, where the outlook is a bit rosier. This is senior Fool analyst David Meyer, interviewing Fluence Energy's president of the Americas, John Zeransky. For those who don't know Fluence Energy,

can you please give us a quick summary of what your business does? Sure. You know, the easiest way to say it is we build big batteries and we build very, very large batteries that are connected to the electric power grid to allow it to work more efficiently, less costly, and produce less emissions, right? So, and when I mean big batteries, think of something on the order of a Home Depot-sized lot of battery systems that are all connected, and effectively a battery that's operating like a power plant. Instead of building a new gas-fired plant, a new coal-fired plant, a new something else, we're building this very large battery. It stores electricity when we are producing a little too much, and then allows us

to use that electricity when we need some more for the grid. I think this is a good jumping-off point to really set the stage for what's happening in the energy market. And right now, if you looked at a headline, probably nine times out of 10, you'd see the word nuclear, because that's been getting a lot of the headlines over the past, let's call it six months. However, you talk about renewables. Renewables is still, if I recall correctly, one of the fastest growing segments of the energy market. So let's talk a little bit about that. How is the renewables market, as well as other pieces of it. How is that driving your business today? Well, I mean, there's a couple of really good pieces of news if you're in this business right

now. The big news really that's occurred over the last year is we're seeing real load growth in the sector. So over the last 20, 30 plus years, we've really only seen load growth as a kind of reflection of population growth. So we've seen general load growth in the neighborhood of 2% per year is sort of what people plan on. Now we're starting to see many of the different utility service areas and transmission service areas starting to predict load growth at 6%, 7%, 8% per year. A lot of this is being driven by increasing use by server farms and AI coming in where we're seeing much more constant power use as these large power consumers are built out. And those large power consumers are looking for places that have low-cost, stable electricity,

a place that they can get access to land relatively easy, and they can permit and build new facilities fast because the growth and the time speed that those systems work on, these data center systems work on, is very, very short. In some cases, it's a little bit of a mismatch to what the electricity system works on because the electricity system is used to being relatively long dated, planned, somewhat slow moving. And so from the beginning of a new project, whether it's a transmission project, a generation project or whatever, you might be looking at five to seven to eight plus years from the beginning of that to when something comes online. And in some cases longer. If you look at large scale transmission projects, some of the newest transmission projects that we're building

across the country, a lot of those projects began their work more than 10 years ago, where they started to get permission to connect. They started to lay out the route and then go into construction and do everything there. So the electricity market has generally been a somewhat slow-moving market in terms of large-scale infrastructure. And we now have this very fast-moving demand cycle from these urgent, fast-moving demanders of electricity. And so finding the ways for those things to fit together is creating a lot of excitement in the electricity sector. So there's a push to demand. That's a big factor. And then the other big transition over the last few years has been that renewables in most places have become the least cost source of electricity.

Which is super interesting, by the way. When we used to talk about solar and wind or hydro or other things, we were mainly talking about it from the standpoint of environmental benefits. And typically, you had some sort of an incentive-based program that created the initiative to create renewables to exist, right? So you might have a direct incentive. You might have some other form of an incentive. Today, renewables have really gotten to the point where they're the least cost source of generation, both wind and solar. Less expensive than putting in a new coal plant, a new gas plant.

You mentioned nuclear. nuclear is very expensive to build, and it's a long cycle. Once you have it in place, the electricity is relatively cheap. And from a carbon standpoint, it doesn't produce any carbon emissions. You have to deal with the waste. But that's why nuclear has come up so much recently, as people are saying, hey, we want large, stable generators at low cost that don't produce carbon. Nuclear seems to fit that. The challenge is the people that want that most also want it tomorrow. You know, and they don't want it 10 years from now. They want it tomorrow.

And so the challenge with the nuclear industry is to get something that you can build more rapidly. And I think what we find mostly being built today is renewables. So a lot of it's solar, some of it's wind in places where we have access to good hydro assets, run of the river or different hydro assets. People are building that. We still see a little bit of geothermal, but it's mostly solar that's driving it. And so it's been a very interesting market. High demand from the base, need for electricity, and then this massive transition to new forms of electric generation and renewables. So, let me put this in some perspective for our investor listeners, and then you tell me, make sure I get this correct.

So, you're seeing a bump up in demand for electricity. On a dollars-per-kilowatt basis, the prices of solar has come down dramatically, which makes it very economic relative to other forms of power, namely, we'll call them fossil fuel and nuclear. Yeah. And I think if you when you combine those two things together, as well as, again, the the intermittency of the renewable generation, that's what's driving you to say, hey, in 2025, we're expecting 40 plus percent revenue growth. Have I got that right? Yeah, that's right. So what energy storage is doing in that whole mix is it's kind of playing this role of a utility infielder. Right. It can run this way. It can run that way. So in the case where we have a lot of conventional generation, fossil fuel generation, we can put storage in and we can improve the operation of that fossil generation. Many of the early projects we did, I've been working in energy storage since 2007.

Back then, it was kind of almost like a dream myth, if you will, that we could put large batteries on the grid and do something with them. But we did. We combined large battery systems with conventional generation. The first ones we did actually were batteries combined with coal plants, and we made the coal plant run more efficiently. And so instead of having the coal plant have to cycle up sometimes when there was more demand and cycle down when there was less, and in some cases have to move very, very quickly to manage intermittencies and outages on the grid, we let the battery do that. And so a bit like hybridizing your car. If you think about what a Prius does, for example, you added an electric motor and a battery, and you also still had a gasoline-powered motor.

By having the battery kind of take the intense pulses, you allow the internal combustion engine to run at a much more consistent cycle. And because of that, you get better fuel economy, you have fewer emissions, you have lower total cost of that system. Effectively, adding a battery to a gas plant or a coal plant is kind of like hybridizing a car that way. So we added it, we lowered the cost, we improved the efficiency, and we improved the reliability. So that was job one, let's say. Then in this cycle, like, you know, you start moving into the 20-teens and the later 20-teens especially, you saw renewables really have a massive uptick. And as we put in lots and lots of solar, lots and lots of wind in places, what we're finding is it's great during the peak part of the day, solar in particular, which is when most of the demand is, right? that people are awake, they're doing things, they're consuming. But as you get towards the

end of the day, for example, people are still pretty active. You're still in your job. Factories are still running. People are cooking dinner. They're watching TV. They haven't really started to completely ramp down yet, but the sun has. And so we end up with this period where we still have relatively high consumption of electricity, but the solar is starting to slack off. And so you don't have production and you have windows of one to two to three to four hours maybe where we still need that peak load of electricity to supply for demand but we don't have it coming from the sun so our choice is turn something else on turn a gas plant on run it for a few hours and then turn it off which is not really the most efficient way to use a plant like that and you you end up having

to have these backup plants right which is costing somebody something but we can put a battery to do that. And the battery can be absorbing some of that excess power at peak. It can use it to fill in that gap in the evening and that gap in the morning. It can absorb stuff overnight if there, for some reason, you need to keep some unit on for minimum load or reliability requirements. That battery can absorb that energy and use it for the morning when things start to ramp up. So the battery system just adds efficiency, if you will, to the whole system. And by adding efficiency, you're essentially reducing costs and you're reducing emissions and you're making the whole system work better. So that's what we're doing. As you get increased demand and as you

have this transition to renewables, there's more and more of those jobs for batteries to do all the time. And so we're seeing very rapid growth. We've been seeing generally 30 to 40% year over year growth in storage demand the last few years. And the third thing, maybe just to mention one more as I throw one thing in here is as we've scaled up the battery industry, we've seen price declines in the cost of battery systems. And so similar to what we saw in solar, as you increase the scale by a lot, you find that you can incrementally take cost out. And this is getting cheaper and cheaper to install batteries alongside other elements in the system. And so that's also fueling demand. So, you know, we're very excited about it. We've been at the forefront of this.

We started this a long time ago and have been one of the real leaders in driving storage into the grid system and see a lot more to come. 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 don't buy or sell anything based solely on what you hear. All personal finance content follows Motley Fool editorial standards and is not approved by advertisers. Motley Fool only picks products it'd personally recommend to friends like you.

For Tim Byers and David Meyer, I'm Dylan Lewis. Thanks for listening. We'll be back tomorrow.