Motley Fool Hidden Gems Investing - What Drives Nvidia’s Growth?

Episode Date: February 27, 2025

The tech giant is a data center business. (00:21) Asit Sharma and Ricky Mulvey discuss: - Why Wall Street is shrugging off Nvidia’s 78% yearly revenue growth. - CEO Jensen Huang’s vision for AI in the coming years. - Short sellers targeting digital ad seller, AppLovin. Companies discussed: NVDA, APP, META Host: Ricky Mulvey Guest: Asit Sharma Producer: Mary Long Engineer: Dan Boyd Learn more about your ad choices. Visit megaphone.fm/adchoices

You've got to try breakfast at A&W. You've got to try breakfast at A&W. And what better way than with the delicious Pret Organic Coffee? Starting with just one dollar, all day, every day, now until December 31st. You've got to try breakfast at A&W. At participating A&W locations in Ontario. NVIDIA reported, the market shrugged. You're listening to Motley Fool Money. I'm Ricky Mulvey, joined today by Asit Sharma. Asit, thanks for being here, man.

I appreciate you inviting me, Ricky. Glad to be here. Well, I wanted to get you on NVIDIA Day because I know you think a lot about artificial intelligence, and this is the market leader. This is the leader in the space. And on the surface, if we don't look at the stock reaction, it seems like NVIDIA shot the lights out. Year-over-year sales growth of 78%, almost 80%. And most of that is coming from data center revenue. Make no mistake, we could and probably will talk autonomous driving, gaming, robotics, but right now, NVIDIA is a data center business. So let's talk about that number. and talk about the growth in the data centers. Where's that coming from, Asit?

So, it's coming from two places, Ricky. First, as we all know, it's coming from the big hyperscalers who are buying NVIDIA GPUs, especially their Blackwell GPU complexes, hand over fist. So, think big companies like Microsoft, which has the Azure platform, Amazon.com, which has AWS. Those companies that are in the business of serving up AI to us are buying this compute. Now, the other half of that group is enterprise businesses, companies that may be in the Fortune 1000 or the Fortune 500. They are slowly but surely digging deeper into their own capabilities to serve AI to their customers. And they're not only renting space on these big clouds, they are doing this internally. So, they're buying GPUs for their own purposes. That's becoming

a little bit bigger business over time than it was at the outset of this generative AI explosion a couple of years ago. There are a few forces that seem to be affecting NVIDIA right now. One is that these large language models that NVIDIA chips power are being asked to do much more than they were just a couple of years ago. Simultaneously, the cost of inferences in doing that is declining. So here are the two forces. One, this is from the call, we've driven a 200x reduction in inference costs in just the last two years. And also pointed out that the amount of tokens generated for an inference compute is already 100 times more than the one shot example. So to break that down, a one shot is if you ask an LLM, what is the capital of Japan? An inference compute

would be, here are NVIDIA's earnings, and I want you to summarize it and debate the bull and bear cases. You're asking for a larger logic chain. So, those are the two forces. The cost is going down, but these machines are being asked to do more. Are these forces in opposition, and what do they mean for NVIDIA? So, they're not really in opposition, if you think of this from NVIDIA's perspective. When the call came to supply this type of answer for consumers. And NVIDIA went from being sort of a test and research development phase company alongside those companies that were building the models, to just jumping into this wide world where suddenly everyone can pay for inference. NVIDIA was talking a lot about scaling laws. The fact

that after you train the model, it required a lot of compute to keep serving up inference. So, the scaling law of inference was much talked about at the time. Now, as we've emerged into bigger and badder models, Ricky, the training actually doesn't stop once a model is released. So, now we're into phases called post-training, where a model keeps learning after it's released to the public, and that really triggers this second law of scaling. And Jensen Huang talked about three laws of scaling last night on the call. That second is the post-training scaling law. That means you have to have a lot more compute

as you post-train a model. So, NVIDIA is starting to win on volume, as you can see. The models emerge, they get better, and more compute is required. Now, there's this third law, which you sort of alluded to, which is when you ask a model to reason, to think in steps, that requires a lot more compute than just answering one question and waiting for that simple answer to a simple query. This is the third law that Jensen was referring to in the call, the inference time scaling law. You ask a model to think, to reason, to take steps, take its own time. Have you tried the latest ChatGPT models that really think? Sometimes that model takes five minutes to return an answer.

As these laws keep evolving, really what's happening is that NVIDIA is winning on volume, as I sort of hinted at before. So, the cost of compute can go down. And to complete this virtuous cycle, what NVIDIA is doing is architecting for more compute that will handle more and more of these scaling laws. And it has to drive down the cost for its customers to want to keep playing. And the customers want to keep playing because they're showing cost savings with each successive, more complex round of GPUs that it throws at the market. What's happening with NVIDIA's networking revenue? We talked about how this is a data center business. So for the newer listeners, how is the networking business different from

the data center business? Because we're going to talk about it. This is one rare area where revenue is actually declining for NVIDIA. Sure. So networking is such a fun thing to think about because none of us really understand it unless we happen to be in this industry. The way I think about it, Ricky, is slinging data throughout a physical space in a way that's efficient, given whatever that end demand is. In this case, AI. NVIDIA bought a company called Mellanox a few years ago, which had a competing standard to the Ethernet standard, which we probably all remember from years ago. That has proven to be pretty good for moving data through AI networks a little bit faster, in some cases, than the Ethernet standard.

Now, NVIDIA has kept innovating on this technology, and it's trying to compete with companies like Cisco, with Juniper Networks, with Arista Networks in some ways. But more overt than that, it's trying to make its own data centers, the ones that it builds in prototype, it actually builds prototype factories to make them more efficient so it can sell more of its GPUs. They've gotten pretty good at this. And what happened this quarter is that a standard that NVIDIA had put into place now is merging over, integrating to a new standard. So, the old standard, which is still quite robust, is shifting to something called NVLink 72. And they are combining that with a technology that they call Spectrum X. With NVIDIA, there's always so many

new products, product names. The gist of this is that they have a transition quarter as they make their networking more capable for the next generations of Blackwell GPUs. They're going to see a slight drop off in this networking revenue, but the company expects that it's going to pick up in the very near future. And anytime you listen to an NVIDIA earnings call, you always get bold visions of the future from CEO Jensen Wong. And I'm hoping you can translate this vision for our listeners. Quote, the next wave is coming. Agentech AI for enterprise, physical AI for robotics, and sovereign AI as different regions build out their AI for their own ecosystems. And so, each one of these are barely off the ground and we can see

them. End quote. What is that vision, Asit? That vision is a vision in which the three most important customers to NVIDIA, those enterprise businesses that I mentioned, then companies that are going to merge in the future from doing stuff that's online and in cloud data centers, they're going to move that into the physical world. So, think the manufacturing community, the automobile autonomous driving community. And third, the only entity left on the planet that has enough pockets to keep NVIDIA growing, if they exhaust the spending of these big hyperscalers and companies and manufacturing companies in the future, that group is the government. Let's quickly break down what this means. Agentic AI for enterprise is the ability

for big companies to spin up their own AI agents throughout their companies and make your work and my work theoretically more easy so that you and I can be more productive and potentially keep our jobs. This is the future that everyone is questioning. Second, physical AI for robots, is it's sort of interesting. From its founding, NVIDIA has been fascinated with the physics of how things work, from the physics of visualization, so how they became a leader in the gaming space with their virtual reality machine learning, and also the way they present graphics on their graphics cards. That's always been something they've wanted to explore, and they've moved that into the physical world. They take something that limits a large language model, the text

modality, and they've expanded that into a lot of video and other modalities. What this means is that they're gathering enormous amounts of data so that the robots can understand physically how things work. They're training robots for the real world. Instead of a robot going out and moving things with its hands. Before that ever happens, they have thousands and millions and billions of simulations based on video data or prior other data, some of it text, so that the robot already knows how to do it pretty well. And they have a platform called Cosmos, which is promoting this, which has millions and millions, actually, I think it's in the trillions of tokens of training already under its belt. And then finally, that sovereign AI piece. Foreign governments

want to catch up in a world where AI can be a level playing field or make things a level playing field. So, NVIDIA's positioning itself is sort of the first end-to-end customer. And they're pitching that to really European countries, countries from the Middle East, all over the globe, to say, we can come in and give you the technology that's needed that can be in-house, so you don't have to go outside of your own country and put stuff on other people's servers. And we can make you as competitive as, say, China or the United States, because you'll be using our latest technology. And it's a big market for them. It's a market that, if it materializes, could be the transition market outside of these few big names that everyone

knows, Alphabet, Microsoft, Amazon, who are the main props right now of the NVIDIA story. All right. I'm going to ask you maybe an unfair question. This is a stock that I have, unfortunately been watching from afar for a few years now. And in my brain, I was like, there's going to be a dip. Right now, NVIDIA is at about 28 times forward earnings. It's flirted with 50 times forward earnings just a few months ago. For a stock that leapfrogs the market cap of McDonald's on certain days, which it has done before, 28 times forward earnings seems awfully mature. Yes, it's a multi-trillion dollar business, but is this a dip worth buying, Asit? It is a tough question. I wake up some mornings, on the mornings I'm thinking about NVIDIA

and wonder if all that growth isn't in the rear view mirror. It seems awfully hard for a company this size to grow at a rate that could be in the teens or the low 20s that would justify a person buying today who feels like maybe it could be an even cheaper company. I bought it 28 times earnings today, but as the growth sputters out, maybe it trades for 15 times earnings in the future. The other argument or way to look at it is that this company has exhibited an uncanny knack for understanding what the future looks like. Proof of this case is this latest Blackwell chip. If you've ever driven by a piece of land and seen the sign build to suit, meaning thereby that the owner of the land will build what you want for you based on your specs, say

a warehouse or a restaurant, NVIDIA is the planet's best build to suit manufacturing company because it works so closely with all the key players, the people that are building the large language models, the people that are building data centers, the hyperscalers, the end customers for its GPUs. It has such a bead on what the future looks like. Forget its own great technology. So, it already has a keen understanding of where things could go. That it's almost unerringly correct at developing in advance the technology that has a lot of demand attached to it. So, even if this phase slows, Ricky, at least based on its track record, I wouldn't be surprised if a dormant company that people don't get excited about anymore

called NVIDIA sometime in 2032 surprises the market again and takes off. I want to move on to the next story. And this is about Applovin, which kind of quietly was the most successful tech company in the stock market of 2024, Asset, not Palantir. It was Applovin. But right now, Applovin is facing the shorts. So for those unfamiliar, Applovin sells ads for mobile games. If you're a fool, you can think of this as the trade desk, but for mobile gaming. And before we get to the allegations, it's been a tremendous rise for the company. Why have investors previously, at least, been so bullish on Applovin? Applovin operates in a pretty difficult market. So this is digital programmatic advertising. If

you're familiar with the trade desks, it's a little bit like that, an ad platform that's served up automatically. But it's confined mostly to the mobile gaming market, which is treacherous. It's a market that it's just difficult to make good money in. There's a company called Unity Software, which is extremely capable. It's having a relatively good year, but that's a case study of the stumbles and problems with trying to compete in a market where ad impressions are are hard to come by. It's hard to actually have unit economic value here. Applovin seemed to do this quite easily with a management change and some new technology. They call it Axon 2.0. It's a model that is continually enhanced to optimize their monetization strategy.

The stock has really taken off with the success of this platform. That's the quick story behind why Applovin has garnered a lot of attention from the business community, the analyst community, and also just retail investors who want in on a company that can prove it can really juice the earnings and revenue growth in such a tough market. There's quite a few allegations in the short reports that one of them came from Fuzzy Panda. A few of them include, basically, one is essentially copying Meta's homework on their users to track their customers, and they think Meta would be very upset about that. Another allegation is that a lot of app-loving ads use nefarious tricks to drive downloads.

That could include placing a little X above a game where you think you're closing out of the ad, but really, you end up opening the app store to download the game. And then additionally, there's some fairly serious allegations on tracking children within these mobile gaming segments. When you were working your way through the short report, Asit, was there anything that just wowed and shocked and awed you as these short sellers would like to do? I like to read short reports of companies that are recommended in services that I work on here at The Motley Fool or that I own personally for the reason that sometimes a short report can have

a kernel of truth that you need to follow and understand more. I've read my share of reports, Ricky, where I just went through everything and felt like nothing here has any potential to even stick. And it just sounds over the top. So I guess what caught my attention was what you mentioned, the allegations of harm towards younger consumers. And that seemed like something that is a typical hook for short sellers when they have a report. So we have to just explain here. short sellers provide a service to the investment community in that they can point out what they think or allege is misunderstood by investors. But they also have a financial interest in investors getting scared and selling out of a company. That

interest is often brought to your attention by the hook. I think that's what leapt out at me as a little maybe too much, because I think it's probably going to be easily refuted by the company. And in fact, there was a blog post from the CEO of Applovin responding to this. And I think he pretty squarely deflected any kind of harm that could come to younger users. But I know we have a lot more to talk about, so let's keep moving. Yeah, this is definitely a question of who do you believe? The short sellers would say that Applovin ads, where if you're playing a mobile game, you see another ad for a mobile game, that there's direct downloading that you are tricked into as a user.

The CEO of Applovin, Adam Ferroghi, says every download results from an explicit user choice. These are statements in direct opposition to each other. And it's up to the, I guess, the investors to decide who they believe, Asit. The blog post itself is pretty interesting, Ricky, because it's short. And this is one of the statements, you know, you've isolated that makes me think I want to study this a little bit more. I haven't made any kind of firm decision on whether these allegations are spot on or just way off. But when an executive puts it in this way, what he's doing is putting the onus of any kind of technical engineering up for subjective interpretation. So in other words, you explicitly, user wanted to play this

game. So you clicked here. Now, what you've also brought up here is if that triggers some installs that the customer doesn't understand or know about, they're disclaiming responsibility, but it doesn't mean they couldn't be culpable in that respect. And we should point out here that Applovin, while they clearly explained how their value is created, they say it's not created by just mere impressions or clicks. Their revenue is based on the value that they drive for those using their service. Well, actually, it's partly install-based. So, the more installs they show, the more revenue they generate. So, that question doesn't exactly address this. The other thing that the blog post doesn't head-on address is this allegation that you've

brought up. Not that you've brought up, Ricky. You and I aren't in the business of writing these short reports, but you've relayed from the reports that maybe this company is sort of reverse engineering important data for Meta. So what's happening here allegedly is that Applovin has sort of a view into Meta's advertising. And from that, it's getting access to important first-party data. So Meta ads will include first-party data that maybe the customer doesn't want anyone else to see a meta platform customer and be served up ads from. So the allegations are that they're sort of looking at all this stream of information and then engineering advertising that makes their ads more successful, which is not kosher for the major platforms from meta

to Apple to Alphabet. And anytime I read a short report, and in this case, I'm reading a short report and i'm reading management's response asset my eyebrows raise in both cases because you have you have two players with tremendous benefit to tell a certain narrative and in the case of app love and as you've mentioned i've been served mobile ads for games that feel a little fishy so i understand where they're coming from on that on the other side for these short sellers they're basically saying that once meta finds this out they're gonna shut down app loving and we know this because of a whistleblower who found this with 13 standard deviations. It couldn't possibly be coincidence. And oh, by the way, we can't really publish that because

they sold that information to a hedge fund. There's some weirdness going on. And it makes me wonder, couldn't Meta just shut down Applovin? And why is a short research firm figuring this out before Meta. So my wife, who has a degree in information science and is really great at reasoning and is often calling me out at the dinner table for stuff I don't understand, she would say, that's the kind of question you should ask. If you're reading a short report, this is the crux of it. Why wouldn't Meta figure this out on its own? Why couldn't they? Why haven't they? Why wouldn't they pull the plug? So I think this is, for investors, just a great question to ponder and to keep us from jumping to a conclusion and getting scared into a response

or perhaps being tricked into a response. We just don't know. But I think we'll find out in the coming quarters if next quarter we see a big drop in revenue and they say, oh, by the way, a certain unnamed platform is sort of clamped down on us, we'll understand what happened. So, I would encourage the investors listening, it's okay to just look at this. Take the information in. It doesn't mean you have to take action on it right now. Asit Sharma, Thanks for being here. Appreciate your time and your insight. Thanks a lot, Ricky. As always, people on the program may have interests in the stocks they talk about,

and The Motley Fool may have formal recommendations for or against, so don't buy or sell stocks based solely on what you hear. All personal finance content follows Motley Fool editorial standards and are not approved by advertisers. The Motley Fool only picks products that it would personally recommend to friends like you. I'm Ricky Mulvey. Thanks for listening. We'll be back tomorrow. Thanks for watching!