Forces Fueling Nvidia's Expansion
Rewritten Transcript:
In this lively chat, Motley Fool analyst Asit Sharma and host Ricky Mulvey dive into:
- Why Wall Street's response to Nvidia's 78% annual revenue growth seems dismissive.
- CEO Jensen Huang's vision for AI in the coming years.
- How short-sellers are targeting digital ad seller AppLovin.
Keen to explore all The Motley Fool's free podcasts? Head to our podcast center. Got investment basics on your mind? Check out our beginner's guide to investing in stocks. Keen to invest? Look into this top 10 list of stocks to buy.
Here's a sneak peek into their conversation, which was recorded on Feb. 27, 2025.
Ricky Mulvey: NVIDIA's earnings didn't impress the market. You're tuned into Motley Fool Money. Hey Asit, I'm Ricky Mulvey, and today, I've got Asit Sharma with me. Thanks for joining, Asit!
Asit Sharma: I appreciate the invite, Ricky. Nice to be here!
Ricky Mulvey: I wanted to talk NVIDIA Day because you're a whiz when it comes to artificial intelligence, and this is the market leader. Let's not forget the mind-blowing 78% yearly revenue growth, most of which comes from data center revenue. Let's nerd out about that, Asit!
Asit Sharma: It's coming from two sources, Ricky. First, we have the big hyperscalers who are snapping up NVIDIA GPUs, especially the Blackwell GPU complexes. Think giants like Microsoft with Azure and amazon.com with AWS. But we also have enterprise businesses like Fortune 1,000 or 500 companies that are digging deeper into AI for their customers. Instead of just renting space on those big Clouds, they're trying it internally, and that's growing.
Ricky Mulvey: There are a couple of factors at play when it comes to NVIDIA. One is that these massive language models require more compute to run every day compared to a few years ago. Simultaneously, the cost of inferences has been plummeting. Here's a quick breakdown: a 200X reduction in inference costs in the last two years, and the amount of tokens generated for an inference compute is already 100 times more than a one-shot example. To clarify, a one-shot example is asking a model to answer a simple question, while an inference compute involves more complex reasoning like summarizing an earnings report and debating the bull and bear cases. Asit, what's going on here? Are these forces working against each other, and what does it mean for NVIDIA?
Asit Sharma: They're not really in opposition if we consider NVIDIA's perspective. When it comes to answering consumer queries, NVIDIA moved from being a research and development phase company to catering to a wider audience that can pay for inference. NVIDIA CEO, Jensen Huang, discussed three laws of scaling during the call:
- The first law refers to the scaling of training models, which requires a lot of compute.
- The second is the post-training scaling law, meaning that once a model is released, it requires more compute for post-training.
- The third law is the inference time scaling law, focusing on how asking a model to think or reason requires a lot more compute than answering a simple query.
These laws are interconnected, as NVIDIA strives to cater to growing consumer demand by offering more powerful GPUs and driving down costs for customers. This is part of a virtuous cycle that benefits both NVIDIA and its customers.
Ricky Mulvey: While the networking revenue is declining for NVIDIA, let's get into the nitty-gritty of what differentiates the networking revenue from the data center business. NVIDIA's network business relies on efficiently slinging data through physical space, and it's a derivative of a competing standard to the Ethernet standard called Envy Link 72. NVIDIA also uses a technology called Spectrum X, but they're going through a transition quarter as they merge their networking technology with Envy Link 72. Asit, what's the deal here?
Asit Sharma: NVIDIA purchased Mellanox a few years ago, and they've been working on innovating on this technology to compete with big names like Cisco, Juniper Networks, and Arista Networks. They're also building their own data centers, and by doing so, they're becoming more efficient at producing and selling their GPUs. This quarter has seen a dip in networking revenue due to the transition, but NVIDIA expects it to pick up soon.
Ricky Mulvey: In every NVIDIA earnings call, Jensen Huang offers bold visions of the future. Let's check out the latest: "The next wave is coming: agentic AI for enterprise, physical AI for robotics, and sovereign AI as different regions build their AI ecosystems." What does this mean, Asit?
Asit Sharma: This vision revolves around three key customer groups: enterprise businesses, the merger of companies that operate online and in Cloud data centers, and sovereign governments. Here's a brief overview of each:
- Agentic AI for enterprise: This refers to AI agents that can automate tasks within a company, making your work and mine potentially more efficient and preserving jobs.
- Physical AI for robotics: AI is moving from text-based models to video and other modalities, allowing NVIDIA to gather substantial data for robots to understand the physical world more effectively. This enables robots to learn and operate in the real world without as much hands-on training.
- Sovereign AI: Foreign governments want to catch up to the AI competition, particularly between China, the United States, and Europe. NVIDIA is positioning itself as a one-stop solution for governments, offering the latest AI technology and keeping data within their borders.
Ricky Mulvey: I must admit, I've been on the sidelines with NVIDIA for a while now. Right now, the stock trades at 28 times forward earnings, which seems mature for this high-growth stock. Is this a dip worth purchasing, Asit?
Asit Sharma: It's a tough call. I find myself questioning whether all that growth is behind us. It's challenging to imagine a company of this size maintaining a growth rate that justifies a 28 times forward earnings multiple. On the other hand, NVIDIA has a remarkable track record for understanding the future and developing cutting-edge technology in advance of demand. Even if this growth phase slows, there might be dormant periods where investors lose interest, but in 2032, NVIDIA may spring back to life as the next industry-disrupting innovation.
Ricky Mulvey: Let's switch gears to AppLovin, which quietly dominated the tech stock market in 2024. But now, AppLovin faces scrutiny from short-sellers. In a nutshell, AppLovin sells ads for mobile games, with a focus on The Trade Desk's programmatic advertising market. Need some background on why AppLovin has been a top pick for investors?
Asit Sharma: AppLovin operates in a tricky market—digital programmatic advertising for mobile gaming. The competition is intense, especially with companies like Unity Software. However, AppLovin has had success, thanks to management changes and their enhanced Axon 2.0 model, which continually optimizes their monetization strategy.
Ricky Mulvey: The short-seller reports are full of juicy allegations against AppLovin. One is essentially copying Meta's homework on their users to track customers, which Meta wouldn't be happy about. Another allegation is AppLovin ads tricking users into downloading mobile games without their knowledge. Lastly, there are serious accusations about tracking children within mobile gaming segments. Asit, did any of these allegations shock you while reading the short reports?
Asit Sharma: I often read short reports for companies mentioned in services I work on or that I personally own. Some reports contain a kernel of truth that requires further investigation, but others come across as over-the-top. The allegations about harming younger consumers seemed particularly sensational, and I think it will be easily refuted by the company. However, I'm not making any firm judgments until I can examine all the evidence.
Ricky Mulvey: Short-sellers claim that a good proportion of AppLovin ads use sneaky tricks to drive downloads, such as displaying an X above a game that looks like it's closing the ad, but actually opens the app store to download the game. The CEO of AppLovin claims every download results from an explicit user choice. When two statements appear to be in direct opposition, it's up to the investor to decide whom they believe, Asit!
Asit Sharma: I think it's essential for investors to look at both sides of the argument and make an informed decision. In AppLovin's case, they've clarified that their revenue isn't just based on impressions or clicks, but on the value they drive for their customers. However, the allegations about reverse-engineering important data from Meta is cause for concern. Without more information, it's difficult to determine the validity of these claims, and it's important for both sides to provide evidence to support their positions.
Ricky Mulvey: Each time I read a short report, I'm reminded that both parties have a vested interest in swaying our opinions. In the case of AppLovin, I've been served mobile ads that feel fishy. I get why they're saying that. On the other hand, short-sellers argue that, once Meta discovers what AppLovin is doing, they'll shut down AppLovin. But why would Meta find out first? Aren't they equipped to catch malicious activities like this on their own?
Asit Sharma: My wife, who has a background in information science, would ask this exact question. Why didn't Meta figure it out first if they have the resources and capabilities to monitor their platform? It's an important question for investors to ponder, as it challenges us to be cautious and not jump to conclusions based solely on the information presented in short reports. We'll see how this situation unfolds in the coming quarters.
Ricky Mulvey: Wise advice, my friend. As always, the people featured on this program may hold stakes in the stocks discussed, and it's essential to do your own research before making investment decisions. I'm Ricky Mulvey. Until next time!
Asit Sharma: Thanks, Ricky!
- Given the growth of data center revenue for NVIDIA, enterprise businesses are increasingly investing in AI solutions for their customer-facing tasks, aiming to boost efficiency and preserve jobs with the use of AI agents.
- Asit Sharma expressed his uncertainty about whether NVIDIA's growth rate justifies its 28 times forward earnings multiple, implying that the stock's maturity might call into question its further growth.
- Short-sellers have been pointing fingers at digital ad seller AppLovin, accusing them of copying Meta's user tracking data and tricking users into downloading mobile games without their knowledge, though Asit believes these extreme allegations can likely be easily refuted by the company.
- The conversation around NVIDIA's revenue revealed that the company is undergoing a transition quarter as it merges its networking technology with Envy Link 72, causing a dip in networking revenue but expecting an uptick soon.
