Latest / Stock Picking with Fexingo: Individual Equities, Research, and Building a Concentrated Portfolio / How NVIDIA Is Becoming an Infrastructure Company
Transcript
- Lucas: So NVIDIA is up about 4 percent over the past five days, even as Microsoft slipped a bit. And I think the market is starting to price in a shift that goes beyond the latest GPU cycle. Luna: A shift how? Because when I look at NVIDIA, I still see a company that sells chips. Yes, they're very expensive chips, but it's still a hardware revenue model. Lucas: Right, and that's been the dominant narrative. But quietly, over the past eighteen months, NVIDIA has been building out something that looks a lot more like an infrastructure company. I'm talking about DGX Cloud, their ai platform as a service offering, and the broader push to make NVIDIA the operating system for AI data centers, not just the silicon supplier. Luna: DGX Cloud — that's the service where companies rent NVIDIA's full stack, hardware plus software, instead of buying the hardware outright, correct? Lucas: Exactly. You don't buy an H100 or a B200 server. You subscribe to a pod of GPUs managed by NVIDIA, running on top of major cloud providers like Oracle and Microsoft Azure. And the key point is: that subscription revenue is recurring. It's not tied to the boom and bust cycle of enterprise data center build-outs. Luna: But isn't that still essentially selling compute time? I mean, Amazon Web Services has been doing that for years with their own Inferentia chips. Lucas: It is similar in form, but the difference is the software layer. NVIDIA has CUDA, which is this incredibly deep ecosystem of libraries and tools that developers have built on for over a decade. If you're a company training a large language model, you're almost certainly using CUDA. Switching to a different chip means rewriting that software stack. That's a massive switching cost. Lucas: Precisely. And that's what makes the infrastructure as a service model so powerful. NVIDIA can charge a premium for the full stack, and once a company is integrated into that ecosystem, they're very unlikely to leave. You see this in their enterprise software revenue, which has been growing at over 50 percent year-over-year for the past three quarters. Luna: Fifty percent year-over-year — that's not just a cyclical bump. But let me play devil's advocate: the hyperscalers — Google, Amazon, Microsoft — they're all building their own AI chips. Wouldn't that eventually erode NVIDIA's pricing power? Lucas: It's a real risk, but I think it's important to look at the time horizon. Building a competitive AI chip isn't just about hardware design. You need the full software stack, the compilers, the libraries, the integration with existing AI frameworks. Google's TPU is probably the closest competitor, but even Google still uses NVIDIA for a lot of its training workloads. And for most enterprises, building a custom chip is not an option. They'll go with the platform that works out of the box. Luna: So you're saying NVIDIA's real competitive advantage is the ecosystem, not just the chip performance. Lucas: Exactly. And that's why I think the market is starting to re-rate NVIDIA. The P/E ratio has actually compressed a bit this year, even as earnings have grown. But if the market starts to see NVIDIA as a recurring-revenue infrastructure play, rather than a cyclical hardware supplier, that multiple could expand significantly. Luna: Let's talk about that headline from today — about AI computing power becoming a tradeable commodity. If AI compute becomes commoditized, doesn't that hurt NVIDIA's pricing? Lucas: It could, but only if the commodity is undifferentiated. The article talks about efforts to create a futures market for AI compute — which is fascinating. But NVIDIA's bet is that they can provide the premium tier, the one with the best performance and the most advanced software tools. Just like in cloud computing, there's AWS for the mass market, but there's also a premium segment where companies pay more for better performance and support. Luna: So NVIDIA wants to be the Apple of AI infrastructure — the premium, integrated experience. Lucas: That's exactly the analogy. And Apple's hardware margins are high, but their services revenue is what drives the multiple. If NVIDIA can get to a point where, say, 30 percent of revenue is recurring software and cloud services, the valuation story changes completely. Luna: It's interesting that you mention the valuation story, because we talk a lot on this show about finding companies with durable competitive advantages. And it sounds like you're arguing NVIDIA is transitioning from a cyclical advantage to a structural one. Lucas: That's the thesis in a nutshell. And I think the evidence is there if you look beyond the headline chip numbers. The DGX Cloud revenue is still small, but it's growing fast. And the enterprise software attach rate is climbing. For a concentrated portfolio, this is the kind of transformation that can drive outsized returns. Luna: You know, speaking of the show and the research we do, we deliberately keep these episodes free of ads. The idea is that you get the analysis without any sponsored interruptions. Lucas: Yeah, that's been the philosophy from day one. And if you find that valuable, and you want to support the ad-free approach, there's a simple way to do it. Just visit buy me a coffee dot com slash fexingo. Luna: No pressure, but it helps us keep the focus on the ideas rather than trying to sell you something. Lucas: Exactly. And back to NVIDIA — the other thing I'd point to is the recent announcement about their partnership with Oracle to offer DGX Cloud directly through Oracle's data centers. That's a big deal because it gives NVIDIA access to Oracle's enterprise customer base, which is a different set of buyers than the cloud-native startups. Luna: So they're going after the Fortune 500 that might not have the expertise to manage their own AI infrastructure. Lucas: Right. And those customers are typically less price-sensitive and more focused on reliability and support. That's a higher-margin, stickier revenue stream. It's also a hedge against the cyclical swings in the data center build-out cycle. When companies pause their hardware purchases, they still need to run their AI workloads. Luna: So if we think about the risk factors, what keeps you up at night with this thesis? Lucas: The biggest risk is execution. Transitioning from a hardware company to a services company is hard. Look at how long it took Microsoft to pivot under Satya Nadella. NVIDIA is essentially trying to do both at the same time — keep the chip innovation engine running while building a new services business. That's a lot of management attention and capital allocation. Luna: And the competitive risk from custom chips, even if it's a longer-term threat. Lucas: Yes. But I'd argue that the custom chip threat is actually more of a risk for the hyperscalers' own suppliers than for NVIDIA. If Amazon builds a better chip for its own data centers, that hurts Intel and AMD more than NVIDIA, because NVIDIA's advantage is broader — it's not just about running inference in one cloud. Luna: That's a good point. So the moat is really about the network effects of the developer ecosystem. Lucas: Exactly. And that's why I think the current moment — with the stock up 4 percent in a week while the broader market is also rallying — might be a good entry point for investors who think long-term. The infrastructure narrative is not fully priced in. Luna: Alright, Lucas. So if someone wants to dig deeper, what's the one number they should watch besides the obvious GPU shipments? Lucas: I'd watch the 'Software and Services' line in NVIDIA's quarterly earnings. It's broken out in their segment reporting under 'Compute & Networking'. If that number grows faster than total revenue for the next few quarters, the infrastructure thesis is playing out. Right now, it's roughly 15 percent of that segment. If it gets to 25 percent, the market will start to re-rate. Luna: Fifteen to 25 percent — that's a meaningful shift. I'll be watching that too. Lucas: It's a subtle change in the business model, but for a stock that's already a mega cap, those incremental shifts can drive the next leg of the story.