Latest / Stock Picking with Fexingo: Individual Equities, Research, and Building a Concentrated Portfolio / How NVIDIA Is Becoming Its Own Best Customer
Transcript
- Lucas: NVIDIA is up almost eight percent this week, trading north of two-ten as we record on Monday July thirteenth. That's a big move even by its standards, and it's not because of a new chip announcement or an earnings beat. Luna: So what's driving it? I saw the headlines about Meta being up even more, but NVIDIA's move was surprisingly broad-based. Lucas: Exactly right. And the driver is something investors are slowly waking up to: NVIDIA is becoming one of its own biggest customers. I'm not talking about the traditional hyperscaler demand from Microsoft, Amazon, or Alphabet. I'm talking about NVIDIA buying its own GPUs to build out AI infrastructure that it either uses internally or leases to others. Luna: So they're basically competing with their own customers? Or is this more about capturing more of the value chain? Lucas: It's the latter, and it's a fascinating capital allocation story. In their most recent fiscal year, NVIDIA spent roughly twelve billion dollars on capital expenditure — that's up from about three billion two years ago. A big chunk of that is going to data center build-outs that house their own H100 and B200 chips. They're not just selling shovels in the gold rush; they're also mining. Luna: So the capex goes up, which normally depresses free cash flow, but in this case it's actually creating a new revenue stream? Lucas: That's exactly the angle. Think of it this way: NVIDIA spends money to build a data center, fills it with its own GPUs, and then either uses that compute for internal R&D — training their next-generation models, optimizing chip designs — or sells the compute time via NVIDIA's DGX Cloud platform. So the capex converts into recurring revenue over time, and the margins on that cloud service are actually higher than their hardware margins. Luna: Right, because they control the full stack. But doesn't that create tension with their hyperscaler customers who also want to sell cloud compute? Lucas: It's a delicate dance. But remember, the total addressable market for AI compute is growing so fast that there's room for everyone. And NVIDIA's argument is that by having its own cloud, it can showcase best practices, push the ecosystem forward, and ultimately expand the pie. The hyperscalers are still buying more chips than anyone else. In fact, Microsoft alone accounted for about fifteen percent of NVIDIA's data center revenue last quarter. Luna: It reminds me of how Apple started making its own chips and then built services on top. But with NVIDIA, the product itself is the infrastructure. So the capital intensity is much higher. Lucas: Exactly. And that brings us to the stock's valuation. NVIDIA trades at roughly thirty-five times forward earnings, which sounds expensive until you factor in that this internal capex is essentially pre-paying for future high-margin revenue. If you strip out the capex that's really an investment in future cloud revenue, the adjusted price to earnings ratio is probably closer to twenty-eight or twenty-nine times. Luna: So the market is starting to price in that virtuous cycle. That eight percent move this week — was there a specific catalyst or just a gradual recognition? Lucas: I think it's a combination. There was a report from a sell-side firm quantifying the size of NVIDIA's internal data center build-out — they estimated it could contribute three to four billion dollars in annualized cloud revenue by the end of fiscal twenty-twenty-seven. And that number is not in most models. So as analysts revise their numbers upward, the stock responds. Plus, the broader market is rotating back into megacap tech after a rough June. Luna: And that rotation is real. The S&P 500 is up half a percent over the past five days, but the Dow is down nearly a percent. The NASDAQ is up six-tenths of a percent. So money is flowing back into growth. Lucas: Right. And NVIDIA is the most liquid way to play AI infrastructure. But the reason I find this particular angle interesting is that it changes how we think about the company's moat. It's not just about having the best chip. It's about having a self-reinforcing ecosystem where the more you invest in your own infrastructure, the better your chips become, and the more your customers rely on you. Luna: So for a stock picker, how do you model this? Do you build a separate division for the cloud business, or just treat it as a margin uplift over time? Lucas: I think the cleanest way is to break out data center revenue into two buckets: chip sales and cloud services. Chip sales grow with unit volumes and average selling prices. Cloud services grow with utilization and capacity. And the cloud business has much higher incremental margins — think sixty to seventy percent versus the chip business's fifty to fifty-five percent. So as the cloud mix increases, overall margins expand. That's a powerful compounding effect. Luna: And that's exactly what you want in a concentrated portfolio. One stock that has multiple engines of growth and a self-reinforcing advantage. You're not just betting on a product cycle. Lucas: Speaking of things that compound, if today's episode helped you think about NVIDIA — or any stock — in a new way, and if that insight was worth the price of a coffee to you, there's a link where you can show that. It's buy me a coffee dot com slash fexingo. That's all lowercase, no spaces. And that kind of listener support is honestly what keeps this show ad-free and focused on the numbers. Luna: Yeah, it's a small gesture that makes a big difference for us. And it lets us keep drilling into these specific angles without having to chase clickbait. Lucas: Exactly. So back to NVIDIA: one risk to watch is that if the hyperscalers decide to build their own custom chips, like Google's TPU or Amazon's Trainium, that could erode NVIDIA's pricing power over time. But in the near term, the switching costs are enormous because the entire AI software stack is optimized for CUDA. Luna: And CUDA is the moat inside the moat. It's not just about the hardware. So the internal cloud business is almost like a way for NVIDIA to monetize that software lock-in even further. Lucas: Right. And that's why I think this stock still has room to run even after the recent move. But for a stock picker, you have to monitor the capex efficiency. If the return on that twelve billion dollars in capex doesn't materialize, the stock could get punished. But so far, the data points look good. Luna: So for listeners who want to look into this themselves, what are the key metrics they should track? Lucas: Three things. First, watch the data center revenue — that's the headline number. Second, look at the capital expenditure guidance and compare it to the growth in deferred revenue on the balance sheet, because that indicates how much of the capex is presold. And third, listen for any disclosure about DGX Cloud revenue. Right now it's lumped into a 'compute services' line, but as it grows, they may break it out. That would be a strong signal. Luna: And what about the competitive landscape? AMD has been gaining some ground, and there are new startups like Cerebras and Groq. Do any of them change the thesis? Lucas: They're worth watching, but none are at the scale where they threaten NVIDIA's dominance in training. Inference is more fragmented, but even there, NVIDIA's market share is around eighty percent. The real threat would be if a hyperscaler fully vertically integrates and stops buying NVIDIA chips. But that would take years and billions of dollars. So in the meantime, NVIDIA is using its cash flow to build a moat that gets deeper every quarter. Luna: And that's exactly the kind of thesis that belongs in a concentrated portfolio. You're not trying to own the whole market — you're betting on a few companies with compounding advantages. Lucas: Exactly. And with NVIDIA, you're getting a company that is both the leading supplier and a growing consumer of its own technology. That's a rare combination, and it's why I think this stock remains a core holding for anyone who believes in the long-term AI trend.