Lila Soto: I bought a Hershey bar on the way in today — kind of intentionally, kind of as a prop for this conversation.
Hugo Vance: Mm, and the prop is doing more work than you think.
Lila Soto: The Hershey Company — thirty-five billion dollars, single-digit growth, very stable, very boring — trades at roughly thirty-one times trailing earnings. Nvidia, five-point-one trillion, eighty-two billion in one quarter's revenue, trades at eighteen to twenty-two times forward earnings. The market is valuing the candy company higher on that multiple than the most valuable listed company in the world.
Hugo Vance: Now, the word that matters in that framing is forward. The forward P/E divides today's stock price by projected earnings over the next twelve months. A lower number can mean the stock is cheap — or it can mean the market simply doesn't believe those projected earnings will materialize. And in Nvidia's case, at eighteen-to-twenty-two times, we are at the lowest forward multiple since early 2019.
Lila Soto: Which is — okay, that's the part that stopped me. Because in 2019 the AI cycle wasn't even really a thing yet, not at this scale. And the market paid more for the stock then than it does now, after Nvidia posted forty-nine billion dollars in free cash flow in a single quarter at a seventy-four-point-nine percent gross margin.
Hugo Vance: Indeed. The earnings have grown faster than the stock price. That is what multiple compression means — the denominator outran the numerator. And the question the market is implicitly answering is: how long does that denominator keep growing?
Lila Soto: So is the market wrong about Nvidia — or is it correctly scared about something the earnings report doesn't show?
Hugo Vance: That is the entire episode.
Lila Soto: But wait — that framing almost makes it sound inevitable, like compression just happens to successful companies. And I'm not sure that's the whole thing.
Hugo Vance: You see, let me try a plain version of it. Think of a restaurant that gets so popular so fast that even with a line out the door every night, investors are asking — can the chef actually cook at that pace forever? The line doesn't mean the price is wrong. It means the question has changed.
Lila Soto: Oh — the question changed. That's the click.
Hugo Vance: The core of it is this: Nvidia's multiple compressed not because the stock fell but because earnings grew so explosively they outpaced even its enormous stock gains. The denominator — earnings — just ran away from the numerator. That is unusual. That is actually almost without precedent at this scale.
Lila Soto: Which means Hershey's thirty-one times and Nvidia's eighteen-to-twenty-two times aren't — I mean, they're the same number but they're describing completely different anxieties.
Hugo Vance: Precisely. Hershey's thirty-one times reflects what Morningstar would call a stable earnings base — the market paying up for certainty, for predictability. Nvidia's eighteen-to-twenty-two times reflects earnings that have already outgrown even optimistic forecasts. The market is saying, well — we believe today's earnings. We are genuinely not sure about tomorrow's. Now, Morningstar has run a discounted cash flow model and argued Nvidia trades below fair value. And I don't dismiss that. But a DCF is only as solid as its long-term growth assumptions. Project the wrong growth rate ten years out and the intrinsic value number becomes — it becomes whatever you need it to be.
Lila Soto: So the Morningstar bull case isn't wrong — it's just load-bearing.
Hugo Vance: Yes. The entire structure rests on growth assumptions that skeptics are actively questioning. That is not a flaw in the method — that is the method working exactly as designed, revealing where the real argument lives.
Lila Soto: And here's where I think the skeptics actually have it right — the growth assumptions only hold if the customers keep buying. But Google is building TPUs. Amazon has Trainium. They're pouring billions into Nvidia chips at the same time they're funding the thing designed to replace Nvidia chips.
Hugo Vance: That is the structural fact I keep returning to. The hyperscaler AI capex cycle is Nvidia's entire revenue story — data center was seventy-five billion dollars of that eighty-two billion quarter. And every dollar of that capex is also, quietly, funding the customer's exit strategy.
Lila Soto: That's — I mean, that's a strange position to be in.
Hugo Vance: Intel lived it. PC makers spent the nineties dependent on Intel silicon, then spent the two-thousands methodically building around it — custom chipsets, alternative architectures — because indispensability, you see, eventually reads as a vulnerability to the people who depend on you. That pattern is not new. What's new is the speed.
Lila Soto: But — wait. CUDA. That's the part the Intel comparison doesn't fully handle, right? Because this isn't just a hardware dependency. An entire generation of AI researchers has built their entire workflow inside Nvidia's software ecosystem. You can't just swap the chip; you have to retrain the people.
Hugo Vance: CUDA is real. I won't dismiss it. A researcher in, say, a university lab in Seoul who's spent four years optimizing models inside that ecosystem — she doesn't switch because Amazon offers her a Trainium instance at a discount. The behavioral lock-in runs deeper than the hardware. That is a genuine circuit-breaker on the substitution timeline.
Lila Soto: So the custom silicon threat is real but slow.
Hugo Vance: Slow at the researcher level. Not slow at the infrastructure level. And the market already told us which one it's watching — during the AI sell-off, Apple gained roughly six hundred billion dollars in market cap as capital rotated out of Nvidia and SK Hynix. Six hundred billion. That's not a debate about long-run substitution. That's investors deciding the AI earnings cycle is — well, episodic is the word I'd use. Fragile. The capex could cool before CUDA loses its grip, and that's the scenario that compresses the multiple further.
Lila Soto: And that becomes an even stranger story when you realize this isn't just about one stock — the part we get to next is how Nvidia's earnings durability has basically become inseparable from the stability of the entire market.
Hugo Vance: And that's the part that actually changes the stakes here. At forty percent AI concentration in US equity weight, Nvidia's multiple isn't just Nvidia's problem anymore. It is, well — it is the market's multiple. If those earnings falter, you're not watching one stock correct. You're watching the index reprice.
Lila Soto: The nervous system analogy keeps coming back to me. Like — you can't price the nervous system the way you price a single organ.
Hugo Vance: And stress propagates. Micron Technology, SK Hynix — they're nodes in the same network. Any crack in Nvidia's demand picture hits memory suppliers inside the same earnings cycle. That's not correlation. That's structural adjacency.
Lila Soto: Wait — so the board raised the dividend and authorized new buybacks alongside the Q1 results, and the multiple still compressed?
Hugo Vance: Yes. Which is the telling detail. Buybacks and a dividend raise are, historically, a floor signal — management saying the cash is real and durable. And it hasn't stopped the compression. The market is looking through the signal.
Lila Soto: Okay, I keep thinking about a portfolio manager — San Francisco, Tuesday morning, Nvidia is now twelve percent of the fund because it kept working for five years. They want to reduce the concentration risk, but if they sell enough to matter, they become the reason the thesis breaks. They're actually trapped by their own success.
Hugo Vance: That bind is — I mean, that's not hypothetical, that's the actual mechanical reason concentration at this magnitude historically precedes mean-reversion rather than a permanent regime change. Everyone who wants to exit needs someone else to hold the position stable while they do. And at forty percent weight, there's no clean out.
Lila Soto: So the calibrated verdict isn't that Nvidia is cheap or expensive in isolation. It's that the market is correctly pricing what it would cost if the earnings don't hold — and at this concentration, that cost isn't a stock story. It's a system story.
Hugo Vance: Well, the place I've landed — and I'll admit it's uncomfortable — is this: if the bull case is right, if those earnings are durable, then Nvidia at eighteen-to-twenty-two times forward is genuinely a gift. You are buying the most consequential infrastructure company of the decade below what the market paid in early 2019, when the AI cycle was, frankly, a speculation. Below Hershey's thirty-one times. That is, by any historical measure, a strange sentence to say aloud. But if the custom silicon era arrives on schedule — if the hyperscaler capex cycle cools before CUDA's grip loosens — then the cheapness is a trap. And the market, I think, knows it cannot tell which one it's in.
Lila Soto: Yeah. And that's — I mean, that's the thing neither the bears nor the bulls can actually resolve right now. It's not whether Nvidia will grow. It's whether the market believes growth matters anymore at this concentration.
Hugo Vance: Exactly that. The question changed. Hershey bar on the table, five-trillion-dollar company on the screen, and the one priced lower is the one that just printed forty-nine billion in free cash flow. I find that genuinely hard to look away from — and genuinely hard to act on. Which is, perhaps, where we ought to stop.
Lila Soto: The prop earned its keep.
Hugo Vance: It did. Thank you for this.
