Nvidia's 36x P/E Is the Question the AI Trade Hasn't Answered
The pitch arrived on 8 June 2026 in the same shape it has arrived for three straight years: buy the dip, trust the buildout, ignore the multiple. Two venture-aligned Telegram channels reposted the line within minutes of each other. By mid-morning Eastern time, Nvidia chief executive Jensen Huang had given investors a more quotable version of the same argument, telling markets via his public commentary that the selloff that began the prior week was a buying opportunity because the buildout of AI has only just begun.
That is the bullish case in its purest form. It is also the case that the market has been asked to accept at a price-earnings multiple that historically belongs to growth businesses with single-product monopolies — not to a chip designer whose customers are simultaneously building the very capacity that could compress its margins. The interesting story is not whether Huang is right about demand. He usually is. It is whether the price investors are paying today prices in a world where he might, for once, be early.
What the multiple is actually saying
A 36x trailing P/E for a $3tn-plus market-cap company is not, on its own, extreme. It is the company that sits on top of that multiple that matters. Nvidia's earnings have grown fast enough, for long enough, to justify a premium most of the time. The bull argument — the one circulating in venture and product-discovery feeds on 8 June — rests on a simple arithmetic identity: if AI capex from the hyperscalers compounds at anything close to the trajectory of the last eight quarters, the dollars flowing through Nvidia's data-centre business will outrun even an ambitious multiple.
The bear argument is also arithmetic, and it is the one the market started pricing in during the week of 1 June. Three of Nvidia's largest customers are now designing or co-designing their own accelerators. Two of them have publicly committed to scaling in-house silicon across multiple training and inference workloads. The third has signalled that custom parts will absorb a meaningful share of its 2027 build. None of that cancels the Nvidia franchise. But it does change the slope of the customer concentration curve, and the slope is what a 36x multiple ultimately discounts.
A more honest framing treats the current multiple as a function of three things: the size of the addressable market, the durability of Nvidia's margin structure, and the cost of capital investors are willing to accept while waiting for the first two to resolve. The addressable market looks enormous. The margin structure is being negotiated in real time. The cost of capital is the variable that is moving.
The Huang counter-narrative
Huang's public posture is not new, but its context has shifted. For most of 2024 and 2025, the "buy the dip" line was made when dips were 6 to 8 per cent and the next earnings print was weeks away. The 8 June messaging landed in the middle of a broader tech drawdown that had pulled in large-cap software, memory, and power-infrastructure names alongside the chip leader. That is a different audience and a different psychological state.
It is also a different competitive landscape than the one in which Huang first made the "buildout has just begun" formulation. In 2023, the constraint on AI deployment was supply of accelerators. By mid-2026, the binding constraints have moved downstream: power, cooling, data-centre shell, and the trained-talent pipeline. None of those bottlenecks are addressed by shipping more GB-series parts. They are addressed by capital deployment, grid interconnection, and the slow, unfashionable work of building physical infrastructure. Huang knows this. The investor feeds repackaging his comments rarely mention it.
The risk in taking the CEO's reassurance at face value is not that he is wrong about the trajectory of AI. It is that he is describing a category — the AI buildout — in which Nvidia's share of the wallet is no longer the same as Nvidia's share of the buildout. The buildout is a fact. The share is a forecast.
What the structural frame actually is
Strip the rhetoric away and the AI-equity trade is a bet on three things compounding at once: hyperscaler capex, model-deployment economics, and the willingness of public-market investors to fund the gap between the two. The first is observable in 10-Q filings and is currently enormous. The second is improving but uneven, and the unit economics of frontier-model inference remain heavily subsidised by the same hyperscalers that buy the chips. The third is a function of rates, sentiment, and the calendar.
The interesting shift of 2026 is that the third leg is wobbling while the first two remain intact. A 36x P/E on a stock that has already compounded at this rate is more sensitive to the discount rate than to the next earnings beat. If long-end yields drift higher for reasons unrelated to AI — fiscal trajectory, term premium, geopolitics — the multiple compresses before the fundamentals do. That is the mechanical reason the recent selloff felt disorderly even to investors who agreed with every word of the bullish case.
There is also a quieter structural question, and it concerns the relationship between Nvidia and the rest of the AI trade. A market in which one stock sets the temperature for an entire factor is, by construction, fragile to single-name shocks. The venture and product-discovery feeds celebrating the 8 June dip were not analysing Nvidia. They were treating it as a thermometer for the entire thesis. Thermometers do not need to be priced at 36x earnings.
Stakes and the next twelve months
The investors who act on the Huang line are not making a mistake. They are making a wager: that the buildout is large enough, that Nvidia's share of it holds, and that the cost of capital stays low enough to sustain a premium multiple while the wager resolves. Each of those three legs has a plausible path to disappointment, and none of them requires Nvidia to be wrong about the trajectory of AI.
The more interesting question is who pays if the wager is wrong. Nvidia itself is cash-generative, founder-led, and structurally well-positioned to ride out a multiple compression even if revenue growth slows. The exposure sits with public-equity investors who bought the trade at the top of the range, with index-tracking vehicles whose flows amplify the move, and with the venture and late-stage private investors whose marks are anchored to the public-market multiple. A 25 per cent drawdown in Nvidia would not break the AI thesis. It would, however, reset the venture book in ways that take eighteen months to work through.
What remains genuinely uncertain is whether the customer-concentration story is already in the price. The sources available on 8 June do not specify the size of the in-house-silicon commitments from Nvidia's three largest customers, nor the timeline over which those parts will absorb share. They also do not speak to the trajectory of long-end yields, which is the variable most likely to move the multiple independent of any company-specific news. The market is, in other words, being asked to underwrite a thesis whose two biggest risks are not yet visible in the data the public can see.
The desk note: Monexus treats the 8 June bull case on Nvidia as a stress test of the AI trade, not as a forecast. The wire coverage of Huang's remarks carried the reassurance; this piece tries to carry the question underneath it.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://t.me/producthunt
- https://t.me/AngelList