Nvidia is the most valuable company in the world, its revenue grew 85% year over year in the most recent quarter, and its stock trades at a forward price-to-earnings multiple of roughly 22 to 23, a level it last saw before the AI boom began. Those three facts do not obviously fit together. A company compounding at 85% is supposed to command a premium multiple, not one that reverted to where it sat when it was a cyclical maker of gaming chips.
That mismatch is the whole story, and resolving it explains what the market actually believes about AI right now, which is more specific and more interesting than "bubble" or "no bubble."
The paradox, stated precisely
For most of the past three years, Nvidia's forward multiple oscillated well above 40, at times far higher, as the market priced it as the indispensable supplier of AI infrastructure. Its five-year average price-to-earnings sits around 38 times. Today the forward multiple is roughly 22, near the low end of its megacap peers and at a level not seen since before the world began pricing the company as the king of AI.
Here is why that is strange. A stock's P/E reflects what investors will pay for a dollar of earnings, and they pay more when they expect those earnings to grow fast and durably. Nvidia's earnings are growing about as fast as any large company's in history. Yet the multiple has compressed to where mature, slow-growing companies trade. The market is looking at record fundamentals and assigning them a pedestrian valuation, which means the compression is not about the current numbers, which are spectacular. It is about something the current numbers cannot show.
The resolution: the market is pricing duration, not existence
The key distinction is between two very different doubts. One is whether AI demand is real. The other is how long it lasts. The multiple tells you which one the market is expressing.
If investors doubted that AI demand was real, they would question Nvidia's earnings themselves, and the stock would fall on the numbers. That is not what is happening. The earnings are being believed and are coming through. What the compressed multiple expresses is doubt about duration, the market is willing to pay for the enormous earnings Nvidia is generating now, but it is not willing to pay a premium for many more years of the same, because it is not confident the current pace of AI infrastructure spending continues indefinitely.
This is a subtle but crucial reframe. A low multiple on high growth is the market saying: I believe the growth is happening, I do not believe it is permanent, so I will pay for the earnings in hand but not extrapolate them far into the future. It is skepticism about the tail, not the present. An Evercore analysis put the same point directly, that Nvidia's valuation compressed despite record fundamentals because investors are repricing the duration of AI spending rather than questioning its existence.
That is a far more rational market position than either euphoria or panic. It is neither "AI will grow forever," which would justify a multiple above 40, nor "AI is a mirage," which would crater the earnings estimates. It is "this is real and I am unsure how long it runs," which produces exactly the pattern visible: high earnings, believed, at a modest multiple.
Why duration is genuinely the hard question
The reason the market cannot resolve the duration question is that it depends on facts that do not exist yet, and this is where Nvidia's position is more uncertain than its financials suggest.
Nvidia's revenue depends on a handful of hyperscalers, Microsoft, Alphabet, Amazon, Meta, Oracle, continuing to spend enormous sums on AI infrastructure. That spending is a bet those companies are making that AI will generate returns justifying it, and so far the returns lag the investment: the revenue AI is currently producing is widely estimated to cover well under half of the infrastructure being built for it. The spending continues on the expectation that demand grows into the capacity. If that expectation holds, Nvidia's growth runs for years and the stock is cheap at 22 times. If the hyperscalers conclude the payback is too slow and trim their budgets, Nvidia's growth decelerates sharply, and the multiple is not cheap at all, because the earnings it is priced on would stop growing.
Notice that this is a decision, not a catastrophe. The risk to Nvidia is not that AI fails or that a scandal erupts. It is that a small number of rational, cash-rich customers choose to spend less because the near-term economics disappoint them. That kind of risk is unusually hard to price, because it hinges on judgment calls inside a few companies rather than on anything observable in Nvidia's own results, which is precisely why the market has settled on a wide-open "we don't know how long" rather than a confident number.
What the multiple is and isn't telling you
There is a real case that the stock is genuinely cheap, and it deserves stating. At 22 times forward earnings with 85% revenue growth, Nvidia trades below the semiconductor industry median multiple despite growing many times faster than the median semiconductor company, and its profitability is extraordinary, with return on invested capital above 100% and operating margins above 60%. On any mechanical comparison, a company growing this fast at this margin should not trade like a mature chipmaker, and several valuation models flag substantial theoretical upside on that basis.
But the mechanical comparison is exactly what the market is deliberately overriding, and it is worth respecting why. A P/E is not only a statement about the past year's growth. It is a bet on the growth to come, and the market has decided that Nvidia's future growth is less certain than its trailing growth, so it will not pay a growth multiple for it. The "cheapness" is conditional. The stock is cheap if AI spending sustains for years, and appropriately priced or expensive if the spending plateaus sooner. The low multiple is not the market missing an obvious bargain. It is the market pricing a genuine uncertainty that the trailing numbers cannot resolve.
That also means the valuation carries an embedded asymmetry worth understanding. Because the multiple is already compressed, a good deal of duration skepticism is priced in, so evidence that AI spending will persist, sustained hyperscaler capex, real enterprise monetization, would justify multiple expansion on top of earnings growth. Conversely, because the earnings estimates still assume strong growth, evidence that spending is slowing would hit both the earnings and the multiple at once. The setup is not one-directional in either direction, which is itself a sign of a market that has thought about this rather than one that is simply euphoric or fearful.
How to read it
The useful way to hold Nvidia's odd valuation is that the market has become sophisticated about this stock rather than emotional. It is not treating AI as a bubble to short or a certainty to chase. It is separating two questions that casual commentary blends together, whether AI demand is real, which it is pricing as yes, and how long the current spending pace lasts, which it is pricing as genuinely unknown, and paying full value for the first while refusing to pay a premium for the second.
For anyone trying to make sense of the stock, the takeaway is that the debate worth following is not "is AI a bubble." It is the duration question specifically: whether the hyperscalers sustain their spending long enough for the earnings Nvidia is already booking to keep growing. That question will not be answered by Nvidia's results, which look excellent, but by the capital-expenditure decisions of a handful of its largest customers and by whether enterprise AI revenue starts to catch up with the infrastructure built to deliver it. The compressed multiple on record growth is not a puzzle or a mispricing. It is the market saying, with unusual precision, that it believes what is happening and does not know how long it will last, and pricing exactly that.