Everyone arguing about whether Nvidia is expensive is arguing about the wrong number.
Nvidia trades at 35.97 times trailing earnings right now, with the stock at $210.06 and a market capitalization north of $5.09 trillion. Say that multiple out loud to someone in 2021 and they’d have nodded – that’s roughly what a “boring” large-cap software business fetched back then. Say it to someone who watched Nvidia trade at 119.8 times earnings in early 2023, right as the ChatGPT trade was catching fire, and they’d assume the stock had gotten cheaper. It has. Massively. The multiple has compressed by roughly two-thirds in three years, even as the share price went up several-fold.
That’s not a coincidence. It’s compounding.
The bear case on Nvidia has never really been “the multiple is too high” – anyone who’s watched this stock for a decade knows the multiple has always looked too high, right up until the earnings caught up and made it look cheap in hindsight. The real bear case, the one worth taking seriously, is about the denominator: can the earnings keep growing into the multiple, or is the AI capital expenditure cycle about to roll over the way every hardware supercycle eventually does?
D.A. Davidson’s Gil Luria has been the most articulate voice on this side of the table, and his argument isn’t “Nvidia is a bad company” – it’s that hyperscaler capex is a cyclical phenomenon dressed up as a secular one, and cyclical peaks are precisely when consensus estimates look most conservative and most wrong in the other direction. It’s a fair point. Capex cycles in semiconductors have never been permanently one-way, not once, not for any company, ever.
The bull case – and this is where Baird, Jefferies, and BofA analysts have converged on language that sounds almost identical across three separate research shops – is that Nvidia is “remarkably cheap” relative to its own growth rate and relative to the platform shift it sits at the center of. On forward-looking metrics, the picture shifts further: Nvidia’s forward P/E sits at 38.3 times and its EV/EBITDA at 33.8 times, against a gross margin that’s expanded to 71.1 percent and a return on invested capital above 135 percent – numbers that simply don’t exist elsewhere in hardware at this scale.
Here’s the tension nobody resolves cleanly: analyst consensus on Nvidia is about as one-sided as institutional research ever gets. Of 31 tracked ratings, 96.8 percent are bullish, with an average price target of $308.81 and a high estimate of $500 against a stock sitting at $210. Even the single most conservative price target on the Street, at $215, is still above where the stock trades today. When was the last time you saw unanimity like that on a $5 trillion company? Unanimity like that should make you nervous on its own terms – not because the bulls are wrong, but because a trade this consensus leaves no room for anyone to be surprised in the bearish direction without a violent repricing.
The prediction markets, for what it’s worth, have been pricing a scenario where Nvidia settles into the low $200s – not collapse, not euphoria, just gravity. That’s arguably the most useful data point in this entire debate, because it’s the only one not produced by someone with a research note to defend or a position to justify.
My read: the multiple compression story is real and it’s the single most under-discussed fact about this stock. Nvidia got cheaper as a business while getting more expensive as a stock price, and those are two different sentences that too many commentators collapse into one. The risk isn’t that Nvidia is “overvalued” in some abstract sense – a 36x trailing multiple on a company growing revenue at the clip Nvidia has posted isn’t obviously stretched. The risk is that the market has priced in a specific pace of hyperscaler spending that assumes zero digestion pause, zero pricing pressure from custom silicon, and zero macro shock. Any one of those assumptions breaking would matter more than the multiple ever will.
Comments are closed.