Nvidia at Its Lowest Forward P/E in a Decade: What That Number Actually Means

By Jake, founder of StockResearch.app

In the aftermath of the July 2026 AI selloff, a striking statistic made the rounds: Nvidia was trading at its lowest forward P/E of the last ten years. Gavin Baker cited it on Invest Like the Best, noting that the only comparably cheap moments for semiconductors were the DeepSeek panic and the Liberation Day tariff bottom, both V-shaped lows in hindsight.

"Cheapest in a decade" is a headline designed to end conversations. This post is about starting one instead, because that statistic contains one number and at least three assumptions, and you cannot evaluate it without taking it apart.

Forward vs. trailing: which E are we talking about?

A P/E ratio divides price by earnings, but there are two very different earnings numbers you can use:

• Trailing P/E uses the last twelve months of reported earnings. It is a fact. It is also backward-looking: for a fast-growing company, the last twelve months can wildly understate (or overstate) the run-rate.
• Forward P/E uses the next twelve months of estimated earnings, typically the consensus of sell-side analyst forecasts. It is forward-looking, which is what you want, but it is a forecast, which is the catch.

For a company growing slowly, the two numbers are close. For a company whose earnings are compounding fast, they diverge dramatically. A worked example: a stock at $100 that earned $2/share last year trades at a trailing P/E of 50, which sounds expensive. If consensus expects $4/share next year, the forward P/E is 25. Same price, same company, half the multiple. The difference is purely which earnings you divide by.

So "Nvidia at its lowest forward P/E in a decade" means: price, divided by what analysts collectively expect Nvidia to earn over the next year, is lower than at any point in ten years. Lower than during the 2018 crypto-bust drawdown, lower than the 2022 rate-shock collapse, lower than the DeepSeek and tariff panics of 2025.

The mathematical meaning of compression on accelerating earnings

Here is the part worth slowing down for. A multiple compresses when price falls relative to expected earnings. There are two ways that happens:

1. Price falls while estimates hold. The market pays less for the same expected earnings. 2. Estimates rise while price lags. The earnings expectations grow underneath a stagnant or falling price.

Through mid-2026, Nvidia was in both camps at once: estimates had been rising, with the business accelerating, supply-constrained, and the next product cycle (Rubin) priced at a premium, while the price fell 40%+ from its highs in the July cascade. Both blades of the scissors were compressing the multiple.

Now the key insight, and it is arithmetic, not opinion: a P/E is the market's confidence in the E, expressed as a number. When the market pays its lowest-in-a-decade multiple for earnings that are currently accelerating, the market is not saying "we did not notice." It is saying, in Baker's words: "The market 100% thinks they're significantly over-earning." The low multiple is the price of a widely held belief that today's E is a cyclical peak: that AI capex will crack, GPU pricing will collapse, and the real, sustainable E is far below the consensus estimate.

Put differently: a historically cheap forward P/E on accelerating earnings is mathematically equivalent to the statement "the market disbelieves the earnings estimates." That is the actual claim embedded in the statistic. Everything else is packaging.

Which resolves into exactly two scenarios

If the estimates are roughly right, if the earnings arrive and keep growing, then the multiple was genuinely, historically cheap, and the July price will look like the DeepSeek and Liberation Day lows do in the rearview mirror: brief windows when a compounding business was available at a skeptic's price. If the market is right, if this is peak over-earning, then the "cheap" multiple is an illusion. Divide today's price by the earnings that actually materialize after an AI capex downcycle, and the true multiple could be average or worse. Cyclical businesses famously look cheapest at the top: memory stocks, oil drillers, and shipping lines have all taught generations of investors that a single-digit P/E on peak earnings is a trap wearing a bargain's clothes.

The honest position is that the statistic cannot tell you which scenario you are in. What it can do is tell you precisely what question to research: is there evidence the E is breaking? That is a fundamentals question with observable inputs (GPU pricing, hyperscaler capex and operating cash flow, supply commitments, token demand), and we will turn it into a concrete checklist in the next post in this series.

The historical base rates, honestly presented

The bull citation: the two prior "cheapest semi moments" of the era, DeepSeek (January 2025) and Liberation Day (April 2025), were both V-bottoms. Extreme multiple compression on intact fundamentals resolved upward, fast, both times. July 2026 also had a mechanical amplifier that those episodes lacked: a significant portion of the decline traces to a single leveraged fund's forced liquidation, which is flow, not information. The flow-versus-fundamentals distinction is readable in data if you know where to look.

The bear rebuttal, which deserves equal airtime: two data points is not a base rate. Both prior V-bottoms occurred with ZIRP-era reflexes intact and before credit conditions tightened. And the one bear case Baker himself says he respects is credit: wider spreads, hyperscaler CDS blowing out, debt-financed data-center buildouts that unwind violently if monetization disappoints. There are also decades of counterexamples where "cheapest in years" preceded "cheaper still": Cisco spent 2001 setting decade-low multiples on estimates that kept being wrong.

And one more caveat the headline hides: forward P/E inherits every flaw of consensus estimates. Analysts revise slowly, anchor on company guidance, and historically miss cyclical turns in both directions. A decade-low forward P/E computed from stale estimates is a decade-low ratio of price to a number nobody should fully trust.

How to actually use a statistic like this

Strip the headline to its useful core and you get a two-axis question:

1. Where is the multiple relative to this company's own history? (Percentile, not vibes: a "low" P/E only means something against the company's own range.) 2. What is the fundamental trajectory doing at the same time? (Accelerating, flat, or rolling over?)

Compression on deteriorating fundamentals is the market repricing a worse business, usually correctly. Compression on accelerating fundamentals is the market making a specific, checkable bet against the estimates. The second pattern is rarer and far more interesting, in either direction.

You can run axis one today: our free guide to charting historical P/E ratios walks through it, and the valuation tools on StockResearch chart daily P/E, P/S, and EV/EBITDA history (up to five years, computed daily rather than quarterly) so you can see exactly where a current multiple sits in its own distribution. Axis two is next post's subject: turning "are the fundamentals actually deteriorating?" into a measurable checklist.

The statistic says the market disbelieves Nvidia's earnings. The research question is whether you can find the evidence that settles it.

---

Jake is the founder of StockResearch.app, where he writes data-first research on valuation dislocations. This article is for informational purposes only and does not constitute financial advice. It discusses valuation statistics and publicly reported figures, not the merits of any investment. Nothing here is a recommendation to buy or sell any security. Always do your own research before making investment decisions.