The AI Bubble: Are we in a dot-com bubble — or something different?
The S&P 500 trades at one of the highest valuations ever recorded. The Shiller PE ratio sits at 42.7×, just 1.5 points below the all-time record set in December 1999 — months before the NASDAQ collapsed 78%. So are we in a dot-com-style bubble?
The answer depends entirely on which metric you trust.
The NASDAQ 100 P/E says: not even close
At the dot-com peak in March 2000, the NASDAQ 100 traded at roughly 75 times trailing earnings. Today it sits at 36.8×. Elevated by historical standards, but barely half of 2000's reading.
Cisco at its peak commanded a price-to-sales ratio of around 29×. Nvidia peaked at a similar ~32× in November 2024. So on company-level multiples, the comparison holds remarkably well — except Nvidia's revenue then exploded, and its P/S has since compressed to ~17×. Cisco's revenue stalled, and its P/S simply crashed by 92%.
The Shiller PE says: yes, this is dot-com 2.0
CAPE uses ten-year average earnings, adjusted for inflation. It's at 42.7×. The historical median is 16×. The only time the market was ever more expensive on this measure was the final months of 1999.
The gap between today's P/E (modestly expensive) and today's CAPE (near-record) is itself the warning: corporate profit margins have roughly doubled over the past two decades, from a historic 6–8% to roughly 12–13% today. If margins stay there, the trailing P/E is the honest reading. If they revert — as they have every previous time they've reached extreme highs — CAPE is the honest reading, and prices need to fall a lot.
The poster stocks tell two stories
Cisco and Nvidia peaked at almost identical price-to-sales multiples. The difference is what happened next. Nvidia's revenue grew into its valuation. Cisco's didn't. That's the entire bull case for AI — the revenue is real this time.
The bear case is that AI capex is currently being funded by a handful of hyperscalers buying GPUs from each other in a circular flow. Cisco was perfectly profitable in 1999. The reason it crashed wasn't the company — it was that valuations assumed a growth path that depended on customers who themselves turned out to be financed by speculative capital.
The data
Three valuation metrics, three different stories
Dot-com peak P/E
~75×
AI era peak so far
~38×
Current (May 2026)
36.8×
% of dot-com peak
~49%
Sources: NASDAQ 100 P/E from GuruFocus (current 36.8×, historical median 24.5×); dot-com peak ~75× trailing. Shiller PE from multpl.com / Robert Shiller (live 42.66). Cisco/Nvidia P/S calculated from market cap ÷ trailing 12-month revenue.
What would it actually take?
The math: 75 ÷ 36.8 ≈ 2.04. To reach dot-com peak valuations from here, the NASDAQ 100 needs roughly a 2× ratio change. That could come from prices doubling, earnings halving, or some combination. Try a few:
If prices change by X and earnings change by Y, what's the new P/E?
Resulting P/E
36.8×
vs current (36.8×)
+0.0×
vs dot-com peak (75×)
−38.2×
Quick scenarios:
new P/E = 36.8 × (1 + price%) ÷ (1 + earnings%). Current NASDAQ 100 trailing P/E: 36.8× (May 2026, GuruFocus). Dot-com peak: ~75× trailing (Mar 2000).
So is it a bubble?
Honestly: it depends on whether you believe today's record-high corporate profit margins are permanent. By current earnings, the AI bubble looks roughly half as extreme as dot-com. By ten-year smoothed earnings, it looks essentially identical. The difference between the two is the entire trillion-dollar debate.
We are not in 1999. But we are also not somewhere historically normal. The honest answer is that the market is priced as though everything currently going right continues going right for another decade. If you believe that, prices make sense. If you don't, you're looking at the second-most expensive market in 155 years.