Bubble watch
AI vs the Dot-Com Bubble: What Rhymes, and What Doesn't
Every AI rally now comes with a dot-com comparison attached. Some of it is fair. A lot of it is lazy. Here's the honest version — what genuinely rhymes with 2000, and where the analogy falls apart — and how to read it without pretending you can time a top.
What rhymes with 2000
The pattern-matching isn't wrong. Several things look uncomfortably familiar:
1. Concentration
A handful of names carry the index. In 2000 it was the large-cap techs; today a few mega-caps drive most of the move on any given day. Our bubble-risk index treats market concentration as one of its five pillars precisely because narrow leadership is fragile leadership.
2. A single, totalising narrative
"The internet changes everything" became "AI changes everything." When one story explains every move, price stops discriminating between companies — and that's when capital gets sloppy.
3. A capex arms race
Telecom over-built fibre into 2001; today hyperscalers are pouring tens of billions into data centres and accelerators. Some of that spend will earn its return. Some is being financed with debt — see the $32bn of AI data-center junk bonds the market is quietly underwriting.
4. Valuations that only work if growth stays perfect
Plenty of AI names trade at multiples that require many years of flawless execution. That isn't a prediction of doom — it's just a high bar. (More on how to read that bar below.)
What's genuinely different
Here's where the lazy version of the comparison breaks:
The leaders actually make money
Pets.com had no profits and no path to them. The 2026 AI leaders are among the most profitable companies in history, throwing off real cash flow. A market can be expensive and built on real earnings — those are not the same risk as 2000's profitless concept stocks.
The infrastructure is being used, not just promised
In 2000, a lot of the "build-out" was capacity nobody needed yet. Today the chips and data centres are running at capacity because there is genuine demand pulling on them. Whether that demand justifies every price is the open question — but the demand is real, not hypothetical.
Rates and balance sheets
The dominant AI names are largely self-funding their spend. The fragility this time sits more in the second tier — the debt-funded data-center and neocloud names — than in the cash-rich leaders.
So… is it a bubble?
Wrong question, or at least an incomplete one. "Bubble" is binary; markets aren't. The useful framing is how stretched is it, and where — which is exactly what a composite score is for. Pockets of genuine froth (story stocks, debt-funded capacity, anything priced for 40%+ growth forever) can coexist with real, cash-generative businesses in the same index.
A high reading on the bubble-risk index is a thermometer, not a timer. It tells you the air is thin. It does not tell you to sell — bubbles, real ones, have inflated for years. The next piece in this series is about exactly that trap: why "expensive" doesn't mean "short."
Santro AI is an informational tool. Market data is delayed ~15 minutes and provided for education, not as financial advice. "Hot" means attention, not direction.