AI bubble
AI bubble vs dot-com bubble: 8 metrics compared with Q2 2026 data
Is 2026 a rerun of 2000? Eight metrics compared with data through Q2 2026: valuations, profits, concentration, breadth, and hyperscaler capex-to-revenue from SEC filings (35% in Q2 2026, up from 14% in Q1 2024).
“This is just like the dot-com bubble” is the most repeated line in the AI market — and one of the laziest. The 2000 crash is a useful reference if you compare the right things. So here are eight concrete comparisons between the AI boom and the dot-com bubble, each pointing in a slightly different direction. Together they give a far more honest answer than the headline.
This is educational, not investment advice (see the disclaimer), and the figures are date-sensitive — treat them as of mid-2026; the capex metrics below are refreshed from SEC filings each quarter (last read: 2026-10-12).
1. Valuations: high, but not 2000-high
At the dot-com peak, flagship names traded at extreme multiples — Cisco above roughly 200 times earnings. In 2026, the AI leaders trade richly but far lower; Nvidia, for instance, well under 50 times earnings. Verdict: stretched, not insane. The single most-cited “it’s a bubble” comparison is the one the data least supports.
2. Profits: the decisive difference
Many dot-com stars had little revenue and no profit — they were stories with tickers. The AI cycle’s leaders are among the most profitable, cash-generative companies on earth. Verdict: this is the strongest argument that 2026 is not 2000. Expensive businesses are not the same as empty ones.
3. Concentration: more extreme than 2000
Here the comparison cuts the other way. The top ten S&P 500 names are around 40% of the index — more concentrated than at the dot-com peak. A few AI-linked stocks now drive the whole market, so a wobble in them is a wobble in nearly everything. Verdict: worse than 2000 on this measure — see how to measure your own concentration.
4. Capex: real spending, unproven return
Dot-com capital famously went into “dark fibre” that sat unused for years. The AI equivalent is the staggering spending on data centres and chips. The difference is that today’s spenders are hugely profitable and can afford it — but the return on that capex is still unproven. Verdict: the key open question. If the spending pays off, it looks visionary; if not, it becomes a drag.
5. Breadth: a narrow boom
The dot-com mania was broad — huge numbers of speculative companies floated and soared. The AI boom is narrower and more concentrated in established giants and a few labs. Verdict: fewer obviously worthless companies, but more risk hidden inside “safe” index funds, where the concentration lives.
6. Hyperscaler capex-to-revenue: the number that moves every quarter
The dot-com build-out put roughly 5–10% of telecom revenue into fibre at the peak. The AI build-out is far more intense: four hyperscalers (Alphabet, Amazon, Meta, Microsoft) spent 35% of revenue on capital expenditure in the quarter ending 2026-06-30 — 165 billion dollars against 471 billion of revenue — up from 13.7% in Q1 2024. The series below is computed directly from 10-Q and 10-K filings on SEC EDGAR and refreshed automatically; the live chart is on the hyperscaler capex indicator.
| Quarter ending | Capex, USD bn | Revenue, USD bn | Capex / revenue |
|---|---|---|---|
| 2024-03 | 44.3 | 322.2 | 13.7% |
| 2024-06 | 52.9 | 336.5 | 15.7% |
| 2024-09 | 58.9 | 353.3 | 16.7% |
| 2024-12 | 72.3 | 402.3 | 18% |
| 2025-03 | 71.9 | 358.3 | 20.1% |
| 2025-06 | 88.2 | 388.1 | 22.7% |
| 2025-09 | 97.3 | 411.4 | 23.6% |
| 2025-12 | 118.6 | 468.4 | 25.3% |
| 2026-03 | 129.8 | 430.6 | 30.1% |
| 2026-06 | 165.1 | 471.2 | 35% |
Verdict: the intensity has more than doubled in two years. Nothing in the dot-com era matched a spend ratio like this among profitable incumbents — which is exactly why metric 4 (unproven return) is the swing factor.
7. Funding source: cash flow versus equity
Dot-com capex and losses were financed by the market itself: IPO proceeds, secondary offerings and vendor financing, which is why the bust propagated the moment equity issuance stopped. The AI build-out is financed mainly from the operating cash flow of the same four companies, with debt issuance growing in 2025–2026 but still a minority of the total. Verdict: a slower-burning fuse. A correction would hit share prices first and the build-out only later — unlike 2000, when the money stopped at once.
8. Time to profit at the leaders
Amazon posted its first full-year profit in 2003, nine years after founding; most dot-com names never got there. The AI cycle’s listed leaders were profitable before the cycle began, while the frontier labs (OpenAI, Anthropic, xAI) remain private and loss-making, funded by those same incumbents. Verdict: the listed market is not where the dot-com-style losses sit; they sit one layer down, in private valuations that public investors see only through the hyperscalers’ capex.
Adding it up
| Comparison | Which way it points |
|---|---|
| Valuations | Less extreme than 2000 (bullish) |
| Profits | Far stronger than 2000 (bullish) |
| Concentration | More extreme than 2000 (bearish) |
| Capex return | Unproven (the swing factor) |
| Breadth | Narrower; risk hidden in indices (mixed) |
| Capex intensity | Far higher than 2000 (bearish if returns lag) |
| Funding source | Cash flow, not equity issuance (bullish) |
| Time to profit | Leaders already profitable; losses sit in private labs (mixed) |
The honest conclusion isn’t “bubble” or “no bubble.” It’s that 2026 is a stronger, narrower, more concentrated market than 2000 — one that can still correct hard, but isn’t built on the same emptiness. The risk is real; the “rerun of the dot-com bust” framing is too simple.
If you want to size your own exposure rather than argue the analogy, a research tool such as Simply Wall St can visualise valuations and concentration (affiliate link — see our disclosure). And for the full treatment, see the pillar on whether we’re in an AI bubble.
Educational content, not investment advice. Market figures are as of mid-2026 and change continuously; verify before acting.
We report facts with their date and caveats. We never label a named company as fraudulent or "AI-washing" as a statement of fact — we present verifiable data and the questions an investor should ask.
Frequently asked questions
Is the AI boom the same as the dot-com bubble?+
There are real similarities — extreme enthusiasm, heavy capital spending and high concentration — but key differences too. The biggest is profitability: the leaders of the AI cycle are highly profitable and generate real cash, whereas many dot-com darlings had little revenue and no profit. Valuations today are high but far below dot-com peaks.
Are AI stocks as overvalued as dot-com stocks were?+
By the headline measure, no. At the 2000 peak some flagship names traded above 200 times earnings; in 2026 the AI leaders have traded at far lower multiples — high, but nowhere near dot-com extremes. That doesn't rule out a correction; it means the comparison to 2000 is overstated.
What could make the AI boom end like the dot-com bust?+
The clearest risk is that the enormous capital spending on AI infrastructure fails to produce the expected returns — spending that turns into depreciation without the revenue to justify it. Combined with extreme index concentration, that could drive a sharp correction even though the underlying businesses are far stronger than in 2000.
Related reading