Comparing today’s AI-led technology market to the final quarter before the dot-com bust
Clients understandably ask whether the run-up in AI-related technology stocks resembles the final months before the Nasdaq’s March 10, 2000 peak. The honest answer is mixed: public-market valuations are considerably less extreme than 2000, but capital intensity, leverage, and a handful of technical warning signs have converged on — or exceeded — dot-com-era extremes. The data below lays out both sides plainly.
FUNDAMENTAL & VALUATION INDICATORS
| Indicator | March 2000 Peak | June 2026 |
| Nasdaq-100 fwd P/E | ~60x | ~24x |
| Nasdaq Composite P/E | >90x (1999) | Not comparably extreme |
| Profitable tech IPOs | ~14% | Public leaders profitable; risk is in private AI cos. |
| Tech capex / GDP | Cycle peak | ~4.4% of GDP — matching dot-com peak |
| IPO market | 476 (’99) → 80 (’01) | Reopening fast; SpaceX IPO >$2T mkt cap |
TECHNICAL & SENTIMENT INDICATORS
| Indicator | March 2000 Peak | June 2026 |
| Index momentum | +572% (’95–Mar ’00) | Multi-year uptrend, capex-driven |
| Breadth | Narrowing into top | ~56% of S&P above 200-DMA vs. new highs |
| RSI (14-day) | Persistently overbought | S&P stretched; Nasdaq Composite ~57 (neutral) |
| Margin debt | Elevated | Record high |
| Insider activity | Heavy selling into top | Sell/buy ratio ~4.8x (Jan ’26), $21B sold in March alone |
Sell/buy ratio: shares sold by corporate insiders for every share bought. A ratio of ~4.8x means insiders sold roughly five shares for every one purchased — the most lopsided reading since 2021, which also preceded a market peak. Routine 10b5-1 diversification explains some of this, but the magnitude still stands out.
WHERE THIS CYCLE DIFFERS FROM 2000
- Earnings quality is real today. The 2000 bust hit public companies directly because the speculative excess was already on the index — unprofitable dot-coms with triple-digit P/S ratios. Today’s largest tech names (Microsoft, Alphabet, Meta, Nvidia) are genuinely profitable; the comparable excess sits mostly in private AI valuations (OpenAI, Anthropic, and peers) that have not yet been marked to a public index.
- The risk has migrated to the balance sheet. Hyperscaler capital expenditure now consumes 45–57% of revenue — ratios more typical of utilities than software companies — financed increasingly with debt. Projections call for as much as $1.5 trillion in new sector debt issuance over the coming years to fund AI infrastructure.
- Breadth and insider behavior warrant attention. Roughly 56% of S&P 500 constituents trade above their 200-day moving average even as the index sets new highs — a divergence pattern that preceded prior corrections. Insider sell/buy ratios have also reached multi-year extremes.
THE BOTTOM LINE
This is not 1999–2000 on a P/E basis. It increasingly resembles 1999–2000 on a capital-intensity and financing-risk basis. The next test is not whether AI is real — the revenue growth at the model layer is genuine — but whether hyperscaler returns on roughly $600–700 billion in 2026 capex arrive before depreciation, debt service, and earnings scrutiny force the market to decide for them.
At THALASSA CAPITAL we have also developed a spreadsheet to help run different Capex scenarios and measure stock market’s sensitivities to the different levels. Should you want a copy of the Excel file, please contact our team.
Prepared by Thalassa Capital LLC. This material is for informational purposes only and does not constitute investment advice or a recommendation to buy or sell any security. Past performance and historical comparisons are not indicative of future results.