On 10 March 2000, the Nasdaq hit 5,048. Two and a half years later it had lost roughly 78% of its value. Today, with hundreds of billions of dollars flowing into data centres, chips and large language models, the question is back on every investment committee’s agenda: are we repeating history?
The short answer: it rhymes, but it doesn’t repeat. Some segments show clear bubble traits, and there are real differences that make a literal comparison misleading. The most useful historical analogy may not be the dot-coms at all, but the telecom boom.
What the dot-com bubble really was
Between 1995 and 2000, the internet was a genuinely transformative technology, and investors knew it. The problem was the price: thousands of companies with no profits, and often little revenue, traded on “eyeballs” and promises. When credit tightened and euphoria faded, buying dried up. Cisco, briefly the world’s most valuable company, fell about 90% from its peak. Amazon dropped more than 90% and survived. The technology won; most investors lost.
Where the parallels are real
1. Stretched valuations at the top of the market. Apollo chief economist Torsten Slok argues that today’s ten largest S&P 500 companies trade at higher multiples than their 1990s counterparts. In early 2026 the S&P 500 traded at about 23x forward earnings, the most demanding level since the dot-com era. In private markets, a Deutsche Bank analysis found some AI start-up valuations relative to revenue exceeding the price-to-sales ratios of around 30 reached by Microsoft or Oracle in 2000.
2. Investment increasingly funded by debt. Goldman Sachs expects hyperscaler capex of about $770 billion in 2026, roughly 100% of operating cash flow, so these companies are leaning on debt and equity issuance and cutting back buybacks. Morgan Stanley estimates AI-related global debt issuance could double to nearly $570 billion this year, with capex topping $1 trillion in 2027. Alphabet, for instance, plans $180–190 billion of capex against roughly $174 billion of operating cash flow. When spending exceeds what the business generates and is financed by credit, the margin for error shrinks.
3. Circular financing. Nvidia’s pledge to invest up to $100 billion in OpenAI, which in turn buys Nvidia chips, shows how invested capital can return partly as the investor’s own revenue. It isn’t illegal or necessarily reckless, but it makes real demand harder to separate from financed demand.
4. Concentrated risk. “Big Short” investor Steve Eisman has warned that OpenAI and Anthropic account for around 70% of AI revenue at major tech companies. And according to Goldman Sachs, nearly half of S&P 500 profit growth in 2026 comes from AI-related investment. If two or three players stumble, the effect is systemic.
5. Euphoria in the issuance market. OpenAI’s latest valuation stands at $852 billion, and Anthropic is reportedly preparing an IPO that could value it near $2 trillion, the largest ever. When the world’s most valuable companies come to market, the cycle is often late-stage. That was exactly the lesson of the 1999–2000 IPO wave.
Where the differences are real
Profits. Today’s tech leaders generate substantial earnings and cash. In 2000, most dot-coms had neither. This is the single most important difference.
Multiples. Bloomberg data cited in late 2025 put the Nasdaq 100 at about 26x forward earnings, versus more than 80x at the bubble’s peak. Goldman Sachs has noted that valuations are high but still some way below dot-com peaks.
Observable demand. AI already produces recurring revenue in the tens of billions. Enterprise usage is real, though its profitability and pace of adoption remain debated.
Mature infrastructure. The internet had to be built from scratch. AI sits on top of existing cloud, networks and payments, which speeds adoption.
The most balanced analyst reading is that there are bubble-like conditions in specific segments (private valuations of model developers, debt-financed infrastructure), not a market-wide frenzy.
The most useful comparison: the telecom boom
In the 1990s, carriers laid fibre far faster than near-term demand required, financed with heavy debt. The internet did grow, but capacity arrived before revenue, and companies like WorldCom and Global Crossing collapsed. Market commentators have already drawn this parallel with hyperscaler capex. It matters because it separates two questions: is the technology transformative (almost certainly) and is the pace and financing of the buildout sustainable (still open)? A bubble can form inside a real technology.
Warning signs to watch
- AI revenue vs. capex. If spending keeps outgrowing attributable revenue, pressure builds.
- Credit. Demand for AI-linked credit default swaps and hyperscaler bond spreads are better thermometers than headlines.
- Source of funding. More debt and more private credit tied to data centres raise contagion risk. We explored this in our article on private credit.
- The IPO window. On 8 October the Nasdaq fell 1.25%, its worst day since mid-August, after a report that OpenAI’s annualised revenue was lower than believed; Oracle fell as much as 5.5%. It shows how sentiment hinges on a few data points.
- Interest rates and power. Higher financing costs or electricity bottlenecks can delay projects.
What it means for investors
- Separate technology from price. An excellent company can be a poor investment if you overpay.
- Watch concentration. If your index portfolio already leans heavily on the same ten companies, real diversification is lower than it looks.
- Value with discipline. Revisit valuation methods and be wary of narratives without cash flows.
- Think long term. Bubbles punish those forced to sell at the worst moment. A long-term strategy with adequate liquidity reduces that risk.
- Be cautious in private markets. Private round valuations aren’t marked daily and can be slow to adjust.
Conclusion
The dot-com parallel is real in psychology, concentration and use of credit, and weak on the current profitability of the leaders. No one can time a turn; what investors can do is measure how much of today’s price already assumes a perfect future. AI will probably reshape the economy, as the internet did. The question isn’t whether, but how much of that future is already in the price.
This article is for information only and is not investment advice. Figures come from public sources and third-party estimates and may change.
Fuentes / Sources
- Intuition Labs, AI Bubble vs. Dot-com Bubble: A Data-Driven Comparison (feb. 2026): https://intuitionlabs.ai/articles/ai-bubble-vs-dot-com-comparison
- CO/AI, Apollo’s Torsten Slok on AI valuations: https://getcoai.com/news/ai-valuations-now-higher-than-1990s-dot-com-bubble-apollo-economist-warns/
- The Decoder, Deutsche Bank analysis of AI valuations: https://the-decoder.com/ai-valuations-soar-past-dotcom-era-highs/
- Axios, AI debt boom and Goldman Sachs capex estimates (jun. 2026): https://axios.com/2026/06/16/ai-nvidia-bonds-debt
- Morgan Stanley AI debt forecast (jun. 2026): https://www.tribuneindia.com/news/business/morgan-stanley-estimates-global-ai-debt-issuance-to-double-to-570-billion-in-2026-report/
- AI Weekly / Bloomberg, Alphabet capex and debt: https://aiweekly.co/alerts/alphabet-spacex-equity-surge-renews-ai-debt-binge-fears
- SiliconANGLE, market reaction of 8 Oct 2026: https://siliconangle.com/?p=849942
- 36Kr / WSJ, AI valuations and IPO plans: https://eu.36kr.com/en/p/3995319475541641
- Trading Random / Bloomberg, Nasdaq 100 multiples: https://www.itiger.com/news/1166181905
- Reuters Live Markets, hyperscaler capex and 1990s telecom parallel: https://www.itiger.com/news/2584923189
