“AI is not a bubble because the technology isn’t going to disappear.” This phrase gets repeated every time someone questions the sector’s valuations, as if the longevity of a technology alone justifies any price. But that response confuses two different levels: the utility of an innovation and the market’s valuation of the companies aiming to capture it.
Internet didn’t vanish either. On the contrary, it ended up transforming the global economy, commerce, media, advertising, banking, personal relationships, and the way we work. Yet, the dot-com bubble burst between 2000 and 2002 caused huge losses. The Nasdaq Composite fell nearly 78% from its March 2000 high, while solid companies like Cisco, Intel, and Oracle experienced stock crashes over 80%. Many companies disappeared. Others survived but took years to regain their valuations.
That’s the lesson often forgotten in the current AI euphoria. A bubble doesn’t require the technology to be false, useless, or temporary. It only requires the asset prices to deviate too far from their real capacity to generate returns. There can be an authentic technological revolution and, at the same time, an overvaluation of those involved.
Internet triumphed, but many investors lost
The precedent of the dot-com bubble remains uncomfortable because it dismantles a very convenient narrative for the market. Amazon survived, reinvented itself, and became one of the world’s most important companies. But its stock price fell around 95% between December 1999 and October 2001. Those who bought near the peak weren’t wrong about the internet’s future, but they may have overpaid, in terms of price.
Cisco offers an even more sober case. In March 2000, it briefly became the world’s most valuable company, with a market cap over $500 billion. It wasn’t a company without real business or a hollow promise. It manufactured essential infrastructure for internet expansion. The issue was valuation. The market priced in perfect, sustained, and nearly unlimited growth. When reality normalized, the stock lost much of its value, and it took decades for the market cap to fully recover.
The image accompanying this discussion, with data from LSEG collected by the Financial Times, illustrates exactly that: some dot-com survivors took years—sometimes decades—to reach their pre-bubble highs again. Amazon achieved this sooner; Qualcomm nearly two decades; Cisco around 25 years to recover its bubble valuation level. Technology was the winner; the entry points for many investors, not.
The railroad history offers a similar lesson. The British railways of the 19th century weren’t an absurd fad. They transformed the economy, reduced transit times, and reshaped cities, trade, and industry. But the Railway Mania of the 1840s was a financial bubble. In 1846, the UK Parliament authorized 263 new railway companies, with proposed routes totaling about 15,300 kilometers. About a third of those lines were never built.
The trains kept running. The bubble burst.
The problem isn’t AI itself, but what’s discounted in the stock market
Generative artificial intelligence has already proven to be genuinely useful. It automates tasks, speeds up programming, improves customer service flows, transforms search, document analysis, design, cybersecurity, scientific research, and business operations. It’s not a financial decoration without product behind it. Unlike many dot-com companies of 1999, a significant part of the current cycle is led by companies with profits, cash flow, customers, and a dominant position.
NVIDIA is the clearest example. The company has become the main beneficiary of the AI data center boom by selling GPUs, complete systems, networks, and software to train and run models. By October 2025, it became the first publicly traded company to surpass $5 trillion in market capitalization, driven by AI chip demand. In 2026, various market estimates placed its value even above $5.4 trillion.
Size matters because it turns a single company into a massive piece of the market. According to a J.P. Morgan report published in 2025, AI-related stocks had explained a very high share of S&P 500 returns, earnings growth, and capital expenditure since ChatGPT’s launch. That alone doesn’t prove a bubble, but it shows a level of concentration hard to ignore.
| Indicator | Relevant Data |
|---|---|
| Nasdaq decline after the dot-com peak | About 78% |
| Amazon’s drop between 1999 and 2001 | Approximately 95% |
| Market cap of Cisco in March 2000 | Over $500 billion |
| New railway companies authorized in the UK in 1846 | 263 |
| Rail lines authorized but not built | About one-third |
| NVIDIA’s market cap in October 2025 | Over $5 trillion |
The core question isn’t whether AI will be important. It will. The right question is whether all current investments, multiples, data center plans, chip orders, and growth expectations will translate into enough profits to justify current prices.
That’s where the danger lies. When a company announces AI spending, the market rewards the promise of future productivity. When an infrastructure provider signs a supply agreement, it’s seen as nearly unlimited demand confirmation. When a tech firm raises its capex, it’s viewed as strategic investment, not potential excess. This climate may be rational for a time, but it can also lead to discounting years of growth before actual revenues materialize.
The bubble can coexist with the revolution
The comparison with internet shouldn’t be overly simplistic. Today’s market isn’t an exact replica of 2000. Big tech companies are far more profitable, have solid balance sheets, established businesses, and a global distribution capacity that didn’t exist 25 years ago. Moreover, AI is already integrated into products used by millions of individuals and companies.
But the lesson from tech bubbles isn’t that all companies are hollow. It’s that even good companies can trade at prices that turn a great technology into a poor short-term investment. Sometimes, the mistake isn’t in the thesis, but in the timing and valuation.
There’s also an industry risk. The race for AI requires data centers, energy, HBM memory, GPUs, networks, liquid cooling, fiber, land, and talent. If final demand doesn’t grow as investors expect, some capacity may remain underutilized or generate lower-than-expected returns. AI doesn’t need to fail for valuations to adjust sharply.
To say “technology is real” when uncertain about valuation is like saying “the house is nice” when asked if the purchase price makes sense. It can be attractive, useful, and well-located but still overpriced.
AI doesn’t need to disappear for a correction to happen. Internet didn’t vanish. Railroads didn’t disappear. Tulips didn’t vanish. Bubbles form not because a technology is worthless, but because the market starts paying as if everything will go perfectly, for everyone, all at once, with no margin for error.
Frequently Asked Questions
Is AI a bubble?
The technology doesn’t seem like a passing fad, but that doesn’t rule out excessive valuations. A bubble is measured by the relationship between price and future returns, not just the utility of the innovation.
Why compare AI to the dot-com bubble?
Because the internet was a real, transformative technology, but many companies traded at unjustifiable prices. Some disappeared, others took years to reach their previous peaks.
Can NVIDIA be a great company and be expensive?
Yes. A company can have an excellent business and still trade at a price that discounts too much future growth. Business quality and valuation are separate debates.
What should an investor look at beyond enthusiasm for AI?
Real revenues, margins, return on invested capital, customer concentration, data center spending, reliance on few suppliers, competition, and the ability to turn technological adoption into sustainable profits.
