AI: Bubble or Revolution? What the Dot-Com Crash Actually Teaches Us

The Numbers That Make People Nervous

Nvidia's share price rose roughly 800% between early 2023 and early 2026. The company briefly became the most valuable in the world — a chipmaker, valued higher than the entire energy sector. Its revenue grew from $27 billion in fiscal 2023 to over $130 billion in fiscal 2026, almost entirely driven by demand for AI training hardware.

Microsoft invested $13 billion in OpenAI. Google restructured its entire product line around Gemini. Amazon poured billions into Anthropic. Meta hired 20,000 engineers and spent over $30 billion on AI infrastructure in a single year. Startups with no revenue raised at multi-billion dollar valuations on the promise that their large language model would be the one that mattered.

If you've read anything about the late 1990s, this feels familiar.

The Dot-Com Playbook

Between 1995 and 2000, the Nasdaq Composite rose over 400%. Companies with no profits, no clear business model, and sometimes no product at all were valued in the billions. Pets.com, Webvan, eToys — names that became shorthand for irrational excess.

The thesis was simple: the internet would change everything, and early movers would capture enormous value. Money poured into anything with ".com" in the name. Valuations detached from fundamentals. Analysts argued that traditional metrics like price-to-earnings ratios were outdated — this was a new paradigm.

Then it collapsed. The Nasdaq fell 78% between March 2000 and October 2002. Trillions in market value evaporated. Companies that had been worth billions were worth nothing. Careers, savings, and retirements were destroyed.

The parallels to today's AI market are hard to ignore:

• Astronomical valuations driven by a transformative technology
• Massive capital expenditure by incumbents racing to dominate
• Startups raising billions with limited revenue and unproven business models
• A belief that traditional valuation metrics don't apply because the opportunity is unprecedented
• Concentration of market returns in a handful of companies

So is AI the next dot-com bubble?

What People Get Wrong About the Dot-Com Crash

Here's the part that's usually left out of the cautionary tale: the dot-com bulls were right about the technology. They were just wrong about the timing and the valuations.

The internet did change everything. E-commerce did become enormous. Online advertising did become the dominant business model for media. Cloud computing did reshape enterprise software. Every prediction the dot-com enthusiasts made about how the internet would transform business came true — it just took 10-15 years longer than they expected, and the winners weren't the companies people bet on in 1999.

Amazon survived the crash — barely. Its stock fell 93% from peak to trough. But the company was real, the model worked, and by 2020 it was worth over $1 trillion. Most of its dot-com peers went to zero.

Google didn't even exist during the bubble's peak. It went public in 2004, four years after the crash, and became one of the most valuable companies in history. Facebook launched in 2004. Netflix pivoted to streaming in 2007. The real internet winners mostly emerged after the bubble burst.

The lesson isn't "transformative technologies are bad investments." The lesson is:

1. The technology can be revolutionary and the stocks can still be overvalued.
2. Most of the companies that ride the hype wave will fail.
3. The ultimate winners are often not the early favourites.
4. The crash doesn't invalidate the technology — it clears out the excess.

Where AI Is Different

AI in 2026 is not identical to the internet in 1999. Some important differences cut against the bubble narrative:

Revenue is real. Nvidia isn't a speculative bet — it earned over $130 billion in revenue in its last fiscal year. Microsoft, Google, and Amazon are integrating AI into products that already generate hundreds of billions in revenue. This isn't Pets.com. The underlying businesses are enormously profitable.

Adoption is faster. ChatGPT reached 100 million users within two months of launch — faster than any consumer product in history. Enterprises are deploying AI tools across legal, finance, engineering, customer service, and healthcare. The gap between "interesting demo" and "production deployment" has been shorter than with any previous technology wave.

The incumbents are leading. In the dot-com era, startups disrupted incumbents. In AI, the incumbents — Microsoft, Google, Amazon, Meta — are the ones spending the most and moving the fastest. They have the data, the compute, the distribution, and the capital. This makes a total market collapse less likely, because the companies driving AI investment can absorb losses that would destroy a startup.

Where AI Follows the Script

But other patterns look uncomfortably familiar:

Capital expenditure is running ahead of revenue. The big tech companies collectively spent over $200 billion on AI infrastructure in 2025. The revenue from AI-specific products and services is growing fast but hasn't caught up. If the return on that investment takes longer than expected — or is smaller than projected — the writedowns will be significant.

Valuations assume perfection. Nvidia's valuation prices in years of continued dominance in AI chips. But competition is intensifying — AMD, Intel, Google's TPUs, Amazon's Trainium, and a wave of AI chip startups are all targeting the same market. Nvidia's margins are extraordinary by any historical standard, and extraordinary margins attract competition.

Most AI startups will fail. There are now thousands of AI startups, many valued at over $1 billion. History is clear on what happens to the majority of companies in a gold rush: they run out of capital before they find a sustainable business model. The picks-and-shovels companies (Nvidia, cloud providers) tend to do better than the miners, but even that isn't guaranteed.

Concentration risk is extreme. If you hold a global tracker fund, the "Magnificent Seven" tech stocks — Apple, Microsoft, Nvidia, Amazon, Alphabet, Meta, and Tesla — represent roughly 30% of the S&P 500 and a significant chunk of any global index. You're not just exposed to AI through a few stocks; you're exposed through the structural mechanics of index investing. A correction in AI sentiment hits your entire portfolio.

What Happened to Dot-Com Investors

This is where it matters for retirement planning. The Nasdaq peaked in March 2000 at around 5,048. It didn't recover that level until April 2015 — fifteen years later.

An investor who put a lump sum into the Nasdaq at the peak in March 2000 and held through the crash would have waited a decade and a half just to break even in nominal terms. Adjusted for inflation, it took even longer.

But an investor who was pound-cost averaging — putting in a fixed amount each month — had a completely different experience. They bought through the crash at dramatically lower prices, accumulated far more shares, and were in profit years before the index recovered its peak.

And an investor who held a diversified portfolio — not just tech stocks but bonds, international equities, and real assets — experienced the crash as a painful but manageable drawdown rather than a catastrophe.

The people who were destroyed by the dot-com crash were overwhelmingly those who were concentrated in tech stocks and made one of two mistakes: they invested a lump sum near the peak, or they panic-sold after the crash and never bought back in.

What This Means for Your Portfolio

None of this tells you whether AI stocks will crash next month or keep rising for another five years. Nobody knows. But it does suggest some things worth thinking about:

Check your concentration. If you're in a global tracker fund, look at the top 10 holdings. If 30%+ of your portfolio is in a handful of US tech stocks, you're making a concentrated bet on the AI narrative whether you intended to or not.

Don't confuse the technology with the trade. AI is almost certainly transformative. That doesn't mean Nvidia at 35x earnings is a good investment, and it doesn't mean it's a bad one. The technology being real is necessary but not sufficient for the stocks to be fairly valued.

Zoom out. The dot-com crash felt like the end of the world for tech investors. In the context of a 30-year retirement plan, it was a blip. Even someone who was 100% in the Nasdaq at the worst possible time recovered within their investment horizon — assuming they didn't sell. The crash only became permanent for people who locked in their losses.

Diversification is boring and effective. A portfolio split across geographies, asset classes, and sectors would have experienced the dot-com crash as a 20-30% drawdown rather than a 78% one. That's the difference between an uncomfortable year and a financial disaster.

Keep investing through it. If you're contributing monthly to a pension or ISA, a crash is mathematically in your favour. You're buying more units at lower prices. The worst thing you can do is stop contributing because the market is down. The second worst thing is selling what you have.

The Only Honest Prediction

AI might be in a bubble. It might not be. The honest answer is that bubbles are only identifiable in hindsight, and anyone who tells you with certainty that this is or isn't one is guessing.

What we can say with confidence, based on over a century of market history:

• Transformative technologies create real value and speculative excess simultaneously
• Most investors in a technology wave lose money, even when the technology succeeds
• Diversification, patience, and consistent investing protect you from the worst outcomes regardless of whether a bubble forms or bursts
• The biggest risk isn't the market — it's your reaction to it

The AI revolution might look like the internet: real, transformative, and ultimately rewarding for patient, diversified investors — but devastating for those who concentrated their bets and couldn't stomach the volatility along the way.

If you want to see what a tech correction would actually do to your retirement plan, you can run a free simulation. Model a 40% drawdown in year one and see where you end up. The answer might surprise you.

Further Reading

• Malkiel, B. (2023). A Random Walk Down Wall Street, 13th Edition. W.W. Norton.
• Chancellor, E. (2000). Devil Take the Hindmost: A History of Financial Speculation. Plume.
• Perez, C. (2003). Technological Revolutions and Financial Capital. Edward Elgar Publishing.
• Shiller, R. (2015). Irrational Exuberance, 3rd Edition. Princeton University Press.
• Damodaran, A. (2024). "Nvidia: Priced for Perfection?" Musings on Markets blog.
• SPIVA U.S. Scorecard, Year-End 2025. S&P Dow Jones Indices.