The dot-com bust's 25th anniversary: AI boom is a 'good bubble' — but the revenue hole is real
Mar 11, 2025
Key Points
- AI companies trade at tens or hundreds of billions with uncertain revenue paths, mirroring the dot-com boom's structure, but the underlying infrastructure—data centers and semiconductors—will likely prove foundational regardless of which companies survive.
- Unlike speculative bubbles in tulips or Beanie Babies, AI spending fuels adoption of revolutionary technology and builds productive infrastructure, similar to fiber-optic networks that became the internet's backbone after the 2000 crash.
- Early failures in transformative technologies often seed talent pools that drive later breakthroughs; General Magic's 2002 collapse produced iPhone and Android architects, suggesting today's AI spending will eventually justify itself even if individual companies fail.
Summary
Twenty-five years after the dot-com crash, the AI boom faces the same question: bubble or breakthrough? The NASDAQ hit its peak in March 2000 after a 500% run in five years. Companies like pets.com, Webvan, and the Globe.com collapsed, but the underlying ideas weren't wrong, just premature. Pets.com's core business became Chewy, a $10 billion company. Webvan was essentially Instacart. MP3.com went public on a domain-name business with no revenue; Spotify is now a dominant force.
The parallel anxiety is real. Leading AI companies trade at tens or hundreds of billions of dollars with uncertain paths to meaningful revenue. They are spending heavily on specialized semiconductors and building massive data centers, betting that productivity gains will eventually justify the spend.
A critical distinction separates good bubbles from bad ones. Bad bubbles are speculative bets on assets that don't make the economy more productive—tulip bulbs, Beanie Babies, Arizona desert real estate. Good bubbles fuel rapid adoption of revolutionary technology and leave behind valuable infrastructure. The fiber-optic networks built during the dot-com boom became the backbone for the internet era, just as railroads, canals, and electrical grids preceded their own booms and busts.
Carlota Perez, author of "Technological Revolutions and Financial Capital," argues that radical innovators must convince suppliers, workers, and financiers to march toward an imagined future. Many fail, but they lay important groundwork. Even if AI companies end up spending billions on data centers they don't know how to use yet, that infrastructure will eventually find purpose.
Nvidia, which makes the GPUs driving AI investment, is now the world's most valuable company at $2.7 trillion market cap. Tangible advances are already visible. Search is smarter, AI bots write code, agents book flights and file taxes, and productivity gains may follow. Not all AI companies will survive. Sonya Capital's David Khan has written about the "massive revenue hole" that AI companies must fill to justify their data center spending, a metric that could trigger a speculative shakeout. Yet Khan and others remain optimistic about the long-term value creation.
The smartphone revolution offers a template. By the time Steve Jobs unveiled the iPhone in 2007, the underlying technologies already existed: wireless networks, the internet, flash memory, touch screens. But it took General Magic, which failed in 2002 trying to build phones a decade too early, to seed the talent pool. Tony Fidel and Andy Rubin, both General Magic alumni, went on to develop the iPod, iPhone, and Android.
Early bets on revolutionary technology often look wasteful in the moment but prove foundational in hindsight. The question isn't whether AI is a bubble—it probably is, at least partly—but whether it's a good bubble that builds the infrastructure for a genuinely more productive economy.