Is AI a Bubble? A Dot-Com Veteran Sees Familiar Warning Signs

A veteran of two major boom-bust cycles — the dot-com crash and the Great Financial Crisis — argues that the AI spending surge shares the same structural DNA as both: explosive real demand layered on top of dangerous leverage. The pattern has repeated across railroads, canals, electricity, and the internet, and the author believes it will likely repeat again with AI, even as he admits he hopes it doesn't.

The demand side of the AI boom is undeniably real. Anthropic's revenue rocketed from $4.8 billion in Q1 to $11.6 billion in Q2, and based on July's numbers, the company is now running at a $65 billion annualized rate — a 7x jump in a single year that the author calls unprecedented in corporate history. The Wall Street Journal has also reported Anthropic is now profitable, dismantling the widely held belief that frontier AI model companies can't make money. OpenAI, meanwhile, hit $40 billion in annualized revenue but saw sequential growth slow sharply — Q2 revenue rose just 18%, to $6.7 billion from $5.7 billion — leading the author to predict OpenAI will become the Netscape of the AI era: the company that sparked the boom and then got passed.

The leverage side is where the alarm bells ring loudest. Alphabet's off-balance-sheet spending commitments jumped by nearly $500 billion in just three months — five times the company's annual free cash flow — and for the first time in its public history, Alphabet burned cash rather than generating it in Q2. Meanwhile, Nvidia has structured deals that function as vendor financing: a $1.5 billion investment plus $105 billion in credit for an Ohio data center whose owner will likely buy Nvidia chips, and a separate $30 billion investment in OpenAI, which will use the cash to lease compute. The circular logic amplifies purchasing power across the entire AI ecosystem — classic bubble leverage.

The author draws a direct parallel to 2000, when internet supply finally caught up with demand, interest rates rose, and the leverage unraveled catastrophically. He notes that even the sharpest investors — citing George Soros's partner Stan Druckenmiller, who famously lost $3 billion on tech stocks after knowing the bubble existed — couldn't time the exit. The same cognitive trap applies today: being right about AI being a bubble doesn't mean anyone can successfully trade around it. AI spending already accounts for roughly a third of U.S. economic growth, so a bust wouldn't just hurt tech investors — it would likely trigger a broad recession.

The practical advice offered is deliberately undramatic: engage with the technology, invest proportionally, keep learning — but don't stake more than you can afford to lose. The AI boom may be in a 1997 moment with years of runway ahead, or it may be late 1999 with the cliff already passed. Nobody knows, and history suggests the people most certain they know are usually the ones who get hurt worst.