Ghosts of Bubbles Past

The ghosts of past bubbles are spooking AI investors. Skepticism is creeping into valuations. Rising interest rates are stranding private-equity funds in zombie territory. Gorging on origination fees for years, nonbanks stumble as refinancing dries up and housing stalls. And investors are haunted by a nagging doubt…Could irrational exuberance be inflating an AI bubble like the dot-com or housing bubbles?

The dot-com boom boosted the Nasdaq nearly sevenfold in less than five years. It climbed 40% in 1998, 86% in 1999, before peaking in early 2000 at around 5,050. The bust was brutal. It left the index just over 1,100 two years later, erasing nearly $5 trillion in market value. The boom and bust left a graveyard of failed companies: Global Crossing, Webvan, Pets.com, Boo.com, and others. And the survivors were severely punished: Priceline plunged from $974 to $6.60; Amazon from $113 to $5.50; eBay lost nearly 80% of its value; Cisco, almost 90%; and Yahoo, over 90%. The Nasdaq didn’t regain its 2000 peak until 2015. The recovery was slow, but damage was mostly limited to equities and the tech sector. No major banks failed, credit markets continued to function, and companies with real earnings eventually climbed to new highs.

That makes the 2008 housing crash more frightening. Markets recovered faster, but the damage to jobs, credit and household wealth ran deep. To ward off deflationary risks following the dot-com bust, in June 2003, under Chairman Alan Greenspan’s leadership, the Fed slashed its benchmark funds rate to a 46-year low of 1.00%. Combined with a flood of subprime lending, this helped launch a speculative boom that saw house prices more than double from 1997 to 2006. The steepest gains came in the final three years before the crash.

In an attempt to cool the economy, Greenspan gradually increased rates in quarter-point steps beginning in June 2004. The end of the rate-climbing cycle occurred in June 2006 at 5.25% under his successor, Ben Bernanke. As cheap credit disappeared, the housing boom began to crack.

Falling home prices exposed a mountain of risky mortgages. By 2007, subprime lenders began failing. Delinquencies, defaults and foreclosures soared, and panic spread across the financial system. Losses spread from mortgages to mortgage-backed securities, and then across a credit system wired together by leverage, short-term borrowing, and circular counterparty exposure.

The collapse of Lehman Brothers on Sept. 15, 2008, drove a final stake through the housing market. Holding a portfolio of real-estate assets that relied heavily on short-term funding, Lehman’s liquidity position was squeezed as home prices fell, defaults soared, and lenders demanded more collateral. When potential buyers backed away, and the government refused a bailout, Lehman filed for bankruptcy—the largest in U.S. history. This shattered confidence and triggered a chain reaction that turned the housing bust into a full-blown financial crisis. It soon spread from Wall Street to Main Street.

As losses mounted, investors questioned securities previously judged safe. Financial institutions stopped lending, and credit markets seized up. Household wealth dropped some 20%, erasing around $13 trillion from its early 2007 peak. Nearly 9 million jobs vanished, and unemployment hit a record 10%. In contrast, the 2000 dot-com crash cost households $2–3 trillion, led to roughly 2 million job losses, and unemployment peaked at just over 6%. It took until 2012 for household wealth to return to its 2007 high, until 2013 for the S&P 500 to reclaim its peak, and not until 2017 for unemployment to return to its pre-crisis level.

The distinction between the two bubbles is critical. A valuation bubble like the dot-com bubble can self-correct, brutally but without lasting institutional damage. A leverage bubble fueled by hidden debt and counterparty exposures spreads through the financial system like wildfire. The housing crash left a complex web of losses that propagated through banks, shadow banks, mortgage-backed securities, and derivatives.

The past is not prologue, but it often rhymes. Today’s AI buildout has echoes of the dot-com era, a transformative technology combined with enormous optimism, and investment racing ahead of profitability. Consider OpenAI. It earned about $13 billion in revenue in 2025 and targets roughly $600 billion in cumulative spending through 2030. While not proof of a bubble, it illustrates the scale of capital being committed before long-run returns are fully understood.

Monster gaps between AI spending and revenues will eventually reveal which companies created sufficient value to justify their investments. Some investments will prove enormously productive. Others could become stranded; the AI equivalent of fiber-optic networks laid during the telecom boom that sat dark for years before broadband finally caught up. More concerning is that the rapid innovation in chips risks an equally rapid obsolescence of data centers.

The most frightening issue is how today’s AI boom is haunted by 2008-style risks: excessive leverage, short-term financing, opaque counterparty guarantees, and circular transactions that make demand look stronger than it is. The Fed recently flagged risks coming from debt-financed AI investments, private credit, and highly leveraged nonbank institutions. It sees hedge-fund leverage at near-all-time highs and found that bank credit commitments to nonbank financial institutions last year reached $2.6 trillion. It also fears that a sharp repricing of AI assets might transmit losses through credit and liquidity exposures into the broader financial system.

This raises ominous parallels with the financial crisis and the “Great Recession” that followed. Economist Scott Sumner and others suspect that what turned the housing bust into a historic recession was tight monetary policy, which contributed to a collapse in nominal spending. In Sumner’s view, the economy initially absorbed the housing downturn, but a severe recession took hold as spending contracted across the economy. If that lesson applies to AI, then the policy challenge would be not to shield investors from losses, but to keep a collapse in aggregate spending from turning an AI asset-price correction into an economy-wide recession.

A shock that hits a market with sound balance sheets and modest leverage is painful but manageable. One amplified by debt, hidden exposures and circular financing is frightening. With AI, the key question is not simply whether valuations are too high. It is how much leverage and hidden financial connections lurk underneath. Markets can survive a graveyard of failed companies. Recovery is harder if no one knows where the bodies are buried.