Is the AI Boom Building to a Crash?

Market Insights

Is the AI Boom Building to a Crash?

The U.S. is not in a 2008-style crisis, and the ingredients that caused that one, reckless mortgages and thinly capitalized banks, are largely absent. But a different set of warning signs is now flashing around the artificial intelligence build-out: fast-rising debt, soaring asset prices, opaque financing, and many companies lending to each other through complex webs. Harvard economist Samuel Hanson recently described the situation as lights "flashing orange." That isn't a prediction of collapse, but it warns us to understand the risks.

What made 2008 a crisis

A recession is painful, but not as challenging as a genuine financial crisis. When the credit structure breaks, even healthy borrowers can't get loans. In 2007–09, years of risky mortgage lending, packaged into complex securities and held with heavy leverage, turned a housing downturn into a global banking panic.

One truism that emerged from that experience is that crises are harder to predict than to explain afterward, but that may not be accurate. Harvard’s Hanson and his co-authors studied 42 countries over more than six decades. They found a consistent, recognizable danger signal: borrowing and asset prices in the same sector both rising rapidly over three years. When both happen together, the odds of a crisis in the following three years jump to roughly 40%, versus about 7% in normal times.

The pre-2008 housing market met both conditions. Housing today has major concerns, but not that. The question is, does AI?

The flashing alarm signals of AI

Historic spending

Columbia economist Stijn Van Nieuwerburgh estimates U.S. investment in data centers, chips, networking, and power will reach $10.3 trillion from 2025 to 2032. That works out to about 3.6% of GDP a year. By his math, it is bigger relative to the economy than the railroad boom of 1870–90 (2.2%), the interstate highway build-out (1.1%), or the late-1990s telecom wave (1.1%).

Reminiscent of the housing boom

The closest comparison this century is the one before 2008. Residential investment peaked at 6.6% of GDP in 2005, nearly twice AI's projected share. But housing started from a large base. Measured by how much each boom grew, AI is bigger: Apollo chief economist Torsten Slok calculates data center spending will rise 2.5 percentage points of GDP from 2023 to 2027, versus 2.2 points for housing from the mid-1990s to 2005. AI spending is also climbing nearly twice as fast. That matters because booms can reverse just as quickly. Residential investment fell from 6.2% of GDP in early 2006 to 3% by the end of 2008, which made the recession far more severe.

Rapid borrowing acceleration 

Morgan Stanley expects AI-related debt issuance to reach about $570 billion in 2026, roughly double last year. Goldman Sachs noted that hyperscaler bond sales in the first months of 2026 had already passed all of 2025. The IMF says these companies now account for about 10% of global investment-grade bond issuance, up from less than 1% in 2024.

In October, research firm Gartner estimated that data center owners would need $8.5 trillion in revenue through 2030 to cover spending, financing costs, and a 12% return. It forecasts AI revenue tied to data centers of no more than $3.6 trillion over that period.

Interwoven spending webs

A Bank for International Settlements report found that 55% of investment dollars going to AI firms from 2021 to 2025 came from other AI firms. For example, Nvidia sells chips to CoreWeave, owns part of CoreWeave, and has agreed to buy its unused computing capacity. Such arrangements work well when demand is strong, but can magnify losses when it isn't. CoreWeave's stock fell roughly a third in the two weeks after the IMF's October warning, leaving it down more than half from its 52-week high.

Where the debt is hiding

In 2008, much of the danger sat in structures few people understood. Something similar is happening now, though on a smaller scale. At first, tech giants paid for AI out of pocket. Increasingly, the bills are going to outside lenders, often through joint ventures and leases that keep the debt off the companies' balance sheets.

The BIS calls this "shadow borrowing." A Nikkei analysis estimated that Alphabet, Amazon, Meta, Microsoft and Oracle have about $1.65 trillion in off-balance-sheet AI obligations, more than the roughly $1.35 trillion of debt they report. The best-known example ties back to our recent post on private credit: Meta's $27 billion joint venture with Blue Owl to finance its Hyperion data center campus in Louisiana.

That is how AI risk and private credit risk become the same risk. Private credit funds, insurers, and their lenders are financing the build-out. According to the IMF, bank lending to private equity and private credit firms rose 26% this year. Meanwhile, private credit is already under strain, with record default rates and redemption queues at most large funds. In May, the Federal Reserve called the risk from those redemptions limited and manageable, but warned they could tighten credit for riskier borrowers. Currently, the concern is less about any single lender than about how fast this kind of credit is growing, and how little anyone can see of it.

Why this may not be a repeat of 2008 

In 2026, banks are much stronger. In its May 2026 Financial Stability Report, the Federal Reserve said bank capital remained near historic highs and rated vulnerabilities from household and business borrowing as moderate. Post-2008 rules have forced banks to hold more capital and take on less risk.

However, IMF officials say risks have shifted from banks to nonbank lenders such as private credit funds, hedge funds and insurers, which are less regulated and less transparent. A loss there is less likely to cause a bank run, but harder to see coming. A stock market drop is also generally less dangerous to the system than a wave of loan defaults. Even so, the IMF has urged limits on margin lending and leveraged funds that investors use to amplify their AI bets.

What could tip the scale?

  • Slower than expected AI advances. AI doesn't have to fail to cause damage. It only has to deliver profits more slowly than lenders and shareholders expect. Apollo chief economist Torsten Slok says the burden is on AI companies to keep posting very high growth.

  • Stubborn interest rates. Inflation remains above the Fed's 2% target, and many economists now expect rate cuts to wait until 2027. The IMF notes that long-term borrowing costs are near the top of their historical range. Higher rates reduce the present value of far-off AI profits and make AI debt more expensive to carry.

  • A concentrated stock market. AI-related companies make up nearly half of the S&P 500's value. Risk manager Richard Bookstaber, who warned of system fragility months before the 2008 crisis, says he is more worried now than then. He suggests an AI crash could cut the broad market by 40% and the most exposed stocks by 70%. Hyperscalers including Alphabet and Meta are now issuing shares to fund growth, a strategy that depends on high stock prices.

  • Deflation later on. Today, the build-out is pushing up prices for power, copper and labor. But investor Paul Kedrosky argues that if AI eventually replaces jobs and pushes wages down, falling incomes would make fixed debts harder to repay, from mortgages to car loans.

The IMF's own downside scenario is not a meltdown. A sharp AI correction would leave global GDP about 0.4% below its baseline by 2028, with the U.S. and Asia most exposed.

What to watch

No one can time a crisis, and Hanson won’t speculate on whether one is imminent. He compares the economy to a dry forest: the point is to manage the fire risk, not guess the date of the spark. For investors, a few signals are worth tracking:

  1. AI revenue versus AI spending. Hyperscaler earnings calls show whether cloud and AI revenue are closing the gap with capital spending.

  2. Credit spreads on AI debt. Widening spreads on hyperscaler bonds or data center deals would signal lender anxiety.

  3. Private credit stress. Redemption queues and default rates at large private credit funds are an early read on the nonbank lenders financing the build-out.

  4. Rates and inflation. A Fed that can't cut rates, or long-term yields that keep climbing, puts pressure on everything above.

  5. Concentration in your own portfolio. With AI stocks near half the S&P 500, an index fund is already a large AI bet.

The 2008 crisis grew out of a credit boom that few people saw clearly until it broke. This time, the IMF, the Fed, and the BIS are all pointing at the same place. That attention is itself a safeguard, but only if investors and lenders act on it.

Book a Private Investment Strategy Call With Tom

Choose a Time Here