The current AI-related bubble in financial assets increasingly echoes warning signs from the global financial crisis (GFC), precipitated by excesses in residential housing, and the spectacular collapse of Enron, “the world’s leading energy company,” whose executives orchestrated a giant financial fraud. The similarities and differences illustrate yet again that while history doesn’t repeat, it often rhymes.
Enron contained a legitimate business, with hard assets and real revenue. It was undermined by the abuse of off-balance sheet structures that hid mountains of losses and liabilities. Enron’s conspiracy was conducted by executives who knew exactly what they were doing, and who richly compensated Wall Street and their own accountants and lawyers to play along.
The GFC and the current AI bubble are more akin to a conspiracy without a conspirator. Unlike Enron, WorldCom, or Madoff, we cannot point to the man behind the curtain pulling the levers and manipulating the markets. Each actor is reacting to financial incentives and market signals that make perfect sense at the time. Each individual may suspect that all this is going to end badly, but what can one do? In the meantime, they must keep playing their role; indeed, that was what they are paid to do.
Enron executives believed they could continue the charade so long as Enron’s share price, which shares served as collateral for off-balance sheet liabilities, kept going up, or at least held steady. The GFC may have never occurred if US housing markets had maintained their decades long trend of steadily increasing in value. Future Federal Reserve Chairman Ben Bernanke’s market-calming statement in 2005 echoed the expectation. “We’ve never had a decline in house prices on a nationwide basis.”
And the AI bubble won’t burst as long as computational demand, and the price users are willing to pay for it, lives up to the TAM (“total addressable market”) projections being used to justify the trillions of dollars being spent on AI infrastructure.
The CEOs of today’s tech companies find themselves unable to stop the madness, much as the bankers did in the months leading up to the GFC. Stop funding and building, and your competitors will happily take your business. Recall Citigroup’s then CEO, Chuck Prince, speaking in 2007, on the eve of the crisis. “When the music stops, in terms of liquidity, things will be complicated. But as long as the music is playing, you’ve got to get up and dance. We’re still dancing.” Prince stopped dancing and resigned four months later, as Citigroup disclosed billions of losses on mortgage-related securities. Citi became the largest of the too-big-to-fail banks to avoid total collapse only by a massive transfusion of capital (i.e., taxpayer money) from the US government.
Some of the warning signs that we’re standing on similarly dangerous ground include multi-trillion dollar liabilities being housed off-balance sheet in special-purpose vehicles (SPVs), a closed-loop AI economy of vendor financing to customers to buy their semiconductor chips or other products, circular reference valuations and corporate earnings driven not by recurring operating profits but by mark-to-market accounting on investments (a favorite Enron trick), and ongoing stress among the private credit companies that serve as the biggest non-bank lenders to the software and tech industries.
Debt issuance in the US has soared, as corporate treasuries have determined to get what they can before liquidity dries up. AI companies alone raised $1.5 trillion in bond markets in the first eight months of 2026. Yet public debt issuance only scratches the surface. Since banks and bond markets can only absorb so much, the already distressed private credit market is where much of Big Tech’s debt has landed. But Big Tech’s record debts are increasingly not on the balance sheet at all, but in off-balance sheet structures and operating lease liabilities.
One leading business publication (The Wall Street Journal) recently calculated over $3 trillion in off-balance-sheet commitments across nine tech companies, of which $1.2 trillion represents leases obligations yet to commence and $1.9 trillion represents purchase obligations for chips, power and construction. Enron’s hidden liabilities totaled twice its reported debt, while Meta’s disclosed off-balance sheet liabilities are 5:1 its consolidated debt.
Companies like Meta are using off-balance sheet SPVs, mostly financed by third-parties, to build and fund massive data centers. Meta does not consolidate the debt on the theory that the vehicle bears the risk. Yet Meta bears the risk of the future lease obligation. If customer demand and pricing over the next twenty years prove out as projected, Meta’s bet will have paid off. If not, the lease obligations will remain Meta’s, without the revenue to cover it.
The analogy with the GFC emerges. The subprime machine did not fail just because everyone turned a blind eye. It failed because risk was moved into opaque structures, sold to unquestioning buyers reaching for yield, and priced on the assumption that the collateral would continue to appreciate. Swap data centers for houses and private credit funds for CDO desks, and the narrative starts to rhyme. Enron’s vehicles hid losses but were collateralized by Enron’s own stock, an imagined perpetual motion machine that flew like a rocket on market momentum, until the fuel ran out, gravity intervened and Enron’s share price collapsed. Today’s SPVs are intended to hold real assets, with real tenants, generating real cash, but similarly rely on the assumption that customer demand and pricing will always be there to support them.
Big Tech’s record profits do not just represent recurring operating income, but sizable mark-to-market gains on investments in other AI companies. While one-off occurrences, they flow into earnings per share, which the market credits with a price-to-earnings multiple of up to thirty times, as if these gains will go on forever. Alphabet’s “other income,” which mostly represents markups on investments in companies like Anthropic and SpaceX, was 71 percent of pretax profit last quarter. Amazon’s other income was 66 percent of pretax profit, driven by a $53.4 billion gain on Anthropic alone. Combined, these anomalies represent about 15 percent of the S&P 500’s second-quarter earnings. Strip these out and Amazon’s 240 percent earnings growth becomes 17 percent; Alphabet’s 300 percent becomes 23 percent.
The bond market is signaling stress. Yields are rising in part because debt issuance—sovereign and corporate alike—is now outpacing demand. The 30-year Treasury yield touched 5.32 percent, the highest since 2007. The 10-year rose above 4.7 percent, also near a high. The Federal Reserve may be forced to raise rates later this year. In the meantime, the US Treasury has started to repurchase bonds in an attempt to lower yields and push liquidity into the markets, an additional warning sign.
Nvidia backstops its customers’ financing, which in turn book paper gains on stakes in AI labs that are also their customers. The labs’ rising valuations justify the marks. The marks multiply the earnings that support the equity that anchors the credit that funds the data centers the labs will rent. On and on it goes. Every node in the system becomes collateral for other nodes.
When the momentum reverses, nodes will fail everywhere at once. The trigger will not likely be an accounting scandal but rather the cost of money. Lease obligations commence. Debt lands on the balance sheet precisely when refinancing turns expensive. Private valuations stop climbing, the marks reverse, reported earnings collapse, and valuation multiples contract.
When a rocket-like volcanic eruption runs out of fuel, the skyward motion continues briefly, until the forces of gravity exceed the remaining momentum. What happens next is referred to as “column collapse.” That collapse is what generates pyroclastic flows, the nasty part that kills people and ravages property. The eruption doesn’t end at the collapse; gravity redirects energy and material downward and outward. The recent explosion in AI-related investment bears resemblance. As with a live volcano, the best thing for an investor to do in these moments is to be as far away as possible.