The artificial intelligence boom could end in recession. Economists at Moody’s Analytics now put the odds at 45 percent.
The warning comes as a separate report from Moody’s Ratings exposes how the country’s biggest tech companies have buried more than half a trillion dollars in future obligations that don’t appear on their financial statements.
Moody’s Analytics laid out the recession scenario in stark terms. The economists said AI companies have taken on dangerous debt loads using financing methods that lack transparency. This assessment gained credibility when Moody’s Ratings uncovered exactly how opaque the situation has become, finding $662 billion in off-balance-sheet commitments from hyperscalers.
The Analytics division sees two main paths to recession.
One involves a stock market crash triggered by inflated expectations around AI technology. Investors have poured money into AI stocks based on promises of future returns. If those returns don’t materialize, a sharp correction could wipe out trillions in market value.
The other path centers on automation eliminating jobs faster than workers can find new employment. If this job displacement happens too quickly, the economy won’t have time to adjust. Mass unemployment could drag down consumer spending and economic growth.
This hidden debt amounts to 113 percent of what these five companies currently report as their adjusted debt. The total lease commitments, including ones already on the books, reach $969 billion. More than two-thirds of that figure stays invisible to investors looking at standard financial reports.
See also Google users push for $2.36B compensation after privacy class action verdict
Accounting loopholes keep massive obligations off the books
David Gonzales works as an accounting analyst at Moody’s Ratings. He said the companies haven’t dodged any requirements through creative bookkeeping. The obligations just haven’t triggered yet because the services haven’t been delivered. But they will be.
Look at Alphabet’s financial disclosures to see how fast these numbers grow. In the second quarter of 2025, the company reported future lease payments of $23.9 billion for data centers not yet on its balance sheet. By the third quarter, that figure jumped to $42.6 billion. The leases will start between 2025 and 2031. Terms run anywhere from one year to 25 years.
The unusual accounting comes from how AI equipment differs from traditional technology. Standard data center leases used to last 10 to 15 years. But the specialized chips and hardware needed for artificial intelligence wear out in just four to six years. Tech companies now want shorter initial lease terms with options to renew later.
Accounting rules date back to the 1930s. Under these rules, companies only report lease renewals if they’re reasonably certain to happen. That means more than 70 percent sure. Nobody can predict AI technology needs years ahead. The firms argue they can’t be reasonably certain about renewals. This keeps those costs off their books.
Meta’s $28 billion guarantee stays hidden from investors
Property owners still need guarantees before building multibillion-dollar facilities. The solution uses something called residual value guarantees. If a tech company walks away from a lease, it pays the landlord the difference if the data center’s market value drops below an agreed amount.
See also Senators propose US law to limit AI and offshore call centers to protect American jobs and consumers
Current rules let companies avoid reporting these guarantees unless it’s probable they’ll pay. Meta Platforms entered leases starting in 2029 worth about $12.3 billion. The company also provided a residual value guarantee with a threshold of $28 billion. Meta decided payout wasn’t probable. Nothing shows up on its balance sheet for that $28 billion promise.
Apollo Global Management’s analysis shows hyperscaler capital expenditure reaching around 2% of GDP in 2026. Source: Apollo
Apollo Global Management tried showing the scale of this spending. Total capital expenditure on data centers hits roughly $646 billion. That’s about 2 percent of the country’s entire economic output. It matches the combined economies of Singapore, Sweden, and Argentina. Defense spending in 2025, for comparison, was around $917 billion.
Alastair Drake, another analyst at Moody’s Ratings, worked with Gonzales on calculating the unrecorded obligations. The two accounting analysts determined that the $662 billion figure represents a massive financial overhang that will eventually land on corporate balance sheets as the leases commence over the next several years.
If AI investments don’t pay off as expected, these companies could face a cash crunch just as the hidden lease obligations come due. That could force cutbacks, layoffs, or fire sales that ripple through the tech sector and beyond.
