- Bloomberg currently puts unpaid AI data center debt at over $500 billion
- CoreWeave isolates each loan into its own separate special purpose vehicle
- Parent companies report only a fraction of their actual total exposure, hiding the rest in fictitious entities.
A growing number of analysts are now warning that AI data center debt increasingly resembles the subprime mortgages that sparked the 2008 financial crisis.
Much of this debt is issued through special purpose vehicles, structures that keep billions of dollars off company balance sheets.
Bloomberg estimates there is more than $500 billion of AI data center debt outstanding, with about $200 billion held by private credit funds.
A debt structure built on theoretical income
Special purpose vehicles (SPVs) borrow to build data centers, then repay creditors only once paying customers start generating revenue.
This structure is exactly why CoreWeave has raised billions through separate SPVs for individual loans, including an $8.5 billion facility tied to Meta’s contract, since each loan remains isolated within its own entity.
The same logic explains why Nikkei Asia reported that Meta, Google, Amazon, Microsoft and Oracle have accumulated about $1.65 trillion in debt over five years, much of it spread across similar vehicles rather than a single balance sheet.
This gap between actual exposure and reported debt exists because these vehicles are jointly owned with outside investors, allowing the parent company to report only a fraction of the risk.
Meta’s Hyperion data center shows the pattern clearly: it is 80% owned by Blue Owl and only 20% owned by Meta itself, so most of the debt resides on paper with Blue Owl, even though Meta is the intended tenant.
Google used the same approach, backing debt-financed data centers built by Fluidstack, Cipher Mining and TeraWulf without those obligations ever touching its own balance sheet.
This type of arrangement is precisely what caught the attention of auditor Ernst & Young, who flagged Meta’s structure as a critical audit topic, questioning who ultimately bears the economic risk.
The stakes extend well beyond the companies involved, as pension funds and insurers are also directly exposed, with many now relying on data center returns to fund their future payouts.
Echoes of the 2008 mortgage collapse
The comparison with 2008 holds water because both bubbles were based on the same erroneous premise: demand would continue to grow forever and would never need to be tested.
Subprime mortgages were proof of that at the time: By 2006, about 20 percent of all new mortgages issued in the United States were already classified as subprime, according to government data.
Rather than seeing this as a warning sign, financial institutions packaged these loans into complex securities, a move that obscured the true underlying risk for both investors and rating agencies.
The financier Michael Milken perfectly captured the mood of the times when he publicly described these securities as a “financial innovation” with the potential to vastly increase national prosperity and employment.
Reality overtook this optimism when mortgage defaults began to rise sharply in 2005, and from there the damage rippled throughout the financial system.
Lehman Brothers epitomizes how unchecked this confidence has become, operating with more than 25 times leverage in 2005 without serious reaction from regulators or rating agencies.
Today’s numbers echo this same pattern of unstudied risk: Analysts estimate banks’ exposure to private credit at more than $1.4 trillion, including $300 billion held by big banks alone.
Some estimates suggest that planned AI data center capacity exceeds actual annual computing demand by a factor of around 15 times.
Unlike in 2008, this risk is not due to derivatives but to the sheer scale of the construction costs of individual data centers.
Whether this debt is paid off gradually or all at once likely depends on how quickly AI’s main customers can pay their bills.
For now, the scale of the exposure of banks, pensions and insurers suggests that the comparison with 2008 is more than just rhetoric.
Via Ed Zitron
Follow TechRadar on Google News And add us as your favorite source to get our news, reviews and expert opinions in your feeds.
