AI Datacenters Are Piling Up Debt. Is a Crisis Coming?

Key points

• Companies including Meta, Oracle, xAI and CoreWeave are borrowing billions of dollars to build AI datacenters at a rapid pace.
• Critics say some of that debt does not appear on company balance sheets, the official financial documents that show what a company owes.
• Analysts comparing this to the Enron accounting scandal, where hidden debts caused a sudden corporate collapse in 2001, are drawing the wrong parallel.
• The debt is real, but the structure is different from past crises, and the companies involved have genuine assets backing the borrowing.

Something big is being built underground, and most people have no idea it is happening.

Across the United States and Europe, technology companies are pouring concrete, running fibre cables and installing rows of specialised computing hardware to support the surging demand for artificial intelligence. Meta, Oracle, xAI (Elon Musk's AI company) and CoreWeave (a cloud computing firm that rents out AI hardware) are among the biggest spenders. The price tags run into the tens of billions of dollars.

That money has to come from somewhere. Much of it is borrowed.

So what is the actual worry?

Some financial analysts are calling it a potential "debt bomb." Their concern is specific: a chunk of this borrowing does not show up clearly on the companies' balance sheets. A balance sheet is simply a snapshot of what a business owns and what it owes. If debts are kept off that snapshot, lenders, shareholders and regulators can miss how stretched a company really is.

The comparison flying around financial circles is to Enron. That US energy company famously used off-book financial vehicles to hide billions in losses. When the truth emerged in 2001, the company collapsed almost overnight, wiping out employees' pensions and shaking global markets.

It is an alarming comparison. It is also, most analysts argue, the wrong one.

Why this is not Enron 2.0

The short answer: Enron hid liabilities that had nothing real behind them. Today's datacenter debt is backed by physical infrastructure that holds genuine value.

When CoreWeave borrows money to build a datacenter, that facility exists. It contains racks of GPUs, the specialised chips that do the heavy number-crunching AI needs. Those chips can be resold. The building can be repurposed or leased. Enron's hidden debts, by contrast, were largely built on fictional accounting entries with nothing solid underneath.

The Guardian AI first flagged the Enron comparison in recent coverage, while also noting that context matters here.

There is a second difference. AI datacenter builders are not hiding losses. They are using a common financing structure called off-balance-sheet leasing, where a company signs a long-term deal to use a facility without technically owning it. Accountants and regulators have known about this practice for decades. The rules around it have tightened significantly since Enron's era, especially after the 2002 Sarbanes-Oxley Act, which forced companies to disclose far more about off-book arrangements.

What should ordinary people take from this?

If you hold shares in any of these companies, or in funds that do, it is worth reading their annual reports with this question in mind: where is the debt, and what backs it? That is always good practice.

If you are simply a customer using AI tools, this debate does not change your day-to-day experience right now. The risk is a financial one, not a service outage.

The broader point is this: large bets on new technology often look reckless from the outside. Some of those bets fail. The datacentre spending wave could overshoot real demand for AI computing power, and some lenders could take losses. That is a legitimate concern worth watching. A sudden Enron-style implosion, driven by outright fraud rather than market misjudgement, is a different and much less likely story.

Common questions

What does "off-balance-sheet debt" actually mean?

A company takes on a financial obligation, such as a long-term lease on a building, but structures the deal so it does not appear as a liability on the official accounts. It is a legal practice, but it can make a company look less indebted than it really is.

Could AI datacenter spending still cause a financial crisis, even without fraud?

Possibly, at a smaller scale. If demand for AI services does not keep growing, some datacenters will sit underused and lenders could face losses. That would be painful for investors, but it would look more like a sector correction than a systemic collapse.

Who regulates this kind of borrowing?

In the United States, the Securities and Exchange Commission oversees public company disclosures. Accounting standards bodies set the rules for what must appear on a balance sheet. Since 2016, tighter international rules have required companies to put most long-term leases back onto their balance sheets, closing one of the older loopholes.