xAI Data Center Debt Risk: What Happens if AI Revenue Doesn’t Catch Up With Spending?

AI companies are spending enormous amounts on data centers before they know exactly how much revenue those facilities will eventually generate.

That gap is increasingly being financed with debt.

Over the past year, Alphabet, Amazon, Meta, Microsoft and Oracle have issued roughly $220 billion of bonds as AI infrastructure spending accelerated. Meanwhile, developers such as Vantage Data Centers have borrowed tens of billions more to build the physical campuses behind the boom.

Debt is not automatically a problem. The risk appears when revenue, utilization or cash flow grows more slowly than the borrowing used to build the infrastructure.

Our guide to the $3.6 trillion AI financing boom explains how bonds, private credit, leases and project finance are funding the expansion. The next question is what happens if those projects earn less than expected.

Debt Still Has to Be Paid When AI Demand Slows

A data center can cost billions before it produces meaningful revenue.

Borrowers must still pay interest, refinance maturing debt and cover electricity, maintenance and equipment costs even if customer demand disappoints.

That makes utilization crucial.

A fully occupied AI campus with long-term contracts can generate predictable cash flow. A facility operating well below capacity may struggle to cover financing costs, particularly if it was built when borrowing rates and equipment prices were high.

Oracle provides a useful example of the tension. Its AI-related backlog has climbed to $664 billion, but the company is also spending aggressively and recently reported negative free cash flow of $5.4 billion. Oracle plans to raise about $40 billion this fiscal year while investing heavily in cloud capacity.

The problem is not that Oracle lacks demand. It illustrates how rapidly infrastructure spending can consume cash even when revenue is growing strongly.

AI business typeRevenue riskUtilization riskRefinancing riskHyperscalersAI services may grow slower than capexLow to moderate because capacity can serve multiple workloadsRelatively low due to strong balance sheetsModerate; older GPUs and infrastructure can depreciate quicklyCloud / AI compute providersHeavy dependence on customer demand and contract renewalsHigh if rented GPU capacity sits idleHigh if growth relies on repeated debt issuanceHigh because GPUs may lose value as newer generations arriveData-center developersRevenue depends on tenants signing and stayingHigh when new campuses open before demand is securedHigh because projects often use large long-term debt packagesModerate; buildings retain value, but specialized AI infrastructure can age fastPrivate AI infrastructure firmsOften concentrated in a few large customersHigh if one major customer reduces usageVery high if cash flow is weak or debt matures earlyHigh when loans are backed by GPUs or project assetsMain warning signRevenue growth lagging capexFalling occupancy or compute usageRising interest expense or difficult refinancingFalling resale value of GPUs and specialized equipment

Higher Rates Make Weak Projects More Dangerous

The economics become more difficult when interest rates rise.

A project financed at 5% may be attractive. The same project refinanced at 8% can produce a much weaker return even if its operating performance is unchanged.

That matters because AI borrowers increasingly compete with governments and other corporations for investor capital. Our explainer on AI companies competing with governments for money shows how heavy Treasury issuance can push required returns higher across credit markets.

The sheer volume of AI borrowing is already affecting bond pricing. Reuters noted that enormous new tech bond issues are creating unusual differences in credit spreads even among bonds from the same companies.

The Risk Could Spread Beyond AI Companies

The debt does not sit only on Big Tech balance sheets.

Banks, private-credit funds, infrastructure investors and special-purpose vehicles increasingly finance data centers. The Financial Times estimates that total data-center investment could reach roughly $7 trillion by 2030, with increasingly complex financing structures spreading exposure across the financial system.

Physical assets may also lose value faster than expected if newer chips make older computing infrastructure obsolete.

That does not mean an AI debt crisis is inevitable. Strong demand, long-term contracts and rising AI revenue could justify much of the spending.