Oracle and the AI bubble: debt-funded AI infrastructure

Why does debt change the AI bubble risk for Oracle?

Oracle supplies databases, business applications and cloud infrastructure to large organisations. For this page, the lens is funding. A company that pays for data centres from cash flow can pause when demand softens. A company that pays with borrowed money must still meet interest and repayment dates, whatever happens to demand.

That difference explains why the same market slowdown looks mild for one builder and severe for another. Debt amplifies results in both directions: it lets growth run faster when customers pay, and it narrows the margin for error when they do not. The five steps below turn the idea into a routine. Each step uses a public number, each has a simple reading, and together they take a short session per quarter.

How do you read the contract backlog behind AI infrastructure?

Backlog is the value of signed customer commitments not yet billed. Find it in quarterly results and compare its growth with the growth of debt and capital spending. A backlog that grows faster than the borrowing supports the plan, while the reverse leaves a gap. Use the same dates for all three lines, so growth rates compare fairly across the same quarters.

Read the terms too. Long contracts with firm minimums protect revenue, and contracts that customers can reduce give less comfort. Ask how much of the backlog is firm and when billing starts, because capacity under construction earns nothing until it opens. The gap between signing and billing is where borrowed money works hardest and earns least.

Which customers stand behind Oracle’s AI backlog?

List the largest customers behind AI contracts. If a few model builders hold most of the commitments, the backlog behaves like a bet on their success. Those builders depend on investors, so their ability to pay rests on future funding rounds.

The EQOS Asia AI study records 95% of AI projects failing in production (MIT 2025) and about $200B a year invested (April 2026). Established companies running steady workloads add stability to the mix. A backlog spread across them is easier to trust than one concentrated in a few young firms. Ask management for the share of the backlog held by the 3 largest customers, and follow that share over time.

How do you test whether income covers Oracle’s interest payments?

Divide operating profit by interest expense. The result is the interest cover ratio: it shows how many times profit can pay the interest. A high number leaves room for a bad year. A number that falls toward 1 means almost all profit goes to lenders.

Look at the direction over several quarters, since new borrowing raises interest cost before new capacity earns revenue. A temporary dip during construction is normal. A dip that continues after sites open points to weak demand or heavy cost. Record the ratio with its date every quarter. Two quarters of decline in a row, with new borrowing still arriving, is the pattern to flag for a closer reading.

When do Oracle’s debt payments fall due?

A debt schedule lists amounts and due dates. Long schedules give revenue time to arrive, while many payments due in a short window create pressure. Look in the annual report for the table of maturities and compare each year’s amount with the cash the business generates.

Check the type of debt as well. Fixed-rate debt gives predictable cost. Variable-rate debt moves with the market and can raise payments without warning. Mixed funding with cash, bonds and partner capital spreads the risk, which a single large borrowing does not. When several payments cluster in one year, compare that year with the expected opening dates of new sites.

Which credit signals warn of stress in debt-funded AI building?

Credit ratings and the extra yield investors demand on a company’s bonds react before accounts do. A downgrade or a rise in the extra yield shows lenders asking for more compensation. Both are public and update often, which makes them a quick monthly check.

Treat a signal as a prompt to reread steps 1 to 4, and avoid reading it as a verdict. Markets sometimes overreact and sometimes underreact. If backlog, customers, cover and maturities still look sound, a credit signal alone is a reason to look closer, not to change course. Keep a monthly note of the rating and the extra yield, so you see a trend instead of a single day.

How can you apply the five-step test to your own decisions?

Use the same steps on suppliers, partners and customers who build with debt. Ask for the schedule of commitments, check who funds them, and keep backup options for critical services. A second supplier for a vital workload is cheap insurance when a provider carries heavy debt. The EQOS Asia method page explains how such checks combine into one score.

This page is general information, not investment advice. The test frames risk and leaves prices and dates open. For a decision that depends on a debt-funded supplier, the Λ simulator gives a first score in 10 minutes, and the 30-minute pre-diagnostic applies it to your case.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile