Oracle's $638 Billion AI Backlog Is Both a Demand Signal and a Leverage Risk

The irony sits right on the surface. Oracle has more demand than it can sign up, let alone deliver — a reported backlog of roughly $638 billion in contracted and awarded cloud and AI work — while revenue grows at a double-digit pace. And the stock is down about 47% over the past year. It is confusing on purpose, and worth untangling, because the same number is doing two jobs at once: it is the strongest evidence that Oracle's AI bet is real, and it is the reason the balance sheet scares people. The backlog is simultaneously a demand signal and a leverage risk.

The $638 billion is a promise, not cash

First, what a "backlog" actually is. Oracle's number includes remaining performance obligations — future revenue already booked under contract and recognized as the services are delivered, not as they're signed. What starts as a contract stretches into a multi-year revenue schedule. Recognize a $100 billion AI deal across, say, five years and roughly $20 billion shows up each year, subject to how quickly the compute is actually switched on. And part of that $638 billion is "awarded but not yet contracted" — softer than a signed deal, and cancellable in principle. So treat it as a queue of future business, not cash in hand.

The buildout is where the risk lives

Meeting that schedule means building data centers at a brutal pace, and the cash math is where the tension becomes visible. Over the trailing year, Oracle spent roughly $55.7 billion on capital expenditure while generating around $32 billion in operating cash flow — leaving free cash flow at roughly negative $24 billion. It is funding a cash-negative buildout with borrowed money: total debt of about $219 billion, net debt near $98 billion, and a debt-to-equity ratio of roughly 3. Every incremental gigawatt in that backlog has to be pre-funded with borrowed capital, and the revenue only arrives later, spread thin.

That is why the stock's normal yardsticks look misleading. The stock traded at roughly 27 times trailing earnings and about 37 times forward earnings — but those figures tax a company that is reinvesting every dollar of cash into building capacity. When a good chunk of the depreciation hasn't hit yet and cash flow is negative, an EV/EBITDA of roughly 19 times is the closer reading of what the market is actually paying. Investors are absorbing today's cash burn to buy tomorrow's cash flow.

Same number, two readings

This is the supply-commitment trap in miniature. A surge in commitments reads as demand strength and as rising leverage and delivery risk at the same time — and Oracle feels both. The backlog is also concentrated: OpenAI and Microsoft anchor a huge share of it. That makes it the strongest demand signal in the industry, and also the part that creates the financing, delivery, and counterparty risk, because a renegotiation or a utilization slump at the biggest customer would land hardest on a book that leans on just a couple of names.

The question for an investor is not whether AI demand is real — the backlog says it is, and the falling stock is not a verdict on demand. The question is whether the near-term return curve, burdened by negative free cash flow and rising leverage, still justifies the position versus putting the money elsewhere. Oracle has a real asset in that queue, and the market has already marked down the risk. What would close the gap is conversion — capex turning into cash, and the biggest commitments turning into recognized revenue — not more signings.