Oracle's $638 Billion Backlog Is Real. The Stock Halved Anyway — Conversion Is the Clock

Oracle just reported the largest pile of signed business in its history, and the market spent most of the past year treating it like a liability. That contradiction is the whole story. The call making the rounds — that Oracle's $638 billion cloud backlog makes it one of the best-performing AI stocks through 2028 — is not wrong about demand. It is wrong about what the market is actually pricing. The stock has already halved. Whatever the backlog is worth was known while shares fell from above $330 toward $160. The bet investors actually face is different: how much of that giant number becomes cash, at a margin that can pay for what it took to build it, before 2028 runs out.

Start with what $638 billion is, because almost every argument about this stock turns on it. In accounting, a company doesn't count a cloud contract as revenue when it's signed. It books it over time as it delivers the service, and every undelivered portion sits on the balance sheet as "remaining performance obligations" (RPO) — the technical name for the backlog. Oracle ended its fiscal 2026 year with $638 billion of it, up 363% from a year earlier, roughly eight years of revenue at the current run rate. In the most recent quarter alone it signed $67 billion of new AI infrastructure contracts, and its cloud-infrastructure revenue grew 93% year over year.

A backlog this size is not a cash receipt. It is a queue of work that still has to be delivered. Only about 12% of it is expected to convert into recognized revenue over the next twelve months, and over half of it is tied to a single customer. That customer is OpenAI, which signed with Oracle to buy more than $300 billion of computing power over roughly five years — a contract that does not begin delivering revenue until 2027, and which alone requires about 4.5 gigawatts of capacity. So the "best-performing through 2028" headline is really a narrower claim: between now and then, one giant, private, cash-burning customer has to start paying, and Oracle has to fund the construction that makes the delivery possible.

That funding is the wall the market has been walking into. Oracle isn't servicing the backlog with cash it already has; it is building the data centers first and paying later. Capital expenditures surged 162% in fiscal 2026 to roughly $56 billion, the company guided to about $95 billion for the current year, and free cash flow turned negative by almost $24 billion — with the burn expected to continue. To bridge it, Oracle plans to raise about $40 billion of new debt and equity in fiscal 2027, on top of roughly $130 billion of total debt, including a $20 billion share sale that directly dilutes existing holders. The worst week of the selloff, a 19% drop in June that was Oracle's largest since the 2001 dot-com bust, followed exactly these headlines: negative free cash flow, the financing plan, and reports that OpenAI was delaying its IPO.

Now the part that decides the trade, and it is the opposite of what the bullish case assumes. The crowd is already long. About 71% of analysts rate the stock a Buy, the highest reading in fifteen years, so the pessimism in the price is not a shortage of believers — it is a market that looked at contract size and then looked at the balance sheet financing it and chose to discount the whole package. This is not a forecast waiting to happen; it is a disagreement that has been repriced in real time for a year.

So the clock you can actually follow has four ordered links, each falsifiable before 2028:

1. Revenue conversion. Does Oracle's reported cloud-infrastructure revenue keep compounding near ~90% and push total revenue toward management's roughly $90 billion target for fiscal 2027? The street only partially believes that number. If the queue turns into recognized revenue at that rate, the derating was overdone. If growth comes mostly from re-contracting existing capacity to one customer on thin margins, the discount is earned.
2. Cash inflection. The $24 billion free-cash-flow hole has to start narrowing. Watch whether capex approaches a peak and whether operating cash flow catches up to the spending — the moment the build stops outpacing the revenue it produces.
3. Financing discipline. Each $20 billion equity sale is a direct tax on shareholder value. The less Oracle has to issue, the cleaner the story; the more it leans on dilution and debt, the more of the backlog's eventual profit belongs to someone else.
4. Counterparty. The deal that fills more than half the backlog is with a private company that has separately committed over a trillion dollars across Oracle, Microsoft, Broadcom, Nvidia, and others. OpenAI's creditworthiness is now Oracle's credit risk, and no balance-sheet line item can make that go away.

Here is the tripwire, and it is unusually close at hand: Oracle reports fiscal 2027 first-quarter results tomorrow, September 10. The single number that matters is not the backlog balance — that will be large and roughly unchanged. It is whether reported cloud and OCI revenue growth actually reaccelerate, and whether management holds the $90 billion target instead of nudging it. If the growth rate in the quarter is real and the cash-outflow story starts to stabilize, the halved stock is a contrarian flag waving at a crowd that already sold. If the quarter just re-confirms the same contract on the same thin margins, then the selloff was the market reading the fine print, not a mispricing.

The $638 billion number is real. It was also real when the stock was $330 and when it was $114. The bet through 2028 is not whether AI demand holds up — it is whether Oracle converts the biggest contract queue in software history into cash fast enough to outrun the cost of building it, and whether the one private customer at the center of the queue stays solvent to pay. Buy the backlog headline and you've bought that chain. Watch its weakest link, because that is where the verdict comes from first.

The wager in one block: - Claim: Oracle's $638 billion backlog makes it a top AI stock through 2028. - What confirms it: OCI revenue keeps compounding near 90%; FY2027 revenue reaches ~$90B; the FCF gap narrows; little incremental dilution. - What kills it: The OCI growth recontracts to one customer on low margin; capex keeps rising ahead of revenue; OpenAI's payment or IPO slips; equity issuance accelerates. - Next date: Sept 10 earnings — does the reported growth rate reaccelerate and hold the $90B target?