Oracle Fell 50% While Its Backlog Grew 363%: Reading the $36B Cash Flow Swing

The stock got cut in half while the business got bigger

Here is my read up front: what broke at Oracle is not the business, it is the timing of cash. Revenue grew 17% to $67 billion, and the backlog of signed contracts grew 363% in a single year to $638 billion. What cut the stock in half is the money that has to leave the building before any of that growth shows up.

Last September, Oracle was one of the hottest stocks on the market. It had more than doubled in a year to around $279 a share. Today it trades at about half of that.

A drawdown that size usually means something is broken. So I went back through the statements line by line looking for the break. I could not find it.

What Oracle actually sells in this cycle

Oracle rents out raw computing power — the kind AI labs like OpenAI use to train frontier models. Through its cloud arm, OCI, it builds the data centers and leases the GPU capacity.

Its edge here is not technical. It is positional. Oracle is neutral.

There are companies that need enormous compute but do not want to rent it from a direct competitor like Amazon or Google, because that means running their workloads on a rival's infrastructure. Oracle is what is left. On top of that, Oracle engineered its cloud specifically for AI training rather than general-purpose workloads, which lets the same training run finish faster and cheaper than it would on the older clouds. I think that combination — neutrality plus purpose-built design — is a more durable moat than the market currently credits.

The number that actually spooked investors

The problem is what those data centers cost to build.

Oracle's capital expenditures exploded more than 1,100% in four years to nearly $56 billion. That pushed free cash flow from a positive $12 billion to almost $24 billion in the red — a $36 billion swing.

In isolation, that is a frightening chart. A company that printed cash now burns it. I take that seriously; it is exactly the pattern that should make you slow down. I wrote about the risk that this era's AI capex never earns its return in the AI capex cliff and concentration risk.

But this spending is against inventory that is already sold

Here is the part the headline number hides. Oracle is not building on hope. It is building to fulfill contracts that are already signed.

The backlog grew 363% in a single year to $638 billion. That is nearly ten times everything the company sold last year across every business line. The capex is closer to a delivery obligation than a bet.

Once that backlog starts converting, the shape changes. Revenue can go from $67 billion today to roughly $140 billion within three years. And as the build-out passes its peak, free cash flow has a real path back to positive over the next two years.

The market is pricing the worst quarter of the spending cycle and giving almost no credit to what the spending is for. Anyone buying here is underwriting the lag, not the survival of the business.

Where I think the real risk sits

The thesis does come with conditions.

First, customer concentration. A large backlog is only as good as who signed it. If a meaningful share sits with a handful of AI labs, then those labs' ability to keep raising capital effectively becomes Oracle's revenue line. This is the most underpriced risk in the story right now, and it gets far less attention than the capex number does.

Second, leverage. Negative $24 billion of free cash flow has to be funded somewhere. If funding costs rise, the effective margin on that backlog gets shaved.

Third, the lag itself. The stock can keep falling during the window between signing and recognizing. Calling something undervalued is not the same as calling a bottom.

Even so, I think Oracle is genuinely undervalued here. The $638 billion is a contract number, not a forecast, and the market is currently handing it over for close to free. For the broader version of this business model, see AI infrastructure as a tollbooth.

FAQ

Q: Is it ever safe to buy a company with negative free cash flow? A: It depends entirely on what is causing it. Cash leaking out of a deteriorating business is a warning. Cash going out ahead of contracts that are already signed is a different animal. For Oracle, the $638 billion backlog is what justifies the spend.

Q: Doesn't a 50% drawdown mean the market knows something? A: What the market usually prices first is timing, not direction. Revenue is still growing 17% and the backlog is up 363%. What is in the price is uncertainty about how long the negative cash flow window lasts, not a collapse in demand.

Q: When does cash flow normalize? A: On the current build cycle, free cash flow has a credible path back to positive within about two years. The caveat is that if bookings keep accelerating, the capex peak gets pushed out with them.

Q: Does Oracle need to beat AWS or Google Cloud for this to work? A: No, and that is the point of the neutrality argument. Oracle does not need to win general-purpose cloud. It needs to remain the credible non-competitor option for labs that will not run on a rival's stack.