Industry Spotlight: Data Centers — When the Customer Becomes the Creditor

Normally, a customer buys something and a vendor delivers it. In data centers right now, that relationship has started running in both directions at once: the companies buying AI infrastructure are also the ones financing it, insuring it, and in some cases owning a piece of the company selling it to them. Read one filing and this looks like an interesting footnote. Read 21 companies in the same window, and it's the defining structure of the entire buildout.

Oracle's Customers Are Also Its Bankers

Oracle's backlog of signed-but-not-yet-delivered business — what's called RPO, or "remaining performance obligations": contracts a company has already won but hasn't recognized as revenue yet — reached $638 billion this quarter, up 363% from a year ago. $67 billion of that came from AI infrastructure contracts signed in a single quarter alone, and most of it was either prepaid upfront or structured so the customer supplies its own hardware.

Put plainly: Oracle now has $75 billion in contracts where the customer paid before the infrastructure existed, or brought the hardware itself. That's not a normal vendor relationship — it's the customer funding construction of the thing it's about to rent.

Alphabet Is Underwriting Data Centers It Doesn't Own

Alphabet went a step further this cycle. It committed $85.2 billion in future data center lease payments, and it started backstopping — essentially co-signing — infrastructure built by other companies, with exposure that can run up to 15 years. In its own words:

Alphabet accounts for these arrangements as credit derivatives — financial exposure to whether a third party's data center gets built and pays its debts, not a purchase Alphabet is making for itself.

Alphabet isn't just a tenant anymore. On some of these deals, it's closer to the bank that guarantees the landlord's construction loan.

Amazon Tied a Vendor's Stock Price to Its Own Spending

Amazon's backup-power deal with Generac is the clearest version of this pattern, because the mechanics are simple once you see them. Amazon committed to buy up to $8 billion in generators, with $2.4 billion of that landing in 2027–2028. But the deal isn't structured as a normal purchase order — it's built around a warrant, meaning Generac gave Amazon the right to receive Generac stock, and how much stock Amazon actually gets depends on how much Amazon itself ends up spending.

Generac's equity is now directly tied to how aggressively its own customer builds out. If Amazon's data center spending slows down, Generac's stock deal shrinks with it. The vendor's upside and the customer's capex plan are the same number.

The Money Flows Back the Other Way, Too

This isn't a one-way street from customer to vendor. NVIDIA — the chip supplier every AI cloud company depends on — put $2 billion directly into CoreWeave's stock. Jane Street separately took a $1 billion stake in the same company. CoreWeave, meanwhile, has its own side of the ledger: it just closed a $4.2 billion bond deal, bringing its total number of separate debt agreements to at least eight, several of which are linked — if CoreWeave stumbles badly enough to trigger one, it can trigger the others at the same time.

So the loop closes at multiple points: the chipmaker owns equity in the cloud provider, the cloud provider's own debt is stacked and interlinked, and the whole structure depends on the same AI demand holding up everywhere at once.

The One Number That Would Break the Story

None of this is described as a warning sign in the report — it's described as a reason demand is stickier than it looks, because a customer who has already prepaid or co-signed has more to lose by walking away than a customer on a normal month-to-month contract. But it comes with a specific, checkable condition attached, not just a general sense of risk:

Only 12% of Oracle's $638 billion backlog is expected to convert into actual revenue in the next 12 months, and Oracle's own profit margin on this business is still guided to get worse next year before it gets better.

That's the test. If Oracle's margins start recovering on schedule, the financing structure worked as designed — customers funded growth, and growth is now paying them back. If margins keep sliding past the point management has guided, the prepayments and warrants and cross-default clauses don't disappear; they just start pointing the other way, turning today's "sticky demand" into tomorrow's shared exposure across everyone in the chain.

None of these five companies would tell you this story on its own. Oracle's filing describes a backlog. Alphabet's describes a lease. Amazon's describes a supply agreement. Only reading all of them side by side, the same quarter, turns three separate disclosures into one structure — and one falsifiable bet on whether it holds.