Oracle's AI demand now comes with a financing schedule. The paper's June 13 business major said Oracle had turned AI demand into a financing test. Sunday's version is more exact: Oracle's own release reports $638 billion in remaining performance obligations, large AI contracts with $75 billion of prepaid or customer-supplied hardware, negative fiscal-year free cash flow, and roughly $40 billion of expected fiscal 2027 financing. [1]

That is the useful receipt. It does not settle whether Oracle is the AI-cloud winner or the capex bubble's cleanest warning. It tells the reader what has to happen next. Demand must become data-center capacity. Prepayment and customer hardware must become lower risk rather than customer leverage. Debt and equity financing must arrive on terms that do not turn growth into strain. [1]

CNBC's March account supplied the caution before the new figures did. Oracle, it argued, was building yesterday's data centers with tomorrow's debt, because concrete, power, chips, and leases move on slower schedules than the revenue excitement around AI capacity. [2] Oracle's release now gives that caution numbers large enough to test every quarter. [1]

The X frame calls this either bubble or proof. Both miss the schedule. RPO is not cash. Debt is not doom. Customer-supplied hardware is not free money. The arrangement may be the way hyperscale AI capacity gets financed, or it may be the place where one customer set gains enough leverage to make the supplier's backlog look safer than it is.

Oracle has a strong story. The company says cloud infrastructure and cloud applications drove record results, and the size of its RPO shows demand that most software companies would envy. [1] The fresh X reaction that matters is not fanfare; it is investors putting the $638 billion backlog beside cloud growth, capex, and cash burn.

The hard question is concentration. If a huge share of the $638 billion obligation depends on a small group of AI customers, the demand is real but narrow. Narrow demand can still build an empire. It can also transfer bargaining power to the buyers who prepay, supply hardware, or define the schedule.

Financing is not a footnote here. Oracle says it raised $43 billion of debt and $5 billion of equity in fiscal 2026 and expects about $40 billion of debt and equity financing in fiscal 2027. [1] That means the AI story has left the keynote hall. It now belongs to lenders, underwriters, rating agencies, power utilities, equipment suppliers, and customers who need capacity on time.

CNBC's earlier debt frame matters because it names the time mismatch. [2] The market can price future AI demand in an afternoon. A data center needs sites, transformers, permits, cooling, fiber, workers, chips, and interconnection. A model can go viral faster than a substation can be built.

That mismatch is why prepaid hardware and customer commitments deserve their own line in the story. Oracle's release says some large AI contracts include $75 billion of prepaid and customer-supplied hardware. [1] That can reduce the amount Oracle must finance alone, but it also turns the customer relationship into infrastructure co-production. The customer is no longer merely buying capacity. It is helping shape the build.

The risk is not only financial. A schedule this large touches local power grids, construction labor, chip allocation, optical networking, water, and permitting. CNBC's debt story is really a story about physical lead times disguised as finance. [2] If any piece slips, the RPO number still sits on the page while the capacity to satisfy it waits for the real world.

Prepaid and customer-supplied hardware reduce one risk and create another. If customers bring money and equipment, Oracle can argue the commitments are serious. If customers become indispensable, Oracle may be financing a buildout whose economics depend on the terms of a few powerful counterparties. The same fact can be comfort and warning.

That is why the next receipt is not another RPO headline. It is financing terms, customer concentration, margin recovery, capex timing, utilization, and evidence that prepaid commitments become durable cash. Until those arrive, Oracle should be covered neither as miracle nor folly. It is an infrastructure bank for AI demand, and the repayment schedule has become the story.

-- THEO KAPLAN, San Francisco