Meta Platforms and Oracle have become the two clearest tests of whether AI capital spending can convert into cash, and both are failing the timing question in opposite directions: one has spent the money, the other has booked the revenue but not collected it.
Oracle's remaining performance obligation — contracted work not yet delivered or recognized as revenue — reached $664 billion at the end of its last reported fiscal quarter, after the company signed more than $30 billion of additional AI cloud contracts, according to company disclosures. Against trailing-12-month sales of roughly $71.8 billion, that backlog equals about 9.2 times annual revenue. Meta, meanwhile, has watched free cash flow fall by more than 90% as quarterly AI compute spending climbed into the tens of billions.
"Restructuring can free up resources; converting that massive backlog into profitable cash flow is what ultimately determines whether Oracle's AI spending creates lasting shareholder value," GuruFocus wrote in its assessment of the company's expanded fiscal-2026 restructuring plan, which now totals roughly $2.8 billion.
The two numbers describe the same problem from different ends of the income statement. Oracle is spending before it earns: free cash flow is running at approximately negative $5.4 billion, and the $2.8 billion restructuring charge equals about 14.5% of its latest $19.35 billion quarterly revenue, covering severance, contract exits and other costs. Management has said customer prepayments and customer-supplied chips can soften the capital burden, but roughly half the $664 billion backlog is expected to convert to revenue only within 36 months — a payback window that stretches past most investors' holding periods.
Concentration sharpens the risk. About $300 billion of that backlog traces to a single customer, OpenAI, the company behind ChatGPT. If OpenAI slows its own buildout, Oracle faces cancellations or a much slower delivery schedule on nearly half its contracted pipeline. Oracle is guiding for revenue growth of roughly 34% this year and earnings growth of about 19%, and the stock trades at roughly 17 times this year's expected earnings — a multiple that already assumes the backlog converts on schedule.
Meta's position is the mirror image. Its spending is not backed by a signed backlog at all; it is funded against advertising revenue that must keep growing to justify the compute. That is why the free cash flow decline matters more than the absolute capex figure: it removes the buffer that historically let Meta absorb a bad ad quarter without cutting investment.
The market has already begun repricing the cost side of the trade. Oracle shares are down roughly 28% year to date and about 57% from their lifetime high, even as its cloud infrastructure business grows. The stock trades 21.12% below a $191.31 GuruFocus value estimate, a gap that reflects disagreement over whether the backlog is an asset or a liability with a long duration.
For investors, the operative metric is no longer capex size but capex payback velocity — dollars of revenue recognized per dollar spent, per quarter. On that measure, Oracle's 36-month conversion window and Meta's collapsed free cash flow both compare unfavorably with peers that are monetizing AI capacity inside 12 to 18 months. Nvidia, Broadcom and the data center supply chain that sells into these buildouts carry the same exposure: if hyperscalers slow orders to protect cash flow, the equipment vendors feel it first.
Oracle's next quarterly report, which will show whether RPO conversion is accelerating or merely accumulating, is the nearest checkpoint. Until that number moves, the $664 billion backlog remains a promise, and Meta's cash flow statement remains the receipt.
This article is for informational purposes only and does not constitute investment advice.