AI Demand Is Proven. Now the $70 Billion Bill Comes Due

Capex Outgrew Revenue

Oracle reported fiscal Q1 2027 revenue of $19.3 billion on September 10, up 30% from a year earlier. Cloud revenue rose 62% to $11.6 billion, and within that, Oracle Cloud Infrastructure (OCI) revenue jumped 121% to $7.4 billion. The company brought 850 megawatts of new data center capacity online in the quarter and signed more than $30 billion in additional AI cloud contracts.

But the more striking number in this earnings report was not on the income statement. It was on the cash flow statement. Capital expenditure for the same quarter hit $28.499 billion, roughly 147% of quarterly revenue. That is about 3.4 times the $8.5 billion Oracle spent on capex a year earlier.

That combination captures the AI infrastructure industry in miniature. Cash goes out first for GPUs, networking gear, power, cooling, land, substations and buildings. Customer usage comes back as revenue and cash flow only over the following years. Building AI data centers is, at its core, a capital-intensive business with a long lag between spending and payoff.

A $664 Billion Backlog Is Not $664 Billion in Cash

Oracle's remaining performance obligations, or RPO, grew to $664 billion. RPO is money already under contract but not yet recognized as revenue. It is a strong signal of demand visibility, but it is not the same thing as cash or profit sitting in the bank today.

Long-term investors need to watch two numbers side by side. RPO shows how visible future revenue is. Free cash flow shows how much cash the company is spending today to build that future. Oracle's free cash flow was negative $23.7 billion for fiscal 2026, and negative $5.4 billion in the first quarter of fiscal 2027 alone.

That demand for AI exists and that AI infrastructure spending generates strong returns are two separate claims.

This is where the question for the AI investment cycle shifts. In 2024 and 2025, the market kept asking whether customers would actually use this much GPU capacity. Oracle's current numbers strongly confirm that contracted demand from very large customers is real. The question that remains is how much capital that demand requires, how quickly the capacity gets utilized, and at what margin it gets paid back.

Act Two of the AI Investment Cycle

Oracle is likely to be the first company to show what this third stage of stress testing looks like in practice. Microsoft, Alphabet and Amazon are massive platforms that generate enormous cash from existing businesses. Oracle is attempting hyperscaler-scale AI infrastructure expansion from a considerably smaller cash-generating base.

Even so, it would be an overstatement to say Oracle is funding all $95 billion of this spending out of its own pocket. The company's reported fiscal 2027 capex plan runs roughly $90 billion to $95 billion, but Oracle's net cash capex burden was described as about $70 billion. The remaining $20 billion to $25 billion is offset through customer prepayments and similar arrangements. In this very quarter, $11.36 billion in customer prepayments reduced the capex burden.

At the end of fiscal 2026, $75 billion of Oracle's large AI contracts involved customers either prepaying or directly supplying GPUs. Oracle is not simply building data centers off its own balance sheet. It is actively using contract structures that share capital risk with large customers.

The Risk Does Not Disappear

Customer prepayments reduce financing risk, but they do not eliminate the return-on-investment question. If data centers do not come online as planned, if customer AI demand growth slows, or if faster GPU generational turnover shortens the economic life of these assets, returns could come in lower than expected.

There is also a clear way to prove the current spending is paying off. OCI revenue needs to sustain high growth for a long stretch beyond its current triple-digit pace, utilization on newly installed power and GPU capacity needs to climb quickly, and operating cash flow needs to catch up with rising capex so that free cash flow turns positive.

This quarter offered some encouraging signs. Operating cash flow surged to $23.1 billion, and as a result, despite $28.5 billion in capex, the free cash flow deficit was limited to $5.4 billion. Oracle's after-hours share price jump was not driven by the size of RPO alone. It reflected that the gap between growth and cash burn was less severe than expected.

What to Watch

Whether OCI can sustain its current pace of growth over multiple years, whether utilization rates on newly built power and GPU capacity climb fast enough to justify the spending, and whether operating cash flow keeps closing the gap with capex so free cash flow eventually turns positive.

FAQ

Is capex exceeding revenue a red flag on its own? Not by itself. Large infrastructure businesses can run capex ahead of revenue during their build-out phase. The real question is whether utilization, margins and cash flow catch up afterward. Oracle is now being tested on that capacity to recover the spending.

Is a $664 billion RPO the same as guaranteed cash? No. Visibility is high, but it is not the same as cash in hand. These contracts can be recognized over several years, and actual profitability depends on data center construction costs, power costs, GPU costs, depreciation, and terms that vary by customer.

Why does this matter more for Oracle than for Microsoft or Amazon? Oracle is pursuing very large AI infrastructure investment from a relatively smaller existing cash-generating base than those companies. That means the same increase in capex can show up as pressure on the balance sheet and free cash flow more quickly.

What explains the positive market reaction this quarter? Not RPO alone, but the $23.1 billion in operating cash flow. The fact that the quarterly free cash flow deficit was contained to $5.4 billion despite massive capex eased near-term concerns about whether the capital would come back.

What would make this spending look clearly worthwhile? If AI demand keeps growing faster than expected, if the share of customer prepayments and direct GPU supply keeps rising, and if new data centers reach high utilization and margins quickly, the capital intensity that concerns the market now could look far less alarming in hindsight.

Insight Times Editorial Desk