Oracle Q4 earnings beat, but $40B AI capex plan bites

Oracle (ORCL) reported $2.11 in adjusted earnings per share and $19.2bn of fiscal fourth-quarter revenue, ahead of estimates for $1.96 and $19.1bn, but the stock fell about 5 per cent in late trading on Wednesday after investors looked past the beat and focused on the cost of its AI data-center buildout.

For management, the quarter showed that demand for Oracle Cloud Infrastructure is no longer a side story. Oracle said in its official earnings release that cloud infrastructure revenue rose 93 per cent from a year earlier to $5.8bn, while remaining performance obligations climbed $85bn during the quarter to $638bn.

For shareholders, the same numbers carried a second message. Oracle ended fiscal 2026 with $55.7bn of capital expenditure, $23.7bn of negative free cash flow and a financing plan that could run to $40bn in fiscal 2027, according to Reuters reporting on the results. The earnings story has become a funding story.

This substantially reduces the amount of capital Oracle must raise to build out our AI data centers.
— Oracle, according to CNBC

Customer contributions may soften the cash burden. They did not settle the market’s main concern. Strong AI demand is valuable, but it is not free, and Oracle’s latest numbers show how quickly the infrastructure bill can outrun even a fast-growing cloud unit.

The beat was not enough

Plenty of the quarter looked like a classic earnings win. Revenue beat consensus. Adjusted EPS beat by 15 cents. Management lifted fiscal 2027 profit guidance, and the RPO figure put a large dollar value on future demand for Oracle Cloud Infrastructure.

Single-name earnings stories often turn on the delta between what the income statement says and what the stock says. Here, that gap was unusually wide. Infrastructure revenue is accelerating, but the cash-flow profile attached to that acceleration is deteriorating. Investors were being asked to value a bigger backlog while underwriting a bigger funding need.

Reuters said Oracle’s AI spending blew past estimates and raised worries over debt. Bloomberg reported that data-center costs overshadowed AI growth. CNBC’s account made the same market point in simpler form: the company beat, but the shares dropped on the plan to raise more capital.

The late-trading reaction was not a rejection of Oracle’s AI thesis outright. It was a demand for evidence that the backlog can turn into high-quality cash flow, not just headline bookings. A $638bn RPO balance looks powerful only if investors believe it converts on time, at attractive margins and without keeping borrowings permanently higher.

The OpenAI concentration question

Oracle’s AI cloud buildout also has a customer-concentration problem. The company has become one of the most visible infrastructure partners for OpenAI-linked capacity, and the size of the backlog makes that relationship material to the equity story.

Rebecca Wettemann, cited in SiliconANGLE’s analysis, put the skeptical case bluntly.

Oracle has too many eggs in the OpenAI basket.
— Rebecca Wettemann, SiliconANGLE

Concentration does not erase the value of OpenAI demand. It does change the discount rate investors apply to Oracle’s RPO. Backlog tied to a small number of large AI buyers, some pursuing their own capital-intensive expansion plans, looks less like conventional software visibility and more like a counterparty test.

Prepaid hardware arrangements and customer contributions can reduce Oracle’s immediate cash burden. They can also make the economics harder to compare with the older software model that made Oracle a high-margin compounder. If Oracle is effectively scaling as an AI infrastructure landlord, investors will watch utilisation, power access, financing terms and customer quality as closely as they watch reported cloud growth.

Hilary Maxson, Oracle’s chief financial officer, has already signaled that the model carries a margin cost. Reuters quoted Maxson as saying gross margins will step down.

gross margins will “step down”
— Hilary Maxson, Oracle CFO, according to Reuters

Maxson’s warning shifts the debate from revenue acceleration to return on invested capital. Oracle does not need to prove that demand exists. The Q4 numbers did that. It needs to prove that each incremental dollar of AI infrastructure spending earns a return that justifies the borrowing and cash burn required to build it.

The sector is learning the same lesson

Oracle is not alone. The AI capex cycle is pushing more of Big Tech’s growth story into capital markets. Amazon secured a $17.5bn line of credit, according to Sherwood’s reporting. Alphabet has been seeking fresh capital as AI data-center spending weighs on free cash flow, CNBC reported. MarketWatch has argued that more hyperscalers may turn to stock issuance as they bankroll an AI boom that no longer fits neatly inside operating cash flow.

Oracle’s position is distinctive because it came later to the cloud infrastructure race and is now trying to use AI demand to close the gap. That gives the stock more torque when OCI revenue jumps. It also leaves less room for investors to ignore the funding side, because Oracle is not funding the same scale of buildout from the same cloud cash-flow base as Amazon Web Services or Microsoft Azure.

The scale is still large enough to move suppliers around it. The Financial Times reported that Oracle planned to spend $70bn on its data-center buildout in the coming year. That kind of number helps explain why “picks and shovels” companies tied to power, cooling, construction and chips have become a separate trade around the AI boom.

Citi’s warning that investors are becoming more selective on data-center bonds, carried by Bloomberg, fits the same pattern. Capital is available for AI infrastructure, but it is no longer being treated as frictionless. Bond buyers, equity investors and customers are all sorting the same problem: which data-center projects will have durable utilisation once the first wave of AI capacity is built.

Physical constraints add another layer. Seattle’s year-long ban on new AI data centers, reported by The Guardian, and Wired’s reporting on data-center water use show that local resistance is becoming part of the sector’s cost curve. Oracle’s Q4 reaction was about debt and margins, but the broader AI infrastructure trade now depends on power, land, cooling and municipal patience.

What investors will watch next

Three numbers will decide whether Wednesday’s selloff becomes a reset or a warning. The first is RPO conversion. Oracle needs to show that the $638bn backlog turns into revenue on a schedule and at margins that justify the capital plan.

Free cash flow comes next. Negative $23.7bn free cash flow in fiscal 2026 can be tolerated if investors believe it is a temporary investment trough. It becomes a valuation problem if fiscal 2027 capex rises toward $70bn and cash generation remains structurally weak.

Financing mix is the third test. Debt is cheaper than equity until it is not. Equity protects the balance sheet but dilutes existing holders. Customer prepayments can help, but they add complexity and may leave investors asking who really owns the risk if demand shifts.

Oracle’s management is not wrong to chase AI infrastructure demand. The company’s cloud unit is growing too quickly, and the backlog is too large, for that to be dismissed as hype. The sharper read is that Oracle has moved into a business where the market will no longer pay a pure software multiple for a capital-intensive buildout without proof of returns.

Wednesday’s selloff was therefore less contradiction than message. Oracle delivered the revenue and earnings investors asked for. Then it handed them the bill.