Oracle Promised an AI Future—Wall Street Saw a Debt Spiral Instead

Oracle’s latest earnings report was supposed to be another victory lap for the AI trade. Instead, it turned into a live-fire stress test of how much cash burn and leverage investors are willing to tolerate to chase that story. The numbers checked most of the growth boxes—double-digit revenue, surging cloud, a half-trillion-dollar order book—but the market’s verdict is clear: the funding plan looks messy, the free cash flow profile is ugly, and the risk/reward has shifted. With the stock down roughly 10–12% and probing support around the psychologically important $200 level, Oracle has managed to drown out even a dovish Fed cut.

On the surface, fiscal Q2 looked strong. Revenue grew 14% year over year to about $16.1 billion, marking the third consecutive quarter of double-digit top-line expansion, even if it came in a bit shy of street expectations. Cloud revenue rose 34% to $8.0 billion, driven by 68% growth in IaaS to $4.1 billion and 11% growth in cloud applications (SaaS) to $3.9 billion. Non-GAAP EPS of $2.26 crushed the consensus $1.64, though that beat was inflated by a $2.7 billion pre-tax gain from the sale of Oracle’s stake in Ampere. Strip out that one-time benefit and the earnings story looks more pedestrian, which matters when the rest of the P&L is under scrutiny.

The biggest attention-grabber was Remaining Performance Obligations. RPO soared 438% year over year to $523 billion, up $68 billion sequentially, fueled by massive long-dated AI infrastructure commitments from the likes of Meta, Nvidia, and other hyperscalers and model providers. The portion expected to be recognized in the next 12 months grew 40%, up from 25% last quarter—evidence that some of this backlog is moving closer to the income statement. Management now expects the Q2 bookings surge to contribute an extra $4 billion of revenue in FY27. On paper, this is the kind of contracted growth that usually commands a premium multiple.

But the market tonight is not paying for paper. It’s paying for cash, and that’s where the story breaks down . Operating cash flow in Q2 was only about $2.1 billion, while free cash flow was roughly negative $10 billion, driven by an eye-watering $12 billion of capex in the quarter. Over the last 12 months, operating cash flow is up 10% to $22.3 billion, but that trailing strength is being completely overshadowed by the current AI build-out. Investors are staring at a business that is growing fast, but currently spending even faster.

Capex is the heart of the concern. Management now expects fiscal 2026 capex to be around $50 billion, roughly $15 billion higher than what it was signaling after Q1 and equivalent to an astounding ~75% of projected FY26 revenue. That spending supports a network of 147 live OCI regions, more than 200 planned regions, and dozens of multicloud data centers embedded in AWS, Azure, and Google Cloud. The company argues the optics are worse than the economics—that most of this spend is on revenue-generating hardware (GPUs, networking gear) late in the construction cycle, rather than on owned land and power infrastructure, and that capacity should become billable quickly as contracted customers ramp.

Even if you accept that logic, you’re still left with the question the market is asking: how does Oracle finance this without over-stretching the balance sheet? Analysts have floated models that imply $100 billion or more of cumulative capital needs to fully build out the AI infrastructure footprint. Management pushed back on that during the call, saying it expects to need “less, if not substantially less” than those estimates and stressing the intent to remain an investment-grade credit. They also emphasized multiple levers: customers bringing their own chips, vendors leasing hardware rather than selling it, and flexible access to public bonds, bank lines, and private credit if necessary.

The problem is that “multiple levers” isn’t the same as a crisp funding roadmap. Investors are being asked to accept negative free cash flow, rising capex, and opaque long-term debt usage all at once. That’s why you’re seeing price targets come down even from bulls. Stifel cut its target from $350 to $275 while reiterating a Buy, highlighting that accelerating RPO and cloud growth are now being overshadowed by uncertainty around funding and capital intensity. RBC trimmed its target from $310 to $250 and framed the issue bluntly: the debate has shifted from top-line acceleration to capital efficiency and the path back to sustainable free cash flow.

There’s also a growing debate around the quality of that $523 billion backlog. The commitments from mega-caps like Meta and Nvidia are reassuring, but some AI-native customers further down the credit spectrum understandably raise eyebrows. Oracle responded by emphasizing its rigorous contracting process—locking in land, power, component supply, labor, and economics before signing—and stressing the fungibility of its infrastructure. If a given AI tenant stumbles, Oracle argues it can wipe and reprovision capacity to another customer within hours, with plenty of demand still waiting in line. It’s a reasonable operational argument, but in a risk-off tape, investors tend to focus on who owes the money, not just how fast you can reassign GPUs.

Strategically, Oracle is trying to reduce long-term technology risk even as it accepts more financial risk in the near term. The Ampere stake sale is part of a pivot to “chip neutrality”: Oracle will keep buying Nvidia’s latest GPUs but make it clear it’s willing to deploy whatever CPUs and accelerators customers want, including AMD and others. On the cloud side, “cloud neutrality” and multicloud partnerships are gaining importance as Oracle embeds its database services inside AWS, Azure, and Google Cloud, driving an 817% year-over-year increase in multicloud database consumption. The idea is to spread dependency across vendors and ecosystems, even if that means higher complexity now.

Guidance suggests management still sees strong growth ahead despite the spending spike. For Q3, Oracle is calling for total cloud revenue to grow 37–41% in constant currency, total revenue to rise 16–18%, and non-GAAP EPS to increase in the low- to mid-teens in constant currency (mid-teens in USD). Full-year FY26 revenue expectations remain anchored around $67 billion, with the incremental RPO from this quarter leaning more toward FY27. The long-term pitch is that AI workloads will ultimately carry 30–40% gross margins over the life of contracts, and that as utilization rises in these new data centers, cash generation will inflect.

The market isn’t there yet. In the near term, the stock is being treated as a referendum on AI capex discipline. Oracle is down around 11–12%, trading near six-month lows and testing the $200 line, and it has pulled a broad swath of AI-exposed names down with it—Nvidia, AMD, big cloud platforms, and software names tied to AI infrastructure demand. The selling pressure has been concentrated in tech and AI, while the equal-weight S&P is down only modestly, suggesting a rotation rather than a full-blown de-risking. Beneficiaries appear to be lower-capex, cash-rich sectors and more diversified cyclicals that can still enjoy the Fed’s dovish tilt without being asked to fund a $50 billion data center build-out.

That’s the bigger backdrop: just 24 hours ago, Powell delivered a quarter-point cut with a tone that markets interpreted as dovish, helping push the S&P to within a few points of its all-time high. Today, that macro tailwind is being overshadowed by a micro-level reality check on what the AI build-out actually costs. The core question from here isn’t whether Oracle has demand—it clearly does—but whether investors are willing to underwrite the journey from negative free cash flow to the high-margin AI end state without a clearer map of leverage, timing, and capital returns.

If Oracle ultimately executes, tonight’s selloff will look like classic capex panic at a key support level. If it stumbles, this quarter will be remembered as the moment the market decided that “AI orders” and “AI economics” are two very different trades.