Oracle Just Posted a Half-Trillion Dollar Backlog—So Why Did the Stock Get Crushed 10% After Hours?

Oracle’s blowout AI order book wasn’t enough to stop the stock from getting drilled after hours. Shares are down roughly 10%, probing support around the psychologically important $200 level, as investors focused less on the impressive growth metrics and more on the cash burn, capex trajectory, and what the company’s AI build-out means for leverage and risk.

On the surface, fiscal Q2 looked strong. Revenue rose 14% year over year to $16.1 billion, slightly shy of consensus but still the third straight quarter of double-digit top-line growth. Cloud revenue hit $8.0 billion, up 34%, with infrastructure-as-a-service (IaaS) jumping 68% to $4.1 billion and cloud applications (SaaS) up 11% to $3.9 billion. Non-GAAP EPS of $2.26 crushed the $1.64 consensus, though that number was boosted by a $2.7 billion pre-tax gain from selling Oracle’s stake in Ampere, its in-house Arm server chip venture.

The headline number that generated the most buzz was Remaining Performance Obligations. RPO soared 438% year over year to $523 billion, up $68 billion sequentially, driven by massive long-term AI infrastructure commitments from Meta, Nvidia, and other hyperscale and model players. RPO expected to be recognized in the next 12 months grew 40% versus 25% last quarter, a sign that a meaningful slice of that backlog should translate to near-term revenue. Management now expects an extra $4 billion of revenue in FY27 from this quarter’s bookings.

But beneath those growth figures lies the part of the story that is hitting the stock tonight: funding this build-out is expensive, and the cash flow profile is ugly in the near term. Operating cash flow in Q2 was just $2.1 billion, while free cash flow was a negative $10 billion, driven by a massive $12 billion in capex. Over the last 12 months, operating cash flow has climbed 10% to $22.3 billion, but that trailing number is being overwhelmed by the current investment cycle.

Oracle argues that the cash burn looks worse than the economics truly are. Most of the capex is going into revenue-generating equipment—GPUs, networking, and other gear—rather than land, buildings, or power infrastructure, which are typically handled through leases and only start hitting cash once a fully built data center is delivered. The company buys hardware late in the construction cycle, aiming to turn that spend into billable capacity quickly as contracted customers come online.

Still, the scale of the spending is substantial. Management now expects fiscal 2026 capex to be $35 billiion, roughly in line with prior projections and about $15 billion higher than it forecast after Q1, in order to support the surge in AI-driven demand. That’s on top of the existing build program across more than 211 live and planned regions and 72 planned multicloud data centers embedded in AWS, Azure, and Google Cloud. For skeptics, the question is no longer whether demand exists today, but whether Oracle can finance this expansion without stretching the balance sheet or diluting returns.

On that front, the company spent a chunk of the call trying to reassure bondholders and equity investors alike. Oracle emphasized that it has multiple funding levers beyond simply issuing traditional debt. Customers can “bring their own chips,” shifting upfront GPU costs off Oracle’s balance sheet. Some vendors are willing to lease chips rather than sell them outright, smoothing cash flows. Management also highlighted access to public bond markets, banks, and private debt as needed, while stressing a “foundational” commitment to maintaining an investment-grade rating. They pushed back directly on analyst models projecting $100 billion-plus of capital needs, saying they expect to require “less, if not substantially less” than that figure to complete the AI build-out.

Another worry in the market is whether Oracle’s enormous RPO stack is as ironclad as it looks. The $523 billion backlog is eye-popping, but investors are questioning the credit quality and staying power of some AI-native customers that have signed multiyear capacity deals. Management pushed back here, too, emphasizing a rigorous contracting process that factors in land, power, component supply, labor, and profitability before any deal is signed. On the risk of defaults, Oracle underscored that its infrastructure is fungible: capacity can be wiped, reprovisioned, and handed to another customer within hours, and large AI customers historically ramp usage in days once given access. In other words, if one marquee AI tenant stumbles, Oracle believes there is plenty of demand waiting in line.

Strategically, the company is also reshaping its technology stack to de-risk the build-out. The Ampere sale is part of a pivot to “chip neutrality,” with Oracle leaning on Nvidia today but making it clear that it will deploy whatever CPUs and GPUs customers want over time. The same pattern shows up on the cloud side: Oracle is doubling down on “cloud neutrality” via multicloud database regions inside Amazon, Microsoft, and Google, which helped push multicloud database consumption up 817% year over year.

Underneath the balance-sheet anxiety, demand indicators remain robust. OCI now operates 147 live infrastructure regions with 64 more planned, and delivered 50% more GPU capacity in Q2 than in Q1. A Texas supercluster with more than 96,000 Nvidia Grace Blackwell GPUs is ramping, and customers are beginning to receive AMD’s MI355 accelerators as well. Management reiterated that AI workloads should ultimately generate 30–40% gross margins over the life of a contract, with the main drag today coming from the mix of new data centers not yet fully utilized.

On the software side, cloud applications revenue reached an annualized $16 billion run-rate, with Fusion ERP up 18% in USD, NetSuite up 13%, and industry clouds growing 21%. Oracle is leaning hard into AI-infused apps and its new AI data platform, pitching enterprises on a unified way to vectorize and reason over all of their data—Oracle and non-Oracle alike—using top foundation models that run in the Oracle Cloud. Application deferred revenue is growing faster than application revenue, a sign of underlying bookings strength.

Guidance suggests management expects the top-line momentum to continue even as the spending ramp peaks. For Q3, Oracle sees total cloud revenue rising 37–41% in constant currency, total revenue up 16–18%, and non-GAAP EPS growing 12–14% in constant currency (or 16–18% in USD). Full-year FY26 revenue expectations remain at $67 billion, with the incremental RPO from this quarter skewing more to FY27.

For now, though, the market is telling Oracle it wants to see cash, not just contracts. A quarter featuring 13% revenue growth, 50%+ EPS growth, and a half-trillion-dollar backlog is being sold off because free cash flow is deeply negative and capex is marching higher. The investment thesis from here hinges on whether investors buy management’s argument that alternative funding structures, fungible data centers, and high-margin AI workloads will eventually turn this period of heavy spending into outsized returns—without breaking the balance sheet along the way.

If Oracle can execute on that promise, tonight’s selloff at the $200 line could eventually look like classic AI-capex panic. If not, the half-trillion-dollar RPO may start to feel less like a moat and more like a margin test.