Sponsored by edgeful

Clean beat across the board. FY27 guide raised. Here are the initial read from the earnings

Oracle reported Q3 FY26 this week and the numbers were straightforward. Revenue came in at $17.19 billion, up 22% year over year (18% in constant currency), slightly above the Street's $16.9 billion estimate. Non-GAAP EPS hit $1.79 versus the $1.70 consensus. Operating margin was 42.9%, a touch above expectations. Operating cash flow was $7.15 billion, well ahead of the $6.35 billion the Street had penciled in.

Management called it the first quarter in over 15 years where both organic total revenue and organic non-GAAP EPS grew at 20% or better in USD. That's a notable milestone for a company that spent most of the last decade being written off as a legacy database vendor.

They reiterated the FY26 revenue guide of $67 billion and raised the FY27 outlook by $1 billion to $90 billion, which implies about 34% growth. Q4 guidance calls for 18-20% total revenue growth in constant currency, with cloud revenue expected to grow 44-48%, which would be an acceleration from the 41% they just printed.

The infrastructure business is the engine right now. IaaS grew 81% year over year in constant currency, ahead of the Street's 79% estimate. AI revenue specifically grew 243% year over year. Multi-cloud database revenue was up 531%, though off a smaller base. They delivered more than 400 megawatts of AI capacity in the quarter, with 90% of committed capacity delivered on or ahead of schedule.

Backlog hit $553 billion, up 325% year over year. That number gets attention, but the sequential growth is worth noting too. It was up about $29 billion quarter over quarter, or 6%, which is a deceleration from 15% sequential growth in Q2 and 230% in Q1. Management said most of the sequential RPO increase came from large-scale AI contracts structured so Oracle doesn't need to raise incremental capital to fulfill them. Customers either prepay for GPUs or bring their own hardware. They've signed more than $29 billion of these types of contracts.

On the capital side, Oracle moved fast on its previously announced $50 billion financing plan. Within days of the February announcement they raised $30 billion through investment-grade bonds and mandatory convertible preferred stock, with a substantially oversubscribed order book. They said they don't expect to issue additional bonds beyond the $50 billion amount in calendar 2026, and the at-the-market equity component hasn't been tapped yet. Capex was $18.6 billion in the quarter versus $12 billion last quarter, up 55% sequentially. That's above the implied quarterly run rate based on their $50 billion full-year guide, which would put implied Q4 capex around $10.8 billion.

Gross margin on the AI capacity they delivered held at 32%, above their 30% guidance floor. They noted that adjacent services in AI data centers (general compute, storage, security, identity) typically account for 10-20% of total spend and carry higher margins. The multi-cloud database business runs in the 60-80% gross margin range. So as the revenue mix evolves, the overall OCI margin profile should continue moving up.

The applications side was a bit quieter. SaaS grew 11% in constant currency, slightly below the Street's 12% estimate. Within that, Fusion ERP was up 14% (versus 17% last quarter), Fusion HCM up 15%, NetSuite up 11% (versus 13% last quarter). There's a deceleration in both Fusion and NetSuite that's worth watching, though management didn't seem concerned. Industry SaaS solutions across healthcare, banking, retail, hospitality, and others combined grew 19%.

They reported over 2,000 customer go-lives in Q3 and said the median time to live continues to decrease. The applications win list included several displacements of SAP and Workday, including Memorial Hermann Health System (over Workday), Great Media (over Workday and SAP), Investec Bank (over SAP), and a major Wall Street bank standardizing entirely on Fusion ERP to replace SAP across all business units.

On the earnings call, CEO Mike Sicilia pushed back directly on the software disruption narrative. His argument is that complex, mission-critical enterprise systems spanning ERP, core banking, electronic health records, and regulatory compliance across entire industry verticals aren't going to get replaced by cobbled-together AI features. Oracle's position is that they're the ones embedding AI into those systems, with over 1,000 agents already live inside Fusion and hundreds more in their banking suite alone, delivered as standard features at no additional cost through regular quarterly upgrades. Their new AI Agent Studio inside Fusion lets customers and partners build their own agents on top of Oracle's application data. Larry Ellison closed the call by describing Oracle's ambition to automate entire industry ecosystems, not just individual applications, using healthcare as the example: hospitals, clinics, labs, payers, insurance, HCM, pharma, and the FDA, all connected.

A few other items worth flagging. TikTok US completed the separation of its US data operations from ByteDance, and Oracle now holds a 15% equity stake with a board seat. No revenue impact from the existing services relationship. The equity investment will be accounted for under the equity method, with Oracle's share of the new company's earnings showing up in Q4 results as non-operating income.

On sovereign cloud, management highlighted Oracle's alloy model as a differentiator, delivering full-stack OCI (not a subset of services) in sovereign configurations that can be as small as three racks or as large as 500. They framed sovereignty as covering data, operations, and contracting, and noted increasing pipeline globally.

On inferencing versus training, Clay Magouyrk made an interesting point about data center location. The latency concern about large centralized data centers being far from population centers is less relevant for most AI inference workloads than people assume. The round-trip from New York to Wyoming adds maybe 40 milliseconds, which is negligible when the model itself takes several seconds to think. The bigger latency lever right now is accelerator architecture, not geography. That gives Oracle more flexibility to build where power and land are abundant rather than chasing proximity to users.