SAP's Cloud Backlog Grew 26% to €24.1 Billion: Enterprise AI Budgets Are Real, and They Land in Software, Not Labour

Total revenue was €10.270 billion, up 13% reported and 12% at constant currencies. IFRS operating profit rose 18% to €2.930 billion and non-IFRS operating profit rose 18% to €3.030 billion. Free cash flow of €1.718 billion was up 36%, and full-year free cash flow guidance held at approximately €10 billion against €8.24 billion in 2025.

Guidance moved in one direction only: cloud revenue toward the upper end of €25.8–26.2 billion at constant currencies, cloud and software revenue toward the upper end of €36.3–36.8 billion, and non-IFRS operating profit unchanged at €11.8–12.2 billion. That combination says the incremental commitment is being recognised as revenue without a matching step-up in profit.

A backlog is only interesting if it converts. Here it converts at almost exactly one-for-one: current cloud backlog grew 26% at constant currencies while cloud revenue grew 25%. A backlog growing materially faster than the revenue it funds would signal discounting, bundling or contracts pulled forward from 2027. A backlog growing slower would signal a thinning pipeline.

Neither is happening. The backlog is converting, not accumulating — which is why this print reads as evidence of demand rather than as evidence of accounting engineering.

Software support revenue falls mechanically as customers migrate from maintenance to cloud subscriptions, so that line proves little on its own. The informative line is services: €1.021 billion, down 3.7% year over year, while cloud and software revenue rose 15% to €9.249 billion. The labour that implements and maintains enterprise systems is not where the AI budget is going.

IBM published the identical split on Oct 21: software revenue of $7.8 billion, up 8% and 9% at constant currency, against consulting revenue of $5.3 billion that was flat, with Data software up 19% and Automation down 3%. Two companies, two business models, the same budget decision to spend on software rather than on services.

Set the contracted book against the pool that infrastructure spending depends on. In the twelve months to June 2026, Microsoft, Amazon and Alphabet together spent $421 billion on property and equipment. SAP's own trailing capital spending was €735 million, or 1.9% of revenue. The application layer converts enterprise AI spending into revenue without carrying the build; the build sits with the cloud operators and the accelerator vendors.

And the build is not self-funding everywhere. Amazon's trailing capital spending of $173.0 billion consumed 107% of operating cash flow and pushed trailing free cash flow to minus $11.6 billion. Microsoft covered its $115.9 billion outlay 1.6 times over with $182.9 billion of operating cash flow, and Alphabet 1.4 times. The difference is one word: contracts.

The chain runs from accelerators to hyperscale capacity to the applications sold on top, and then to the firms that implement and buy them. Each link is paid differently, and this cycle has moved money away from the layer that sells hours.

The connective tissue is inference cost. SAP's cloud gross margin fell 0.7 points to 73.9% even as cloud revenue grew 26%, and management has said AI server and developer costs weighed on its operating margin. Serving AI workloads is now a cost-of-revenue line inside a subscription, which is the first place the AI bill shows up in the P&L of a software vendor.

One company's backlog proves nothing on its own, so the peer set matters. ServiceNow grew contracted revenue due within twelve months 21% to $13.2 billion in the June quarter and grew subscription revenue 23% at constant currency, with AI products past $1 billion of annual contract value. Workday went the other way: fiscal 2027 subscription revenue guided to $9.94–9.95 billion, 12% to 13% growth against 14.5% delivered in fiscal 2026, with third-quarter contracted revenue growth expected at 11% to 12%.

This is not a story about who adopted AI fastest. It is about how the AI gets priced: per transaction and per process, or per seat. SAP sells systems of record that get migrated and then consumed by transaction volume; seat-based suites sell access to human users, and agentic software is a deflationary force against that.

SAP's ADR closed at $231.86 on Oct 21, 2026, a market capitalisation of about $270 billion and an enterprise value near €230 billion. That is 6.0 times trailing sales and 17.9 times trailing EBITDA, on a 3.6% free cash flow yield and a 13.1% return on invested capital, with net cash and no leverage.

Nothing in the multiple prices an AI re-rating. On these numbers the market is paying a mature software multiple for a contracted AI-era revenue stream, which means the backlog defends the valuation rather than expanding it. The upside case is that the Extension Suite inflects above 8% constant-currency growth; the downside case is that services keep shrinking and the growth premium compresses toward the cloud infrastructure cohort.

Three things are settled by this print, and one is not. Settled: enterprise AI spending is contracted rather than promised; the split between software and services is visible in two independent companies on the same day; and the infrastructure pool is growing faster than the contracted pool it serves.

Not settled: how much of SAP's growth is AI revenue rather than S/4HANA migration and pricing uplift. SAP publishes no Business AI revenue line, so the question cannot be settled from outside until it does.