SAP executive Jan Gilg has offered an answer to investors who fear that artificial-intelligence agents will hollow out traditional software subscriptions: enterprise AI will make business software more valuable, and the market pendulum will swing back. The strongest evidence for that case is not a product demonstration. It is SAP's contracted cloud demand.
The company ended the second quarter with €22.9 billion of current cloud backlog, up 27% from a year earlier. That gives SAP time, cash flow visibility and a large installed base through which to distribute AI functions. But backlog is not an AI price tag. It combines cloud migration, core ERP demand and other products, and SAP does not separately report how much revenue or profit its AI layer creates.
The backlog buys time
SAP's Q2 2026 results show current cloud backlog rising 26% at constant currencies to €22.929 billion. Cloud revenue increased 22% as reported to €6.281 billion, while total revenue rose 9% to €9.878 billion. Cloud ERP Suite revenue grew 25%. These are not the numbers of a customer base abandoning the platform overnight.
Backlog matters because large enterprise migrations are multi-year operating decisions. Moving finance, procurement, supply-chain or human-resources systems changes data models, permissions and business processes. Once contracts are signed, implementation and renewal create a distribution channel that a stand-alone AI application must work to acquire.
Yet the same inertia limits what the data proves. A company can retain demand because replacing its system of record is expensive, even if customers remain uncertain about the incremental value of its AI. Current cloud backlog measures contracted revenue expected over the next 12 months. It does not separate voluntary AI spending from contractual migration or bundled functionality.
Cloud growth and cloud margin moved in opposite directions
The revenue transition remains strong, but its economics were not uniformly stronger in Q2. IFRS cloud gross profit rose 22%, while cloud gross margin slipped 0.5 percentage point to 74.3%. IFRS operating margin also declined 0.5 point to 26.8%, even as operating profit increased 8%. Software licence revenue fell 32% and software support revenue fell 8%, illustrating the deliberate erosion of the legacy model as cloud expands.
SAP still expects €25.8 billion to €26.2 billion of 2026 cloud revenue at constant currencies, growth of 23% to 25%. It reduced the range for non-IFRS operating profit by €100 million at each end to reflect the dilutive effect of the Dremio and Prior Labs acquisitions, forecasting €11.8 billion to €12.2 billion. Management also expects constant-currency current cloud backlog growth to decelerate slightly from the 2025 rate.
Those disclosures do not refute the AI strategy. They show that building and buying the data layer has a near-term cost. For the strategy to improve valuation rather than merely defend it, AI must eventually appear in better retention, higher contract value, lower delivery cost or faster migration—not only in a larger product catalogue.
AI inherits the system-of-record advantage
SAP's strategic advantage is context. An AI agent asked to approve a purchase, close a ledger or reroute supply must understand permissions, master data, process history and audit rules. Those are already embedded in enterprise systems. SAP's Business AI Platform announcement combined its technology platform, data cloud and AI functions, alongside an Autonomous Suite and migration tools.
That architecture turns ERP from an application interface into a governed execution layer. If customers trust SAP's semantic models and controls, agents can act closer to the underlying transaction than a generic assistant operating through connectors. Distribution also becomes cheaper: AI features can reach customers through existing contracts and implementation partners.
The acquisitions support that reading. Forrester interprets Dremio and Prior Labs as an attempt to make Business Data Cloud a control plane for enterprise AI. Dremio can broaden access to data outside SAP, while Prior Labs adds automated analysis. The strategic goal is not necessarily to own the best general model. It is to control the trusted context through which models touch business operations.
The same data layer can become a tollbooth
The countercase begins with customer control. A platform that centralizes data semantics and agent governance can reduce integration work, but it can also deepen switching costs and make AI access another reason to accelerate an expensive cloud migration. Buyers may resist if value arrives mainly through bundles, consumption charges or conditions attached to moving legacy systems.
Competition also comes from above and below. Hyperscalers and independent data platforms can orchestrate agents across multiple software vendors. At the other end, AI is making bespoke applications cheaper to build, a pressure described in a recent TechRadar analysis. If the user interface shifts from clicking through seats to asking an agent for an outcome, the vendor owning the workflow may not automatically own the customer relationship.
SAP's installed base is therefore both moat and test. Deep process knowledge makes its data useful, but customers will compare the cost of using SAP's AI layer with connecting other models to the same records. Open interfaces can expand the ecosystem while also making it easier for value to migrate elsewhere.
Incremental revenue must replace narrative evidence
Demand already visible in backlog weakens the most extreme “software disappears” scenario. The unresolved question is whether AI improves the economics of that demand. A standalone AI revenue figure would help, but integration may make it artificial. Other disclosures could be more revealing: renewal rates for customers using AI, contract uplift after adoption, consumption revenue, migration time, inference cost and measurable gross-margin effects.
Evidence that would strengthen the thesis includes faster cloud conversion without higher implementation expense, widening cloud margin, or higher net retention specifically linked to Business AI. Evidence that would weaken it includes slowing backlog with stable AI usage, customers shifting agent orchestration to neutral data platforms, or continued margin pressure without corresponding contract uplift.
Gilg's optimism is plausible because SAP owns a valuable place in the enterprise stack. The €22.9 billion backlog shows that customers are still committing to that place. It buys the company a runway for AI; it does not yet tell investors how profitable the destination will be.