SAP’s strategy is to lower the cost of delivering a successful AI task, not necessarily its total AI bill: it selects the cheapest model that meets the required performance standard as usage expands.210 Completed-task pricing lets SAP retain the benefit of lower delivery costs, but the projected more than €2 billion in two-year efficiency gains should not be treated as a disclosed €2 billion increase in operating profit.2

• How the savings work. SAP has replaced expensive models in its internal consulting assistant and increasingly selects task-appropriate alternatives from more than 70 models, including those from OpenAI, Anthropic and Google.210 Less expensive models can reduce inference costs without sacrificing the performance needed for a particular workflow; greater adoption can nevertheless push aggregate AI spending higher.210 The objective is therefore better economics per task and healthy margins at higher volumes—not an across-the-board reduction in AI spending.2

• Why task pricing matters. Customers pay SAP for completed tasks rather than the computing resources consumed.2 Economically, that separates the selling price from the underlying model cost: if SAP delivers the same acceptable result more cheaply, it can retain the difference rather than automatically passing the compute saving through to the customer.2 Internal use also produces efficiency benefits rather than direct task-pricing revenue, so the commercial pricing mechanism and internal productivity gains are distinct parts of the strategy.2

• Why the cheapest model cannot always be chosen. Government customers and regulated industries can restrict the countries from which models originate, while European customers increasingly want local options.2 Model selection must consequently satisfy origin and customer requirements as well as performance and price—the relevant choice is the cheapest eligible model, not simply the cheapest available globally.2

• What remains undisclosed. SAP has not identified the specific expensive models removed or their replacements, nor quantified how much of the anticipated efficiency gains will ultimately flow into margins.2 Insufficient evidence exists to translate the headline gain into a particular margin increase or attribute all of it solely to model substitution.2

• How it fits the October 21 results. SAP’s third-quarter revenue exceeded €10 billion for the first time; cloud revenue rose 26% to €6.67 billion, and non-IFRS operating profit increased 18% to €3.03 billion.1 The cloud increase was 25% at constant currencies, an important distinction when comparing it with guidance.3 SAP expected full-year constant-currency cloud revenue near the top of its €25.8–€26.2 billion range, while leaving its €11.8–€12.2 billion constant-currency operating-profit outlook unchanged.1

• Why growth expectations remain measured. One-off effects added roughly one percentage point to third-quarter cloud growth, and SAP expected fourth-quarter cloud growth of about 23% at constant currencies.16 Thus, cheaper-model deployment fits a broader effort to make expanding cloud and AI activity profitable; the results and unchanged profit guidance do not establish that the projected two-year efficiency gains have already translated into an incremental margin uplift.12

SAP’s strategy is to lower the cost of delivering a successful AI task, not necessarily its total AI bill: it selects the cheapest model that meets the required performance standard as usage expands.210 Completed-task pricing lets SAP retain the benefit of lower delivery costs, but the projected more than €2 billion in two-year efficiency gains should not be treated as a disclosed €2 billion increase in operating profit.2

• How the savings work. SAP has replaced expensive models in its internal consulting assistant and increasingly selects task-appropriate alternatives from more than 70 models, including those from OpenAI, Anthropic and Google.210 Less expensive models can reduce inference costs without sacrificing the performance needed for a particular workflow; greater adoption can nevertheless push aggregate AI spending higher.210 The objective is therefore better economics per task and healthy margins at higher volumes—not an across-the-board reduction in AI spending.2

• Why task pricing matters. Customers pay SAP for completed tasks rather than the computing resources consumed.2 Economically, that separates the selling price from the underlying model cost: if SAP delivers the same acceptable result more cheaply, it can retain the difference rather than automatically passing the compute saving through to the customer.2 Internal use also produces efficiency benefits rather than direct task-pricing revenue, so the commercial pricing mechanism and internal productivity gains are distinct parts of the strategy.2

• Why the cheapest model cannot always be chosen. Government customers and regulated industries can restrict the countries from which models originate, while European customers increasingly want local options.2 Model selection must consequently satisfy origin and customer requirements as well as performance and price—the relevant choice is the cheapest eligible model, not simply the cheapest available globally.2

• What remains undisclosed. SAP has not identified the specific expensive models removed or their replacements, nor quantified how much of the anticipated efficiency gains will ultimately flow into margins.2 Insufficient evidence exists to translate the headline gain into a particular margin increase or attribute all of it solely to model substitution.2

• How it fits the October 21 results. SAP’s third-quarter revenue exceeded €10 billion for the first time; cloud revenue rose 26% to €6.67 billion, and non-IFRS operating profit increased 18% to €3.03 billion.1 The cloud increase was 25% at constant currencies, an important distinction when comparing it with guidance.3 SAP expected full-year constant-currency cloud revenue near the top of its €25.8–€26.2 billion range, while leaving its €11.8–€12.2 billion constant-currency operating-profit outlook unchanged.1

• Why growth expectations remain measured. One-off effects added roughly one percentage point to third-quarter cloud growth, and SAP expected fourth-quarter cloud growth of about 23% at constant currencies.16 Thus, cheaper-model deployment fits a broader effort to make expanding cloud and AI activity profitable; the results and unchanged profit guidance do not establish that the projected two-year efficiency gains have already translated into an incremental margin uplift.12