SAP swaps costly AI models for cheaper alternatives, targets over €2 bln in gains | Reuters

Open original

SAP, the enterprise software giant, is replacing some of the expensive AI models that power its products with cheaper alternatives. The company expects the switch to generate more than €2 billion in AI-driven efficiency gains.

  • SAP is moving some of its product line off costly AI models onto cheaper substitutes.
  • The company projects the change will produce more than €2 billion in AI-driven efficiency gains.
  • SAP's software runs inside many of the world's largest companies, so the cost shift matters across a big customer base.

SAP, the German enterprise software company whose products run inside many of the world's largest firms, is swapping some of the expensive AI models that power its offerings for cheaper alternatives. The company expects the change to generate more than €2 billion in AI-driven efficiency gains. The excerpt available here is short, so the headline and lead are the main details to go on: the core move is replacing costly models with lower-priced ones, and the payoff is a target figure above €2 billion. Why it matters: AI inference is expensive, and for a software vendor selling to thousands of enterprises, model costs flow straight into margins. Running frontier-grade models for every task gets pricey fast. Routing work to cheaper models where they perform adequately is one of the more practical levers available, and a figure of this size suggests SAP sees it as a material business decision rather than a minor cost trim. The article's framing of these savings as "efficiency gains" also hints that the company is treating AI spending as something to be optimized like any other line item. The excerpt does not spell out which models are being replaced, which vendors supply the replacements, over what time frame the €2 billion is expected, or whether any customer-facing performance trade-offs are involved. Those gaps matter. A saving that comes from quality drops in areas customers care about would read very differently from one that comes from better model selection. Readers wanting the specifics should check the full Reuters piece, which likely contains more detail than this excerpt shows.

Ask about this article