Europe's legacy tech groups cash in as AI shifts from models to deployment

The prevailing narrative around artificial intelligence held that the spoils would go to the companies building the largest models. Quarterly results from Europe's biggest established technology groups tell a different story. SAP, Capgemini, Sopra Steria and OVHcloud have each reported stronger demand, faster growth or upgraded guidance as corporate customers move from pilot projects to embedding AI across finance, supply-chain, defence and healthcare operations. The common thread is not model supremacy but the difficult, regulated work of wiring probabilistic software into deterministic, legacy-heavy environments.

SAP's backlog signals where the capital is flowing

SAP's cloud backlog, a forward-looking indicator of contracted future revenue, rose 26% at constant currencies to €22.9 billion. The increase reflects continued migration of critical finance, procurement, supply-chain and human-resources systems onto the company's platforms, systems that increasingly serve as the foundation for AI deployment. Christian Klein, the chief executive, has reinforced that positioning with the acquisitions of Dremio, a data-lakehouse specialist, and Prior Labs, an AI company focused on tabular data. Both deals target the same bottleneck: making decades of structured enterprise data accessible to generative and predictive models without rewriting the underlying applications.

The German group's numbers matter because they measure committed spend, not experimental budgets. When a multinational moves its general ledger or its global procurement engine onto a cloud tenant that now hosts AI agents for invoice matching or demand forecasting, the switching costs become enormous. That lock-in is precisely what investors have been waiting to see from European software incumbents.

Consulting and integration firms ride the implementation wave

Capgemini and Sopra Steria, the two Paris-listed services groups, are benefiting from the labour-intensive phase that follows model selection. Capgemini raised its annual revenue-growth target after bookings climbed 9.2% in the first half, while Sopra Steria upgraded its full-year outlook after organic growth accelerated to 5.3%. The work driving those numbers is unglamorous but essential: integrating models into existing workflows, cleansing and governing data, building audit trails and compliance layers, and maintaining the hybrid estates that result.

That integration burden is heaviest in sectors where software cannot be ripped out and replaced. Defence procurement systems, air-traffic-control platforms, hospital patient-record architectures and power-grid management tools all run on codebases that predate the transformer architecture by decades. AI value in those environments depends on wrapping models around specialist logic, not on benchmark scores. The consulting firms that understand the regulatory and operational constraints of those domains are capturing a disproportionate share of the implementation spend.

Sovereignty requirements create a structural tailwind

A second trend is reinforcing the incumbents' position: growing demand for deployment environments that remain under European legal and operational control. Arthur Sadoun, chief executive of Publicis Groupe, has said clients increasingly want advanced models running inside infrastructures where they retain sovereignty over both the technology stack and the data it processes. The preference is most acute in defence, aerospace and critical infrastructure, where extraterritorial legislation such as the US Cloud Act creates legal risk for data hosted on American-controlled clouds.

Airbus illustrates the shift. The aircraft manufacturer has selected Scaleway, the cloud subsidiary of French telecoms group Iliad, to host sensitive industrial and defence applications alongside AI tools developed with Mistral AI. Airbus expects around 70 critical applications to run on Scaleway by the end of 2028. The decision reflects a procurement logic that weighs regulatory exposure as heavily as technical performance, a logic that favours European providers with data-centre footprints and ownership structures that keep them outside foreign jurisdiction.

OVHcloud offers early proof of infrastructure demand

OVHcloud, the French-listed cloud provider, reported a 20.2% rise in public-cloud revenue in its third quarter. While the absolute scale remains modest compared with the hyperscalers, the growth rate provides early commercial evidence that the sovereignty argument is translating into contract wins. The company's positioning, European-owned, European-operated, with no US parent company, directly addresses the compliance requirements that are now appearing in tender documents across the public sector and regulated industries.

The infrastructure layer is where the sovereignty debate is most concrete. A model trained in California but inferenced in a Frankfurt data centre owned by a US corporation still leaves the customer exposed to foreign legal process. OVHcloud and Scaleway sell the assurance that no such exposure exists. Whether that assurance commands a durable price premium remains an open question, but the revenue trajectory suggests buyers are willing to pay for it today.

The multi-model reality favours orchestration over ownership

Enterprises are not standardising on a single foundation model. They are running Llama for code generation, Mistral for multilingual document processing, proprietary fine-tunes for domain-specific tasks, and closed models from US labs where licensing permits. That fragmentation creates demand for a layer that handles routing, observability, cost control, versioning and compliance across heterogeneous model fleets. SAP's business-technology platform, Capgemini's RAISE framework and the managed-services offerings from the French integrators are all competing to become that orchestration layer.

The economics of orchestration differ from the economics of model training. Training rewards capital intensity and compute scale. Orchestration rewards domain knowledge, integration depth and the ability to operate at the intersection of IT and OT, operational technology, environments. European incumbents have spent decades building the latter. They are now discovering that the AI boom monetises those assets in a way the previous cloud transition did not.

Margins and sustainability remain unproven

The enthusiasm in the earnings releases is tempered by two structural risks. First, the same AI tools that create integration revenue also automate lower-value consulting and coding tasks. If a junior developer's work can be done by an agent, the billable-hours model that underpins services margins comes under pressure. Capgemini and Sopra Steria have yet to demonstrate that higher-value AI governance work expands fast enough to offset the compression at the bottom.

Second, the current demand cycle may be front-loaded. Enterprises are spending to become AI-ready, modernising data estates, standing up governance frameworks, negotiating sovereign-cloud contracts. Once those foundations are in place, the incremental spend per quarter could decelerate. SAP's backlog visibility extends several years, but the services firms operate on shorter horizons. The next four quarters will test whether the growth rates are structural or cyclical.