European Tech Groups See Stronger Demand as Enterprise AI Moves Into Production

SAP, Capgemini, Sopra Steria, and OVHcloud are reporting stronger AI demand despite fears that model developers would capture most of the market’s value.

The development challenges a familiar google techmeme narrative. OpenAI, Anthropic, Google, and other model builders created the technology attracting investment. Yet established European vendors increasingly control the difficult work of connecting that technology to corporate data, workflows, infrastructure, and regulatory controls.

Recent earnings and executive comments indicate that enterprise customers are moving beyond isolated demonstrations. They now want AI systems that can operate inside finance, manufacturing, customer service, human resources, and public-sector environments.

That transition favors vendors with existing customer relationships and access to operational systems. SAP owns a substantial layer of enterprise application data. Capgemini and Sopra Steria manage complex technology programs, while OVHcloud supplies European-controlled computing infrastructure.

The shift does not mean Europe has overtaken American model developers. European companies remain dependent on foreign chips, models, and cloud technologies in many deployments. It does suggest that creating a capable model is only one part of the commercial contest.

The more difficult question is who can turn that model into a reliable production system. Europe’s established technology companies now have a credible answer.

Four European Companies Are Seeing the Same Demand Signal

The important change is not another AI product announcement. It is the appearance of related demand across software, consulting, integration, and infrastructure.

SAP reported second-quarter cloud revenue of €6.28 billion, up 24% at constant currencies. Its current cloud backlog, which represents contracted cloud revenue expected during the following 12 months, reached €22.93 billion.

Cloud ERP Suite revenue rose 27% at constant currencies to €5.53 billion. Software license revenue fell 32% to €131 million, reinforcing the company’s continuing transition from upfront licenses to recurring cloud services.

SAP maintained its 2026 cloud revenue outlook, even while reducing its expected non-IFRS operating profit. The adjustment reflected the dilutive impact of its Dremio and Prior Labs acquisitions, according to its quarterly results.

Dremio develops data lakehouse technology, which helps organizations analyze information stored across distributed systems. Prior Labs develops foundation models for structured data, the tables and records underlying many business applications.

These acquisitions illustrate SAP’s approach. It is not trying to win only through a general-purpose chatbot. It is assembling data, applications, and AI systems around the business records already stored in its software.

Capgemini is observing demand from another position. The consulting and technology services company reported first-quarter 2026 revenue of €5.94 billion. That represented 11% growth at constant exchange rates.

Bookings reached €6.05 billion, up 6.2% at constant currencies. CEO Aiman Ezzat said the performance supported Capgemini’s cloud and AI strategy, particularly for complex transformations deployed across large organizations.

The company has also argued that AI expands its addressable spending pool. AI projects can draw money from operating, research, customer-service, and product budgets, not only conventional information technology accounts.

Capgemini therefore benefits when clients progress from a demonstration to a company-wide system. A pilot might need a model and a small development team. Production deployment demands integration, security, data preparation, process redesign, monitoring, and employee training.

Sopra Steria occupies a similar position, with particular exposure to European governments, financial institutions, aerospace companies, and other regulated organizations. It plans more than 8,500 hires worldwide during 2026, including roles involving data, AI, cloud engineering, and cybersecurity.

Nearly 90% of those positions are expected to be based in Europe. Hiring plans do not guarantee future revenue, but they show where management expects client work to develop.

OVHcloud adds the infrastructure layer. Its first-half fiscal 2026 revenue reached €555.3 million, representing organic growth of 5.5%. Adjusted EBITDA was €227.2 million, or 40.9% of revenue.

The company also launched an AI lab focused on vertical agentic AI. An AI agent is software that can select and execute actions toward a defined objective, rather than only returning text.

Together, these signals span the enterprise technology stack. SAP provides applications and governed business data. Capgemini and Sopra Steria implement systems. OVHcloud supplies computing environments designed for customers concerned about European jurisdiction and operational control.

That breadth is what makes the story more significant than one strong earnings report. Demand is appearing in several parts of the production chain at once.

Why Google Techmeme Attention Is Moving Beyond Model Builders

AI value is moving toward deployment because a model cannot independently navigate the systems, permissions, and accountability structures of a large organization.

The first phase of generative AI rewarded companies that could train large models and attract users. Performance benchmarks, model releases, and funding rounds dominated technology coverage.

The enterprise phase has different requirements. A company must decide which records a model can access, which actions it can perform, and who remains accountable when it makes a mistake.

It also needs to connect the model with databases, identity systems, audit logs, and established applications. Those connections determine whether an AI assistant can complete useful work or merely generate plausible suggestions.

Consider a manufacturer using AI to adjust a production schedule. The system needs current inventory records, supplier commitments, sales forecasts, equipment availability, and employee permissions. It also needs rules preventing an uncertain model output from silently changing a critical order.

SAP already manages many of those records and workflows. That installed position gives the company leverage that cannot be reproduced by improving a model benchmark alone.

The same dynamic applies to consultants and systems integrators. Companies rarely deploy enterprise software by connecting an application programming interface and walking away. They redesign processes, classify data, negotiate employee responsibilities, and test failures before approving broader use.

Capgemini reported that AI gives it access to spending outside traditional technology budgets. This matters because production AI affects the operating model of a business, not just its software department.

A customer-service deployment might involve call transcripts, consumer privacy rules, workforce planning, and quality controls. A financial deployment needs access restrictions, documented decisions, and mechanisms for human review.

These projects reward institutional knowledge. Capgemini and Sopra Steria know how their clients’ systems were built, which regulations apply, and where previous transformation programs failed.

OVHcloud benefits from the computing consequences. Production systems generate persistent inference demand, meaning the repeated processing required whenever a deployed model responds or takes an action.

Experiments can be started and stopped. Production workloads require capacity, reliability, data residency, and predictable operations. They create recurring infrastructure demand when adoption succeeds.

This shift also explains the renewed relevance of sovereign AI. The term describes AI systems whose data, infrastructure, operations, and legal exposure meet a country’s or organization’s control requirements.

For European public agencies and regulated businesses, sovereignty can influence vendor selection. They must consider where data is stored, which legal jurisdiction governs it, and whether a foreign policy decision can interrupt access.

OVHcloud has positioned itself as a European alternative for these workloads. Capgemini and Sopra Steria combine European delivery capabilities with platforms from American and European providers.

That does not create complete independence. Nvidia remains central to advanced computing, while American companies supply many leading commercial models and cloud platforms.

However, customers do not need perfect technological independence before directing spending toward European implementation and hosting partners. They need workable controls that reduce specific operational and legal risks.

This is the commercial opening established vendors are pursuing. The model remains important, but deployment determines whether a customer receives measurable value.

The Real Contest Is Model Ownership Versus Workflow Control

The primary conflict is no longer old European companies against young AI startups. It is model ownership against control of enterprise workflows.

Model developers hold significant advantages. They possess research teams, model-training expertise, consumer distribution, and large pools of investment capital. Their products can also reach customers without a conventional enterprise sales process.

Yet a general-purpose model usually sits outside the business process it is meant to improve. It does not automatically understand a company’s product hierarchy, approval rules, contractual limits, or historical decisions.

Enterprise software vendors control more of that context. SAP’s systems support finance, procurement, supply chains, human resources, and other operational functions. Those records give an AI system the structured context needed to make a useful recommendation.

This does not guarantee SAP will capture the resulting value. Customers can connect competing models to SAP data, while cloud providers can offer their own agents and integration tools.

SAP must therefore make its data layer useful without trapping customers inside a narrow model selection. Its Business Data Cloud and Joule assistant form part of that strategy.

The company says its applications contain mission-critical domain knowledge. That is a defensible observation, but the commercial result still depends on adoption, accuracy, and additional revenue.

Capgemini and Sopra Steria face a different version of the same conflict. AI systems can automate coding, documentation, testing, and parts of consulting delivery. Those capabilities threaten labor-intensive service models.

At the same time, automation can improve margins and make previously uneconomic projects viable. It can also increase the volume of technology change that customers attempt.

Capgemini’s argument is that the additional demand outweighs the automation threat. Its first-quarter revenue supports that position, but one quarter cannot settle the issue.

Services companies must show that AI produces new, repeatable offerings rather than merely reducing the hours billed for established work. They also need to retain ownership of customer outcomes while models become easier to access.

Sopra Steria’s focus on sovereignty and regulated industries provides one response. Customers in government, defense, banking, and critical infrastructure cannot always adopt a public AI service without substantial controls.

The company expanded its work with open-source technologies to help customers deploy AI across different environments. Open source can improve portability because customers can inspect or relocate more components of a system.

Its partnership with SAP also targets sovereign cloud environments for sensitive European workloads. This combination places Sopra Steria between application software and infrastructure, where integration requirements are greatest.

OVHcloud is moving closer to the model layer. CEO Octave Klaba told Reuters that the company planned to train a family of frontier models, meaning advanced models trained from the beginning on large datasets and computing clusters.

Klaba estimated that a project once requiring roughly €1 billion might now be attempted with €150 million to €200 million. He attributed the change to improved chips, training methods, and synthetic data.

OVHcloud has completed pre-training work using Europe’s Jupiter supercomputer, although the company had not published detailed performance results. It intends to release models as open source after they reach an acceptable level.

The move is ambitious because it pushes OVHcloud beyond hosting. It also exposes the company to greater capital requirements and direct competition with dedicated AI laboratories.

Still, its broader goal fits the workflow-control thesis. OVHcloud wants to offer customers infrastructure, model access, data protection, and deployment within a European operating framework.

The likely enterprise architecture will include several providers. A company might use an American model, SAP data, Capgemini implementation services, and OVHcloud infrastructure for selected workloads.

The winner will not necessarily own every layer. It will own the interfaces, policies, and customer relationship that determine how the layers work together.

For readers following google techmeme coverage, that is the central reversal. The AI market is expanding beyond the companies that initially made generative models visible.

Strong AI Demand Still Has to Survive the Production Test

Earnings language can show customer interest, but it does not prove that deployed AI systems are producing durable returns.

Several uncertainties remain. Cloud revenue can grow because customers are migrating existing software, even when AI features contribute little direct revenue.

SAP’s cloud backlog is therefore an important demand indicator, but it is not a clean measure of AI sales. Cloud ERP migrations, contract timing, and broader application spending also affect the number.

The company’s reduced operating-profit outlook highlights another tension. SAP is spending to acquire AI and data capabilities while protecting margins expected from its cloud transition.

Its acquisitions might improve long-term differentiation. They can also raise integration costs and distract management if products overlap or customers do not adopt them.

Capgemini and Sopra Steria have a measurement problem as well. Consultants increasingly describe large transformation programs as AI-related, even when conventional cloud migration and data engineering account for much of the work.

Investors need clearer evidence that AI is generating incremental bookings. They also need to know whether automation improves project margins or pressures revenue tied to employee hours.

Headcount offers an imperfect signal. Sopra Steria’s planned hiring suggests demand, but service companies can face utilization problems if projects arrive later than expected.

Skills also become obsolete quickly. Hiring thousands of people for AI and cloud work creates training obligations, while tools may automate parts of those roles before projects reach maturity.

OVHcloud carries more direct infrastructure risk. AI demand requires expensive graphics processors, networking equipment, memory, storage, and electricity. Capacity must be installed before all customer revenue becomes certain.

The company’s first-half capital expenditure was €238.5 million, equivalent to 42.9% of revenue. Its levered free cash flow was negative €14.2 million during the period.

Those figures do not invalidate its strategy. They show why infrastructure demand must remain durable enough to support continuing investment.

OVHcloud also faces much larger competitors. Amazon Web Services, Microsoft Azure, and Google Cloud can spread infrastructure costs across wider customer bases and invest more heavily in specialized chips.

Google Cloud reported strong growth connected to its AI portfolio. AWS has said demand continues to exceed available capacity. These hyperscalers can bundle models, databases, security, and developer tools in one purchasing relationship.

European providers answer with sovereignty, local control, and interoperability. Those characteristics matter most when customers treat them as requirements rather than preferences.

OVHcloud’s AI deployment services illustrate its intended position across data preparation, training, and production. However, product availability does not guarantee customers will move critical workloads away from established hyperscalers.

Security and reliability present another production test. An experimental assistant can be corrected by its user. An autonomous agent connected to business systems can create financial, legal, or operational consequences.

Companies need identity controls, logging, evaluation, and clear escalation paths. They also need to monitor changes in model behavior after updates.

These requirements create work for established vendors, but they can slow deployments. A long approval cycle may produce consulting revenue without creating significant recurring AI usage.

The strongest evidence will come from customer outcomes. Vendors should disclose how many systems moved into production, which workflows they handle, and whether customers expanded usage after initial deployment.

Until then, stronger demand should be described as a promising commercial signal. It is not proof that Europe’s established technology groups have solved enterprise AI.

Europe’s Sovereignty Push Strengthens the Incumbent Position

European technology sovereignty is becoming a purchasing condition, giving local software, services, and cloud companies a practical route into AI projects.

Europe’s dependence on American cloud platforms and AI models has concerned policymakers for years. Recent geopolitical disputes have made continuity and legal jurisdiction more immediate business questions.

The European response does not require every component to be developed domestically. Capgemini CEO Aiman Ezzat has rejected the idea that Europe can achieve complete technological autonomy.

Capgemini instead works with AWS, Google Cloud, and Microsoft while developing offerings intended to meet European sovereignty requirements. This approach accepts technical dependence while adding local governance, hosting, and operational controls.

That compromise reflects how enterprises actually buy technology. Replacing every foreign component would delay projects and reduce access to leading capabilities.

Using foreign components without safeguards creates another risk. Customers could become dependent on a platform whose availability, contract terms, or legal exposure they cannot fully control.

Sopra Steria and OVHcloud offer a more European-centered route. Their expanded partnership focuses on industrializing AI with open-source principles and controlled infrastructure.

Sopra Steria can integrate applications and governance. OVHcloud can host data and compute under European law. Their AI partnership targets customers that need more control than a standard public-cloud service provides.

SAP also plans substantial investment in sovereign cloud capabilities. Its collaboration with Sopra Steria addresses public administrations and organizations operating sensitive infrastructure.

This matters because European governments are both regulators and major technology buyers. Procurement decisions can provide local vendors with reference customers and recurring demand.

Regulation adds complexity, but it can favor companies familiar with compliance. The European Union’s AI Act creates obligations based on an AI system’s risk and intended use.

Organizations deploying higher-risk systems need stronger documentation, human oversight, and controls. Those tasks fit the experience of enterprise software vendors and systems integrators.

Regulation does not automatically protect European companies. American hyperscalers maintain extensive compliance teams and European data centers. They can adapt products while retaining scale advantages.

The more credible European opportunity lies in specialization. Local providers can focus on government, defense, healthcare, finance, and industrial customers with demanding requirements.

They can also offer hybrid systems, which combine private infrastructure with public cloud services. This lets customers keep sensitive workloads controlled while using external models for lower-risk tasks.

Production AI is likely to deepen this hybrid pattern. Companies will select infrastructure based on data sensitivity, latency, cost, and model availability.

That complexity creates an orchestration market. Someone must decide which workload runs where, maintain consistent access policies, and monitor performance across providers.

Europe’s established vendors already perform related work in conventional cloud programs. AI increases the number of decisions, but it does not eliminate the need for integration.

The sovereignty opportunity is therefore less dramatic than a fully independent European AI stack. It is also more commercially realistic.

European companies can capture meaningful revenue by controlling regulated deployments, even while using technology developed elsewhere. That is enough to alter the distribution of value.

What the Next Three Reporting Cycles Must Confirm

The next test is whether reported demand becomes recurring AI revenue, better services economics, and sustained infrastructure utilization.

The first signal to watch is SAP’s cloud backlog and AI-related contract expansion. Investors need evidence that customers are adding AI capabilities to existing cloud agreements, rather than only completing planned ERP migrations.

SAP should also demonstrate how its data acquisitions strengthen products used in real workflows. Adoption inside finance, procurement, and supply-chain operations would support its claim that business data creates an advantage.

A slowdown in backlog growth would weaken the argument. Strong backlog accompanied by identifiable AI expansion would make the incumbent case more persuasive.

The second signal is the economics of Capgemini and Sopra Steria’s AI work. Bookings should translate into revenue without requiring proportionate headcount growth.

That would indicate that internal automation and reusable platforms are improving delivery. It would also answer concerns that coding agents will reduce the value of traditional services.

Falling utilization, weaker bookings, or pressure on project margins would point in the opposite direction. Those outcomes would suggest that AI is disrupting service providers faster than it creates new work.

The third signal is OVHcloud’s infrastructure utilization and cash generation. The company must turn capital expenditure into recurring workloads without allowing equipment and energy costs to overwhelm growth.

Its frontier-model effort adds another benchmark. OVHcloud needs to publish credible evaluations, deployment details, or customer use cases before the initiative can be treated as more than a strategic promise.

The company’s half-year performance showed improved profitability alongside continued investment. Future reports must demonstrate that this balance can survive a larger AI buildout.

Competition from hyperscalers should also remain visible. Google, Microsoft, and Amazon can respond with sovereign controls, local partnerships, and more flexible deployment options.

If they remove enough jurisdictional and operational concerns, European providers lose part of their differentiation. If customers continue demanding local control, the European position strengthens.

Developers and enterprise buyers should care because vendor selection now shapes more than model quality. It determines where corporate knowledge resides, how actions are approved, and whether systems can move between providers.

Teams should preserve evaluation records, deployment decisions, and user feedback as projects expand. A searchable AI knowledge base can help maintain that institutional context across vendors and model changes.

The Reuters story surfaced through Techmeme because it reverses an assumption that defined the early AI market. Companies building models were expected to dominate, while older European technology groups appeared vulnerable.

Production deployment is changing that balance. Models remain essential, but business data, system integration, infrastructure, and accountability now determine whether enterprises can use them safely.

That is why google techmeme readers should watch the less glamorous layers of the AI stack. SAP, Capgemini, Sopra Steria, and OVHcloud do not need to defeat every model laboratory.

They need enterprise customers to keep moving from experiments into governed, recurring operations. The next few earnings cycles will show whether that transition is producing a durable European advantage or only a temporary rise in AI-labeled demand.