Airbus is moving an initial 70 critical applications from Amazon Web Services to French cloud provider Scaleway by the end of 2028. The broader programme may eventually cover up to 900 applications across aircraft design, manufacturing and corporate operations.
The deal is not only an infrastructure migration. It is meant to support AI tools Airbus is co-developing with French startup Mistral for military applications and certified aviation systems. Airbus's digital chief said Mistral models already running on Scaleway would accelerate that AI approach, while keeping intellectual property, research and sensitive data with European partners.
Airbus will retain AWS and other American providers for less sensitive workloads. This selective split, rather than a full break from global technology, is the revealing pattern.
Digital sovereignty is becoming a workload-allocation discipline. Companies are deciding which systems, and which AI models, can remain on the most efficient global platform and which require a second environment under local legal and operational control. The result is a more expensive architecture built around business continuity, not wholesale decoupling.
Start with the minimum viable company
Cloud consolidation once looked rational: broader services, deeper engineering and lower unit costs than most regional alternatives. Economic security asks whether the company can still design, produce and decide if access, legal protection or vendor support becomes uncertain.
Airbus frames its critical estate as the applications needed for a "Minimum Viable Company." Start with the function that must survive, then work backward to infrastructure, skills, models and legal control. Customer analytics may stay on a global platform when interruption cost is manageable. Defence data, industrial IP, factory systems and production AI can justify a higher threshold.
AI turns cloud choice into a model choice
As AI moves into design, manufacturing and defence, sovereignty widens beyond server location. It includes who trains and hosts the model, who can compel production of prompts and weights, and whether a foreign legal order can interrupt a critical tool mid-cycle.
Airbus selected Scaleway after scoring more than 150 technical and legal requirements, including protection against a foreign "kill switch" and against extraterritorial laws. Technical scoring covered advanced cloud services, interoperability, scalability and AI capabilities.
That is why the Mistral partnership matters to the cloud tender. A second platform is easier to justify when it can run the preferred model under the preferred jurisdiction. Duplicate infrastructure becomes the price of keeping model, data and operational control in the same legal stack.
A second platform looks inefficient until the model, the data and the factory depend on it at the same time.
Selective migration is becoming the model
France is replacing Microsoft Azure with Scaleway for its national Health Data Hub, assessed against more than 350 technical criteria. Germany's Schleswig-Holstein is replacing Microsoft products for 30,000 civil servants; Austria's economy ministry moved 1,200 employees to Nextcloud while retaining limited Teams use for external partners.
The European Commission awarded four sovereign-cloud contracts worth up to EUR 180 million over six years in April 2026, including one led by Proximus that uses European-operated services built partly on Google Cloud technology. In June 2026 it proposed a Cloud and AI Development Act that treats cloud capacity and AI compute as one sovereignty problem, with assurance levels from data location to ownership and freedom from third-country interference.
These projects do not follow one architecture, and none pursue a full break with global technology. Some replace a platform for specific use cases; others segment workloads or pair a European model with European infrastructure. What they share is a willingness to pay more for control where dependence has become material.
Apply two tests to the sovereignty premium
A server inside Europe does not automatically create European control. Ownership, legal obligations, keys, support teams, processors and model weights can still sit elsewhere. Data residency is easy to advertise; operational and AI sovereignty depend on auditability, portability and exit.
The first test is the cost of interruption: factory output, safety, payments, regulated records, strategic IP or production AI. Include lost revenue, restart time, legal exposure, model unavailability and customer impact. For less critical systems, the recurring premium may exceed the expected loss.
The second test is the cost of exit. A sovereign platform has limited value if workloads and models cannot move onto it or staff cannot operate it. Austria's limited Teams use shows how partner lock-in can survive an internal migration.
These tests separate useful resilience from protected overcapacity. Targeted duplication is more credible than immediate technological autarky.
The premium still needs to earn a return
Willingness to pay for controlled AI and critical workloads creates revenue for sovereign providers without requiring them to displace global platforms. Investors should distinguish contracted critical workloads from capacity built ahead of customers. Contract duration, utilisation, pricing discipline and capital intensity still determine whether strategic demand becomes durable cash flow. Political support does not suspend return-on-capital discipline.
Airbus is paying for an operating option covering critical applications and the AI tools attached to them, without abandoning its first platform. The durable infrastructure will be the second route that remains technically usable, legally controlled and commercially supported when the cheapest route is no longer available.
