Airbus Picks Scaleway for EU-Only Aircraft Design and Manufacturing

Airbus has chosen French cloud provider Scaleway to manage business-critical tasks such as aircraft design, engineering, and manufacturing, underscoring the growing importance of data sovereignty for European companies. The selection process considered technical capabilities, operational excellence, and legal governance, with a focus on shielding against non-European legislation, notably the US CLOUD Act. Airbus EVP Digital Catherine Jestin highlighted the necessity of protecting critical data assets from foreign laws, while Scaleway CEO Damien Lucas emphasized the importance of infrastructure that offers performance alongside trust and control for AI in advanced industries.

Airbus's strategy involves assigning each workload to the environment best suited to its technical, operational, and regulatory needs—a sovereignty tier for jurisdictionally exposed data alongside existing hyperscaler capacity. The stack is cloud-native, incorporating Kubernetes, Terraform, and S3-compatible APIs, with legal exposure as the key differentiator. This move follows Airbus's partnership with Mistral AI in May to run European models on European infrastructure and Scaleway's selection in April as one of four providers under the EU Commission's €180 million Cloud III framework for EU institutions. The €180 million is the total value across all four providers, not Scaleway's individual allocation. Additionally, Scaleway acquired HPC provider Qarnot in July.

Operational metrics for the deployment, such as latency, inference cost, and throughput benchmarks, have not been released. The Airbus contract value also remains undisclosed. The decision is influenced by real-world scenarios, such as the ICC chief prosecutor's reported loss of Microsoft email access after being sanctioned by executive order in 2025, an account Microsoft's president later disputed cutting off, and export-control actions that suspended access to Claude Fable 5 and Mythos 5 in June. These examples show that regional datacenter placement does not resolve parent-company legal exposure.

The pressure is evident in real deployments. European practitioners have been blocked from adopting AI models on US-parent cloud infrastructure for the same reasons Airbus cites, including a Dutch bank that declined approval and an oncology IT architect requiring guaranteed EU-hosted inference. A US-based SaaS founder described losing European customers due to the need for an EU supplier.

The architectural cost of this approach is a split-brain infrastructure, with sovereign clouds lacking the depth of managed AI serving stacks offered by hyperscalers. Running a sovereignty tier alongside a multi-cloud introduces data gravity penalties, cross-border orchestration complexity, and the reality that European jurisdiction does not equate to equivalent GPU density or inference optimization. Larger vendors can establish EU legal entities to neutralize jurisdictional risk, while small US SaaS vendors often cannot, leading to a bifurcated market.

Practitioners should tier their inference stack by jurisdictional exposure, placing model weights, training data, and sensitive prompts on infrastructure with zero US-parent legal exposure, and latency-tolerant commodity inference elsewhere. They must also budget for the operational tax of maintaining two control planes and ensure the boundary is hardened to prevent context from crossing the wire uncontrolled.