AWS says its European Sovereign Cloud continued operating during an October 24 exercise that disconnected two systems linking it to the company’s wider infrastructure. Its November 1 account reports that workloads kept running and customers retained access to their data during the several-hour test.
The important detail is how connectivity changed: traffic normally carried by AWS’s global backbone moved onto the public internet. The exercise tested operation without specific AWS global systems, while internet access remained available.
AWS launched the European Sovereign Cloud on January 15, 2026, describing a cloud physically and logically separate from its other Regions. Its first Region is in Brandenburg, Germany. The offering targets organizations with additional requirements around data location, operational control and continuity.
In its September 28 exercise announcement, AWS identified two connections used in normal operation: its private global backbone and a locally controlled system for limited transfers of operational AWS data. The latter supports vetted source-code mirroring and software updates, alongside approved outgoing routine information. AWS distinguished those transfers from customer content and customer-created metadata.
AWS’s subsequent report says both systems were switched off. EU-resident staff working within the EU used the sovereign cloud’s own hardware and software resources. The company reports no effect on service availability in the sovereign cloud or other AWS Regions, or on private connectivity through Direct Connect.
What the exercise establishes
For customers assessing continuity, a live exercise offers a more concrete discussion than architecture promises alone. Its scope still matters. AWS’s advance notice specified dedicated connectivity from European internet providers, preserving access to the worldwide internet. That makes the test materially different from a scenario in which external internet connectivity disappears.
The published outcome is AWS’s account of its own exercise. It should be read as vendor-reported operational evidence; the post does not present an independent audit of the event. Nor does disconnecting two technical systems settle every legal, governance or contractual question covered by the term sovereignty.
Security and procurement teams can use the report to ask narrower questions: which dependencies were removed, which customer operations were checked, how performance was measured, and what evidence is available for their particular workloads. A useful assessment should connect the demonstrated conditions to the organization’s actual disruption scenarios.
