Happy dependence? Why Europe will not leave the hyperscalers in 2026

Let us start with the uncomfortable observation. The major analyst firms' forecasts for 2026 converge: despite growing interest in digital sovereignty and rising distrust of American cloud providers, no European company will turn away entirely from the hyperscalers in the short or medium term. The reasons are structural: unmatched depth of services, skills available on the market, ecosystem effects, and the sheer cost of a total migration — which nobody, in reality, seriously proposes. Claiming otherwise condemns the sovereignty discourse to unrealism, and therefore to inaction.

The all-or-nothing trap

The European debate has long oscillated between two equally sterile postures: resignation ("we will never catch up, let's just optimise the contracts") and incantation ("let's migrate everything to European players"). Both share the same flaw: they treat dependence as a binary state, when it is in fact a portfolio of heterogeneous risks. Dependence on a generic compute service has neither the same gravity nor the same reversibility as dependence on a proprietary database, an identity service, or a frontier-model API. Practicable sovereignty starts with that granularity.

Five moves that change the position, not the scenery

First move: selective geopatriation — identified by analysts as a structuring trend of 2026 — that is, repatriating critical workloads to qualified offerings (SecNumCloud-type) or on-premise, while accepting that the rest stays with the hyperscalers. Second move: encryption with externalised keys, held out of the provider's reach, for everything that remains — real protection at rest and in transit, while staying clear-eyed about processing. Third move: contractualised and tested reversibility — an exit plan never exercised is a fiction; an annual restoration test at a second provider is an insurance policy. Fourth move: architectural discipline — proprietary managed services only where the gain justifies them, open standards (containers, standard SQL, S3-compatible) everywhere else. Fifth move: steering growth — failing migration of the existing stock, direct new projects towards the emerging European capabilities, gigafactories included; it is demand that will make the supply exist.

Sovereignty as a direction, not a state

There is nothing heroic about this programme, and that is its strength. It turns sovereignty from a slogan into a management metric: share of critical workloads under qualified control, share of data encrypted with own keys, tested switchover time, share of new projects on European capabilities. Four figures an executive committee can track quarter after quarter — and that a regulator, a customer or an insurer will soon know to ask for.

Happy dependence does not exist; clear-eyed dependence does. In 2026, Europe will not exit the hyperscalers — but every organisation can exit ignorance of its own exposure. It is less spectacular than a sovereign revolution. It is infinitely more useful. And it is, perhaps, the only path by which the revolution will eventually no longer be necessary.