Why now — Palantir alternatives for Europe

Tendered October 2025, awarded April 2026: four European providers, not a single winner — for the EU institutions, bodies and agencies, selected under a published Cloud Sovereignty Framework: eight objectives, scored at set assurance levels. Not a policy paper. A contract.

Sovereignty moved from the preamble to the award criteria — with a methodology anyone may reuse.

You can copy the scoresheet into your own tender tomorrow.

And note what it bought: the cloud layer. The AI and data layers above it — where the decisions live — are still the open question. They are this session.

The interior intelligence services of the two largest member states — the most demanding analytical users on the continent, and historically reference customers for the American platforms.

If a European platform can carry their workload, capability is no longer the automatic counter-argument.

“Not possible in Europe” stopped being an argument and became an excuse.

“Vendor-operated” means the supplier’s staff hold the administrative keys and run the system on your behalf — which is how Foundry-class platforms are usually delivered.

A requirement written this way disqualifies a delivery model, not a logo — and every future tender can copy the clause.

Who operates the platform is now on the scoresheet, next to what it does.

A communiqué builds nothing by itself. What it builds is cover: every procurement officer who was already uneasy now has a policy to cite — and budget lines follow policies.

Belgium, the Netherlands, Poland, Spain, the Nordics — less vocal than Paris and Berlin, with the same exposure and less leverage. A national champion with a different flag is the same problem in a different language.

The institutions that kept functioning were the ones whose data, and the ability to decide on it, stayed available and under their own control. A bank, a grid operator, a hospital runs on exactly the same asset.

Nobody here will fab their own silicon or train every frontier model — that dependency is real, and it is not the point. The point is dependence: someone else deciding how you use what you run on. Conflating the two is how organisations conclude sovereignty is impossible — and sign the renewal.

The test: when the vendor changes the terms, what stops working — and how fast?

Whoever owns that layer owns your switching cost. Choose it like infrastructure, not like a licence.

The Commission’s Cloud and AI Development Act (June 2026) sets four assurance levels for sensitive public and critical-sector workloads. The top levels cannot be retrofitted: ownership and supply chain are decided the day you sign.

Sovereign infrastructure with a non-sovereign superstructure is half a solution.

Pipelines, lakehouse, governance, catalogue. A decade of European data-platform investment lives here, and nothing above it works without it. But this layer answers what is true, not what should we do.

Entities, relationships, state and rules: an aircraft, its parts, its crew, its mission — as living objects, not rows in seventeen tables. This is the layer Palantir genuinely got right, and the one most alternatives quietly skip.

An agent gets an identity and scoped permissions, retrieves from the ontology rather than raw tables, and every action is recorded. Capability is the easy half; governance is the layer. And keep the model swappable: a model you can replace in an afternoon is a supplier — one your processes have grown around is a landlord.

Most organisations’ processes live in siloed systems or in people’s heads. Process flows put them on the model: each step acts on the ontology under policy, and the outcome is written back to the systems where work happens — with an audit record a reviewer will accept. This is where a data platform becomes an operations platform.