The interview ran in Politico around mid-April 2026. Karsten Wildberger — the CDU minister leading the Federal Ministry for Digital and State Modernisation since the Merz government took office in May 2025 — described his role in the metaphor that does the most political work in the federal IT debate: a video referee in football, watching for violations of common standards and intervening when they happen. The standards he had in mind were not procurement procedure. They were the architectural ones.

Four elements stood out. Microsoft and Palantir were named as targets for reduction — not “diversification”, reduction. The federal AI platform KIPITZ would become mandatory for authorities. A European Palantir alternative was given a two-to-three-year timeline. And the BMDS would receive de facto veto rights over IT projects of other ministries above €500,000 per year or €3 million in total project costs.

The political framing was sovereignty. The structural framing should be that this is also the largest centralisation of federal IT authority since the original IT-Konsolidierung programme of 2015.

What the targets actually look like

Microsoft is a usage curve. The federal administration paid €481.4 million for Microsoft licences in 2025 — the figure Lenhard’s parliamentary question to Parliamentary State Secretary Thomas Jarzombek extracted in February. “Less Microsoft” is therefore measurable. Wildberger has not given a specific reduction target in public, but a usage curve produces an annual data point against which the framing can be tested.

Palantir is binary. By May 2026, Vice Admiral Thomas Daum — head of the Bundeswehr’s Cyber and Information Domain Service — had said in public what Wildberger had said in private. Palantir, Daum told reporters, was “not being considered at all right now” because Germany would not permit employees of an American private company access to national defence data. The Bundeswehr was already evaluating three replacement vendors: Almato in Stuttgart, Orcrist in Berlin, and ChapsVision in Paris. Contract awards were expected by year-end 2026.

KIPITZ is the third lever. If federal authorities must use KIPITZ for AI workloads — not “should” but “must” — then federal AI infrastructure is centralised at ITZBund regardless of ministerial preference. The infrastructure layer is being built on the €250 million federal AI cloud contract awarded to the T-Systems and SAP consortium. ChatGPT and equivalent commercial systems remain excluded for data-protection reasons.

The veto, and what it captures

The veto right applies to IT projects above €500,000 per year or €3 million in total. In practice, this threshold captures all federal-level cloud and platform decisions, major application procurements, AI platform decisions (which would default to KIPITZ regardless), and identity and access management projects. It does not capture sub-€500k departmental tooling, existing contract renewals at the same scope and price, or embedded technology in specialised systems — most importantly, classified Bundeswehr and intelligence-service architecture.

The threshold is set carefully. It captures enough to meaningfully shape federal IT direction without provoking immediate constitutional questions about ministerial autonomy under the Ressortprinzip. The video-referee metaphor sits exactly here: Wildberger does not run the play, he calls it back.

This is well-designed institutional aggrandisement. That is not a criticism — institutional aggrandisement is how ministries become effective. But three things follow from concentrating federal IT decisions in a single ministry with limited parliamentary oversight of individual veto decisions.

The Ressortprinzip exists because German constitutional design distrusts central administrative authority. Wildberger’s framework partially supplants that for understandable reasons, but the reasons matter less than the precedent. The next minister inheriting these powers will have a different political position. Mandatory KIPITZ creates a single point of capture: if KIPITZ is the only sanctioned AI infrastructure for federal authorities, then capture of KIPITZ — by a future coalition, by a supplier with extraordinary influence, by a security incident — has consequences across all federal AI use. Single-vendor architectures have well-documented failure modes, even when the vendor is federally owned.

The two-to-three-year European Palantir alternative is the most fragile of the four elements. On the current funding trajectory, the timeline is rhetoric. If it is rhetoric, then Palantir reduction without a credible replacement means the Bundeswehr loses data-fusion capability it currently has. Daum’s named alternatives — Almato, Orcrist, ChapsVision — are start-ups and mid-sized firms. Their combined revenue does not yet approximate Palantir’s federal-government segment. The substitution is not yet sized to the requirement.

What the targets benefit, structurally

The cui-bono pass is short. BMDS, institutionally, acquires veto power over established ministries’ IT spending. ITZBund gains guaranteed workloads at scale through mandatory KIPITZ, justifying continued investment and political relevance. T-Systems and SAP have just won the €250 million contract whose downstream beneficiary is exactly the platform now being made mandatory. The German sovereign-IT ecosystem broadly gains opportunity as Microsoft dependency reduces. Mistral, indirectly, feeds into the KIPITZ-aligned ecosystem through the SAP–Mistral cooperation deepened at the 30 May Berlin German–French sovereignty summit.

The non-obvious party not on either list: smaller federal ministries that previously used IT decisions as a way of asserting policy autonomy. They lose that lever to BMDS.

What this article is not

It is not a claim that reducing Microsoft and Palantir is wrong. The architectural arguments for both reductions are sound — Daum’s framing of why the Bundeswehr cannot work with Palantir is, on its own terms, hard to disagree with.

It is not a claim that KIPITZ is bad. Federal AI infrastructure is necessary if commercial generative AI is to remain out of federal workflows. Without a mandatory federal alternative, shadow IT routes around any “should not” policy on ChatGPT.

It is not a claim that Wildberger acted in bad faith. Centralising authority to enforce policy preferences is how ministries operate. The question is whether the centralisation outlives the present minister’s preferences.

What to watch first

Three measurable signals will determine whether this is operational policy or a Politico cycle.

The first is how often BMDS exercises the veto in the first twelve months. A low number means rubber-stamping; the veto framework is administrative paperwork. A high number means real influence; ministries are being told no.

The second is whether Almato, Orcrist or ChapsVision actually win the Bundeswehr contract by year-end 2026, and at what scope. The Palantir replacement is the place where the rhetoric meets the procurement budget.

The third is what happens to the Microsoft licence-spending curve. If “less Microsoft” turns into licence reduction across federal authorities, the target is being met. If it turns into a migration to Microsoft Sovereign Cloud under a different label, the target is being met cosmetically — and the next Lenhard question will show it.

Sources

Topic overview: Digital Sovereignty in Europe Related articles: Sovereignty Washing Explained, Sovereignty as procurement law