Refuse the Terms, Not the Technology: Europe's Palantir Procurement is Forfeiting Strategic Optionality

On 28 April 2026, Vice Admiral Thomas Daum, Inspector of the Bundeswehr’s Cyber and Information Domain Service, told Handelsblatt that Palantir software is “not being considered at all right now” for the German armed forces, and that granting “employees of a private American company” access to national datasets was “unthinkable” (Ground News, 2026). The remark landed as an outlier. In the seven months bracketing it, NATO Allied Command Operations completed the fastest software procurement in the alliance’s history to field the Maven Smart System NATO at SHAPE in Mons (NCIA, 2025); the United Kingdom signed a £240.6 million enterprise agreement with Palantir under a national-security exemption, without competitive tender, in addition to a separate £1.5 billion strategic partnership, announced during the September 2025 Trump state visit (PublicTechnology, 2026; The Register, 2026); France renewed the DGSI contract running since 2016 (Business Wire, 2025); Poland’s defence minister signed a letter of intent on data, artificial intelligence and cybersecurity (Bloomberg, 2025); and Palantir reported first-quarter revenue of $1.633 billion, an 85% year-on-year increase, raising 2026 guidance to $7.65 billion (Yahoo Finance, 2026).

Daum’s position, on this evidence, is not eccentric. It is the only procurement posture in Europe that treats strategic optionality as a constraint rather than a slogan. The operative question facing European governments is no longer whether to adopt Palantir, but whether the terms on which adoption is occurring preserve the future ability to choose. The argument that follows is critical of those terms, not of the technology, and it can be stated bluntly: the current European procurement architecture trades short-term capability gains for permanent reductions in the future choice set, and it does so through specific, identifiable, and correctable contractual choices.

The expansion is faster than the safeguards

The first thing to register is sequencing. Palantir’s European footprint has expanded by an order of magnitude in less than a year. The Maven Smart System NATO contract, signed on 25 March 2025 between the NATO Communications and Information Agency and Palantir, took six months from requirement to deployment, described by NCIA itself as “one of the most expeditious in NATO’s history” (NCIA, 2025). Within thirty days, Allied Command Operations was using the system. Lieutenant General Markus Laubenthal, SHAPE chief of staff, framed the rationale as accelerated decision-making (SHAPE, 2025). The United States, in parallel, consolidated 75 separate Army contracts into a single Enterprise Service Agreement worth up to $10 billion in August 2025 (Breaking Defense, 2025), a procurement vehicle whose stated design goal is to maximise “buying power” through “volume-based discounts” and “eliminate redundant contracting processes” (InsideDefense, 2025).

That is not a neutral procurement template. It is a specific bet that the gains from consolidation exceed the costs of foreclosing future competition. In the United States, where Palantir is a domestic supplier and the Defense Department exerts direct legal and political leverage, the bet may be defensible. Replicated in Europe, the same architecture has a fundamentally different risk profile. The UK MoD’s £240.6 million follow-on contract, signed on 30 December 2025 and effective 1 April 2026, was awarded directly under a national-security exemption in the Procurement Act 2023 (PublicTechnology, 2026). The Spanish Ministry of Defence’s €16.5 million intelligence-fusion contract in October 2023 used the same negotiated procedure without public tender. The pattern is consistent: large commitments, long durations, no published competitive justification, and contractual sovereignty assurances rather than architectural ones.

The lock-in is in the procurement architecture, not the code

It is tempting to frame the strategic risk in technological terms, such as proprietary formats, the ontology layer, dependencies on Foundry-Gotham-AIP integration. That framing, while not wrong, misses the operative variable. Palantir’s net-dollar retention rate in the third quarter of 2025 of 134%, meaning existing customers spent 34% more than in the equivalent period a year earlier, is the financial signature of a switching-cost mechanism that compounds with use (StrategyPunk, 2025). But the mechanism is institutional before it is technical. The ontology embeds doctrine: target categories, threat schemas, escalation logic, the conceptual structure through which an organisation makes sense of its data. Once that conceptual structure has been encoded for several years, the cost of switching is not the cost of installing new software but the cost of rebuilding institutional thinking.

The canonical illustration is the New York Police Department’s 2017 exit attempt. Having developed an alternative system named Cobalt, the NYPD complained publicly that Palantir would not cooperate with the extraction of analyses the department had itself generated (Burry, 2026, on the documented record). The dispute, in Michael Burry’s reading of the public file, “suggests the ‘moat’ might just be Palantir’s obstruction of data transfer at relationship’s end” (Burry, 2026). Whether or not that characterisation is exhaustive, it identifies the operative procurement failure with precision: contracts that fail to mandate, in advance, the export of customer-generated ontologies and analytics in open formats hand the vendor a one-sided exit veto.

European contracts publicly disclosed to date do not contain such mandates. The only safeguard the UK government has cited is Defence Minister Luke Pollard’s statement that “UK Defence data used and developed in Palantir’s software remains sovereign and under the control of the MOD” (UK Defence Journal, 2026). That is a contractual promise, not an architectural guarantee, and as the Swiss Armed Forces’ own assessment makes clear, the two are not the same thing.

The Swiss precedent

The Swiss case is the most important counter-data point Europe has, and it is largely absent from the UK and Polish procurement debates. In December 2025, the Swiss outlet Republik and the WAV research collective revealed, on the basis of 59 freedom-of-information requests across 41 federal offices, that at least nine Swiss federal agencies had rejected Palantir over a seven-year period (Republik, 2025). The Swiss Armed Forces’ 20-page risk assessment of Palantir’s logistics module concluded that US jurisdiction created an architectural sovereignty problem that the CLOUD Act could not be patched away by contract; that data leakage could not be technically prevented; and that the requirement for Palantir specialists to be permanently on-site “could limit the army’s ability to act in crisis situations” (Republik, 2025).

The conclusion of the assessment was operational, not ideological, and “The Swiss Army should consider alternatives to Palantir.” The relevance for European NATO members is that Switzerland conducted the technical due diligence the UK, Poland and Spain did not, and reached a different conclusion. Whatever else can be said about the divergence, it cannot be attributed to capability scepticism alone; the Swiss had access to the same product specifications. What they did differently was assess CLOUD Act exposure as an architectural problem rather than a public-relations one. That distinction matters because the 2018 US CLOUD Act compels US-headquartered companies to comply with US law enforcement and national security data requests for data wherever stored, including on European servers (FOTI, 2026). The implication is not theoretical. MEP Raquel García Hermida-van der Walle, in Parliamentary Question E-001361/2026, put it in operational terms: “I could imagine a situation where a member state does not want US grabbing biometric data of its citizens, the US could easily withdraw Palantir from a member state, on which the country’s intelligence services or police became dependent” (Hermida-van der Walle, 2026). The Future of Technology Institute’s Cloud Defence — An Exposed European Flank report (FOTI, 2026), published in Brussels on 17 April 2026, classified sixteen of twenty-eight European countries as high-risk on US cloud-layer dependency alone, explicitly noting that its scope did not extend to the analytical-software layer above, which is to say: the Palantir layer.

The operationalised alternative

Two further European data points complete the picture. On 15 May 2026, the Süddeutsche Zeitung–WDR–NDR consortium reported that Germany’s domestic intelligence service, the Bundesamt für Verfassungsschutz, had selected the French firm ChapsVision’s ArgonOS rather than Palantir for its data-fusion platform, an air-gapped deployment processing more than 300 data sources inside a sovereign environment (Modern Ghana, 2026; UNITED24 Media, 2026). Marc Henrichmann, chair of the Bundestag’s parliamentary oversight committee, described the choice as “a clear signal for European digital sovereignty” (Modern Ghana, 2026). In parallel, the Bundeswehr is testing three alternatives, Almato (Stuttgart), Orcrist (Berlin) and ChapsVision (Paris), with a contract decision expected by the end of 2026 (Xpert, 2026).

These are not yet operational successes, but they answer, empirically, the most powerful objection raised against the sceptical case, that no European alternative exists. They also reveal what proper procurement looks like: a published competitive evaluation against operational requirements, with architectural sovereignty treated as a binding constraint rather than a contractual flourish.

Capability and interoperability are real, and limited

A fair reading of the pro-adoption case acknowledges three serious arguments. First, capability speed: NATO’s six-month MSS procurement contrasts with alliance procurement timelines typically measured in years. Second, battle validation: Palantir’s role in Ukrainian targeting since 2022, and the Pentagon’s disclosure that Operation Epic Fury against Iran struck 13,000 targets in 38 days using the Maven Smart System (Breaking Defense, 2026), establish operational credibility under fire. Third, interoperability: European forces operating under NATO command will interact with MSS NATO regardless of national procurement choices.

Each of these arguments has limits the procurement debate has not yet absorbed. Speed of delivery and slowness of exit are two sides of the same architectural coin; the same integration that lets capability stand up in six months is what makes it impossible to stand down in twelve. Wartime validation in Ukraine reflects survival pressure, not peacetime trade-off optimisation; the relevant comparator for European procurement is not the Ukrainian battlefield in 2022 but the Swiss procurement office in 2024. And interoperability with NATO does not require national operational deployment of Palantir, as Daum’s own NATO/national distinction, articulated to Handelsblatt, demonstrates with precision (Ground News, 2026).

What follows

The case for refusal of terms, rather than refusal of technology, can be made narrow and disciplinable. Six contractual requirements, each with a European precedent already in practice, would preserve strategic optionality without forfeiting access to the capability. First, mandatory competitive tender or formal published justification for direct awards under national-security exemptions. The Bundeswehr’s parallel testing of Almato, Orcrist and ChapsVision is the template. Second, ontology and data portability clauses with penalty provisions, i.e. the lesson the NYPD documented in 2017 should not need to be relearned every five years. Third, source-code escrow with a European trustee, sovereign technical inspection rights, and independent third-party security audit. Fourth, architectural CLOUD Act mitigation through air-gapped or genuinely sovereign deployment, with no foreign-national administrator access to production data, the BfV/ArgonOS model. Fifth, sunset clauses limiting contracts to three to five years, with mandatory re-competition; ten-year enterprise agreements modelled on the US Army’s $10 billion ESA template should be presumptively rejected for European customers, because they explicitly trade competition for volume discounts. Sixth, a second-source obligation requiring that at least one critical national workflow run in parallel on a non-Palantir system.

None of these requirements is technologically unreasonable. Each is operationally attainable. Each has been implemented somewhere in Europe in the past eighteen months. The reason they are not standard practice in the UK, Spanish, Polish, and NATO contracts already signed is procurement convention, not technical constraint. That is the diagnosis. The prescription is to make these terms non-negotiable for every new Palantir contract and every renewal across the European theatre, and to accept the consequence that some contracts will take longer to sign, and a few will not be signed at all.

That is not a high price for the future ability to choose. Vice Admiral Daum’s position, far from being an outlier, will be remembered as the moment one European government stated the proposition plainly and acted on it. The remaining question is how many others do the same before the choice set has narrowed too far for the question to matter.