Berlin swaps one integrated Palantir suite for a field of 30 candidate AI tools

TNW's account of Ukraine puts the problem plainly: the volume of imagery coming off drones and satellites has grown far beyond what analysts can process manually ^[8], which is the strongest available argument for buying whatever is fielded today. The counter-argument in the same reporting is that control of the software layer carries leverage, especially when the software sits next to intelligence and targeting decisions ^[9].

The shape of the buy is what decides who pays for that leverage. One of Palantir's selling points is that its capabilities sit inside a single system ^[11], which means the vendor owns the seams. Assemble a field instead and the seams are the buyer's, and TNW notes that military software programmes often come apart at exactly that stage rather than at the demo ^[11].

Daum's own dates size the wait. About 18 months to a prototype, then roughly another year to full deployment if the programme holds ^[7], is about 30 months of calendar ^[19]. Maven, meanwhile, is already deployed ^[7] with the Pentagon and several NATO members ^[6].

The programme sits inside the cyber command, which TNW reads as Berlin treating AI as part of the military's data and network infrastructure before treating it as a targeting system ^[12]. That choice shows up in the question being asked first: which tools can process the data, rather than which system a commander would actually rely on in the field ^[13]. Throughput is straightforward to score in year one, while reliance shows up only once a decision depends on the output, a dimension early evaluations don't measure.

The supplier base is one clear beneficiary. Germany took 90 percent of Europe's record defence tech funding last year ^[14], leaving a tenth for everyone else ^[20], and its domestic supplier base has grown significantly since 2022 ^[15], with startups such as Comand AI raising money to build command software for European forces ^[16]. A procurement of this shape turns that money into software a military can actually use, which is a real objective, and a different one from handing an officer something usable in 2027.

The aversion extends beyond American vendors. After France and Germany pledged to build a European alternative to Palantir, German officials went on to warn that the French-led project could create another dependency ^[10]. A domestically weighted field is what declining both dependencies looks like, and it explains why no single European champion is being asked to carry the load.

The reporting is silent on what the Bundeswehr runs for decision support between now and the prototype, or whether any one candidate covers Maven's scope ^[21]. So the cost of substitution here is measurable as a gap against armies operating a deployed system, not as a void.

The portable test is a pair of written commitments. First, the date functional parity is expected, which Berlin has actually supplied: a prototype in Q2 2027, deployment in 2028 ^[2]. Second, the named team that owns the seams, with budget, before the shortlist is cut. TNW sets out what happens when only the first exists: 2028 arrives with a collection of capable tools that do not form a useful decision-support system, and the case for buying the American one gets made simply by pointing at the calendar ^[18].