Europe wants its own defence AI giants, but the stack is the hard part

An ICDS paper argues Europe needs home-grown AI 'neo-primes' to match Palantir and Anduril, and that autonomy means owning the whole stack, not just apps.

A new paper from the International Centre for Defence and Security, The Neo-Prime Time, makes a blunt case: AI may be the next military technology that separates winners from laggards, and Europe is at risk of ending up in the second group. The author's answer is to grow European 'neo-primes', AI-first defence firms in the mould of Palantir and Anduril, rather than leaving the field to American vendors.

The urgency is partly political. The paper notes that the Dutch defence state secretary has asked for a fully operational alternative to Palantir, and that a UK parliamentary report calls such dependence an unacceptable weakness. Then came the June 2026 restrictions on foreign access to advanced US models, which the paper presents as proof that frontier AI can become a strategic vulnerability overnight. Politico's coverage of the suspension gives the background.

The neo-primes sell at the top of the stack but depend on the two layers beneath it. Image: Unsplash — gray and black SUV

What we found most useful is the framing around the three-layer AI stack: infrastructure and data, models, and applications. Neo-primes mostly live in the top layer, yet autonomy means controlling or reliably accessing all three. Anduril's tie-up with OpenAI shows the US model of reaching down the stack. The EU's counterpart is the EuroStack idea, which the paper sets against American cloud providers holding around 70% of the European market and Europe's roughly 9% share of global chip manufacturing.

There are real European contenders. Germany's Helsing is building the CA-1 Europa around its Centaur software and has partnered with France's Mistral, as both companies announced. French and German agencies have picked ChapsVision's ArgonOS over Palantir, per Politico. The paper reads this as evidence that European firms are moving from software towards systems integration, a job once reserved for the traditional primes.

Whether any of this scales depends on money, and the excerpt we have covers only the start of that argument: defence has been reclassified as compatible with sustainable finance, the European Investment Bank is now lending into the sector, and the EU budget framework is the next signal to watch. The financing section is cut off in the material we received, so we will not guess at its conclusions.

Our take: partnerships between a defence start-up and a model lab are a good sign, but they are the easy part. The hard part is the unglamorous bottom layer, the data centres and chips, where Europe is weakest and where a single regulatory decision elsewhere can turn off a capability. Building a European Anduril is a funding problem. Building the ground it stands on is a decade-long industrial one.

Opinion · The number that worries me

The figure that sticks with me is the 70%: roughly the share of the European cloud market held by American providers, according to the paper. Everything else in the neo-prime story, the drones, the targeting software, the partnerships, runs on top of that layer.

You can build a sovereign application and still have someone else hold the switch underneath it. The June 2026 restrictions on foreign access to US models made that risk concrete rather than theoretical, and 70% is a lot of switch to hand over.

Related articles from LaRebelión

Wikimedia says unapproved OpenAI agents edited wikis and hammered its APIsMistral unveils Le Chonk, a large open-weight model you can run on your own servers

Original source: International Centre for Defence and Security

Produced with AI support and reviewed by the newsroom

Byline · Chief editor · English edition · London

“Europe can fund a Helsing, but it cannot fund its way out of a cloud market it does not own.”