Warsaw-based Expeditions, backed by BAE Systems and the NATO Innovation Fund, has closed a €197 million fund to fuel the next wave of European defence startups. The number itself marks an acceleration: BAE had already committed €50 million to outside VCs, and venture capital is pouring into the continent’s defence tech at record speed.

But the figure is just the surface. For those watching AI deployment in sensitive contexts, the real signal points toward infrastructure that is increasingly local, isolated, and far from the public cloud. This is not abstract speculation. Defence applications — from autonomous surveillance to signal analysis, predictive logistics, and decision-support weapon systems — rely on LLMs and models that process classified data. Moving them to the cloud violates security requirements by design, while NATO regulations and national doctrines impose air-gapping and physical sovereignty.

The winners here are manufacturers of specialized hardware (GPUs and low-latency inference accelerators), orchestration platforms designed for disconnected environments, and integrators that can build certified on-premise stacks. The losers are cloud providers banking on defence as a growth vertical: long-term contracts will shift toward bare metal and self-hosted solutions, where hardware control is non-negotiable. Even the general-purpose chip market faces selective pressure: defence demands accelerators with verifiable supply chains and tamper resistance, not just raw teraflops.

This connects directly to the AI-RADAR ecosystem. For those evaluating on-premise deployment, patient capital in defence acts as an amplifier for the entire supply chain: public procurement volumes drive down hardware costs, accelerate the development of serving frameworks for constrained environments, and force software houses to test quantization and optimizations on real stacks rather than synthetic benchmarks. It is no coincidence that BAE, one of the most conservative prime contractors, chooses to invest indirectly through a venture vehicle: it wants privileged access to technologies that will run inside its own hardened data centers, bypassing external roadmaps.

The Expeditions fund, therefore, is not just a financial indicator. It proves that the AI defence game is fought on the terrain of proprietary hardware and local deployment, with TCO a secondary concern after compliance. This upends established hierarchies: capital expenditure (CapEx) returns to center stage, while the consumption-based operating model (OpEx) retreats before secrecy constraints. For technology decision-makers, the reminder is stark: preparing for scenarios where inference can never leave the physical perimeter is not an option — it is the starting point.