The geography of European tech is being reshaped with record figures. British hyperscaler Nscale secured a £670 million credit facility, while Germany's Proxima Fusion closed the largest European investment ever in nuclear fusion with a €411 million round. These are just two deals in a week that saw over 70 transactions totaling more than €2.8 billion, marking a venture capital rebound tracked by Invest Europe.

Where is all this capital heading? For those observing the world of on-premise and local deployments for Large Language Models and generative AI, the signals are clear. The infrastructure race isn't just about chips. The energy question becomes central: powering GPU clusters for training and inference requires increasing amounts of electricity, and the promise of fusion—still long-term—is to provide abundant, stable, and low-impact energy. If and when it reaches commercial maturity, the energy cost, a hefty component of the Total Cost of Ownership (TCO) of an on-premise data center, could plummet. Proxima Fusion, with its €411 million, is betting precisely on this scenario, and the presence of European institutional investors signals a desire to avoid dependence on external energy supply chains.

Nscale, for its part, operates on a different but complementary front: offering hyperscalable compute capacity, including through private cloud hosted or dedicated environments. For companies weighing whether to keep inference in-house or rely on the public cloud, the existence of European players like Nscale introduces intermediate options that can meet data sovereignty and GDPR compliance needs, while reducing latency. It's no coincidence that banks, defense, and healthcare are accelerating trials of self-hosted LLMs: physical control over data remains a strong negotiating requirement.

The same week brought developments in robotics and edge AI. WaiV Robotics is solving one of the biggest challenges of maritime drones, likely related to autonomous decision-making and intermittent connectivity—a typical scenario where inference models must run locally on embedded hardware, without cloud access. Similarly, the startup digitizing battlefield medicine works on portable devices that must operate in hostile, disconnected environments, imposing a fully on-premise, often air-gapped, deployment.

These examples, along with the new Expeditions fund (backed by BAE) for European defense startups and the €2.4 million raised by Porelio for industrial water treatment, show a clear trajectory: AI applied to regulated and infrastructure sectors pushes toward distributed, local, and resilient architectures. This is not a nostalgic return to the server under the desk, but a choice driven by latency, security, and TCO. In this new landscape, energy cost and low-latency compute availability become competitive levers: those who can combine clean sources, efficient hardware, and local inference frameworks will have a strategic edge.

The venture capital rebound reported by Invest Europe is not just a macroeconomic data point: it indicates that investors are betting on a continental tech ecosystem capable of sustaining the entire AI value chain, from silicon to application. For teams evaluating on-premise deployments, AI-RADAR offers analytical tools to navigate these trade-offs, but the direction of travel seems clear: infrastructural autonomy will be the theme of the coming years.