Europe’s pursuit of sovereign artificial intelligence is producing a financing challenge as demanding as the technology itself, with German infrastructure company Polarise securing a high double-digit-million-euro debt package from SWI Stoneweg Icona Group to expand its AI cloud platform.

The financing replaces an earlier plan under which SWI had proposed taking a majority equity stake in Polarise. That transaction will not proceed, but the two companies have retained a financial relationship through debt funding, giving Polarise additional capital without the ownership change originally envisaged.

Polarise already provides GPU and AI compute through sites in Oslo and Munich, and it is planning additional facilities including an AI factory in Amberg, Bavaria. The company now says the Amberg site is intended to begin at 65MW with scope to scale to 120MW, although earlier planning material had described a smaller first phase.

The change in financing structure is important because AI infrastructure is expensive long before revenue becomes predictable. Buildings, electrical systems, cooling, accelerators, networking, land, and grid connections all require large upfront commitments, while customer demand and chip availability can change quickly.

Sovereignty acquires a balance sheet

European governments and companies are spending more time on where AI workloads run, who controls the infrastructure, and which legal jurisdictions apply to data and operations. Those concerns have created demand for domestically operated or Europe-controlled compute, but sovereignty does not make the economics of data centres any easier.

Polarise is trying to build around that demand with a distributed model rather than relying only on one very large campus. Its Munich and Oslo capacity, together with planned expansion in Bavaria, gives customers the option of keeping workloads within European infrastructure while buying GPU capacity as a service.

The company has also positioned itself as part of the broader European AI-factory push, where governments and private operators are attempting to increase access to high-performance compute without depending entirely on US hyperscalers. Capacity remains the central constraint: the most capable accelerators are expensive, power-hungry, and in high demand.

Debt financing can support that expansion without diluting existing shareholders, but it adds its own discipline. Interest, covenants, and repayment schedules place pressure on utilisation, meaning new facilities need customers quickly enough to support fixed infrastructure costs.

Modular deployment tackles only part of the constraint

Polarise argues that modular infrastructure can be built and expanded in smaller increments, reducing some of the risk attached to committing immediately to a giant campus. That can make capacity planning more responsive to demand, although power and grid availability remain outside the operator’s direct control.

Amberg illustrates the scale of the issue. Even at 65MW, the first phase would represent a substantial electricity load, while expansion towards 120MW would move the project into the territory of major hyperscale facilities. Securing energy supply, grid reinforcement, planning permission, and cooling capacity can take as long as procuring the servers themselves.

The Munich market has already shown how quickly new sovereign capacity can be absorbed. Deutsche Telekom’s Munich AI cloud has already reached full capacity, reinforcing the argument that European demand is not purely political but operational.

Yet demand alone does not guarantee attractive returns. AI hardware depreciates rapidly, model architectures change, and customers may move between private infrastructure, hyperscalers, and specialist providers as prices and performance shift. Operators therefore have to balance long-lived buildings and power systems against short technology cycles.

Polarise’s new debt package gives it additional room to build through that uncertainty, while the replacement of the proposed majority investment removes one layer of corporate complexity. The next test is whether sovereign-AI demand can keep new European capacity sufficiently utilised to support the capital structures now being built around it.