The numbers tell a story that European policymakers are beginning to find uncomfortable. Over eighty per cent of AI compute consumed in Europe runs on infrastructure controlled by three American hyperscalers. Every inference call, every training run, every enterprise AI deployment that touches AWS, Azure, or GCP sends data through architectures governed by US law — subject to the CLOUD Act, shaped by Silicon Valley’s incentive structures, and priced in dollars. For sectors where data control is non-negotiable — defence, healthcare, critical infrastructure, financial services — this dependency is not a market inconvenience. It is a strategic vulnerability.
Victor Caduc saw this clearly before most. A French engineer with a background in international operations — he previously led the South African subsidiary of a French industrial group, managing complex projects and scaling operations in demanding environments — he brings an operator’s instinct to a problem that is usually discussed in the language of policy. Sovereignty is not a white paper. It is a deployment decision. And deployment decisions require infrastructure that does not yet exist in Europe at the scale and form factor the moment demands.
The Micro Factory Thesis
Proxima’s proposition is architecturally distinct from the hyperscale model. Where AWS and Azure build warehouse-scale data centres — consuming hundreds of megawatts, requiring years of planning permission, and concentrating compute in a handful of locations — Proxima builds what Caduc calls micro AI factories: compact, immersion-cooled compute units designed to be embedded directly in the communities they serve.
The design is deliberately sober. Each unit is immersion-cooled, eliminating the massive air-conditioning systems that make conventional data centres energy-intensive and architecturally imposing. The waste heat — which in a hyperscale facility is simply vented into the atmosphere — is captured and returned to the local community as a resource: heating for buildings, swimming pools, greenhouses, industrial processes. Compute becomes not just a service but a neighbour — useful through the energy it returns, not just the calculations it performs.
This is not green marketing. It is an engineering decision with strategic consequences. Distributed micro factories can be deployed faster than hyperscale facilities because they do not require the same planning approvals, grid connections, or water supplies. They can be placed closer to the data sources — a hospital, a manufacturing plant, a government agency — reducing latency and keeping sensitive data within physical as well as legal boundaries. And they can run open-source models, decoupling European AI deployment from the proprietary model ecosystems that create a second layer of dependency beyond the infrastructure itself.
The Operator’s Advantage
Caduc’s background is unusual for a deeptech CEO. He did not come from a cloud computing or AI research lab. He came from industrial operations — managing a subsidiary in South Africa, running complex projects in environments where logistics, regulation, and local conditions are not abstractions but daily constraints. This operational grounding shapes Proxima’s approach in ways that distinguish it from the typical European sovereign-tech pitch.
Most European sovereign AI initiatives start with policy and work down toward implementation. Caduc starts with deployment and works up toward strategy. The question is not “what does sovereignty mean in theory?” but “what does it take to put sovereign compute in a specific building, in a specific city, serving a specific customer, this quarter?” The gap between these two questions is where most European sovereignty initiatives stall. It is also where Proxima operates.
The ambition is explicit: to make Europe a global leader in trusted, sovereign, and sustainable AI infrastructure. Not by building a European AWS — the hyperscale race is already lost, and attempting to win it would mean competing on terrain optimised for incumbents — but by building a different kind of infrastructure entirely. One that is distributed rather than centralised, energy-positive rather than energy-consuming, and sovereign by architecture rather than by policy declaration.
Why Vienna
The Human × AI Conference takes place in a city — and a country — that sits at the centre of the European sovereignty question. Austria occupies a unique position in the European AI landscape: small enough to be agile, wealthy enough to invest, strategically positioned between Western Europe’s regulatory frameworks and Central Europe’s emerging tech ecosystems. Vienna itself is a ViennaUP hub for exactly the kind of cross-border, sovereignty-conscious technology deployment that Proxima represents.
Caduc’s session will be concrete. Not a policy argument for sovereignty — that argument is already won in Brussels — but an operational account of what it takes to build sovereign AI infrastructure from the ground up. How immersion cooling changes the economics and geography of AI deployment. Why open-source models are a sovereignty requirement, not a cost-saving measure. What happens when compute becomes a local resource rather than a remote service. And why the next generation of European AI leadership will be built by operators, not just by policymakers.
Implications
- For infrastructure strategists: The hyperscale model is not the only model. Distributed, immersion-cooled micro factories offer a deployment pathway that is faster, more energy-efficient, and sovereign by design — particularly for regulated sectors where data locality and control are non-negotiable. The question is no longer whether Europe needs its own AI infrastructure but what form factor that infrastructure should take.
- For European founders: Sovereign compute is not just a government concern — it is a competitive differentiator. Startups building AI products for regulated European customers can offer data residency guarantees, latency advantages, and compliance assurances that hyperscaler-dependent competitors cannot match. The infrastructure layer is becoming a go-to-market argument.
- For conference attendees: Expect an operator’s perspective on Europe’s most debated AI question. Caduc brings engineering pragmatism and international operational experience to a conversation that often stays at the policy level — with a concrete model for how sovereign, sustainable AI infrastructure can be deployed at the community scale.
Victor Caduc joins Human × AI on May 19, 2026, in Vienna.