A Serious European Bet on AI Compute
VOLT's proposed Rotterdam AI campus would be among Europe's largest, if built. The real scale, and why Dutch soil isn't the same as durable Dutch control.
Michael Dell posted a message last week about sovereign AI infrastructure “built in the Netherlands, run on Dutch soil”. It read almost as though a ribbon had already been cut.
The announcement is more modest, but also more interesting. VOLT plans to open a Dutch AI Cloud at a NorthC data centre in Amsterdam in October 2026, using infrastructure supplied by Dell Technologies. It presents this as the first operational step towards something much larger: an AI gigafactory planned for the Port of Rotterdam, eventually intended to accommodate roughly 250,000 GPUs and hundreds of megawatts of computing capacity.
Calling that merely a data centre undersells what is being proposed.
A different kind of facility
Conventional data centres already do much more than store data. They host cloud services, databases, networks and all kinds of business computing. But most were not designed around the extraordinary density and internal connectivity required by large AI clusters.
An AI factory concentrates far more computing power into each rack. A rack used for intensive AI workloads can draw 100 kilowatts or more, compared with perhaps 5 to 15 kilowatts in a traditional enterprise environment. That changes almost everything: power distribution, cooling, network architecture and the relationship with the electricity grid.
These facilities are therefore becoming something closer to industrial plants. Their product is not steel, fuel or chemicals, but usable computation.
The actual scale
VOLT has described a Rotterdam campus with an eventual capacity of around 800 megawatts and space for approximately 250,000 GPUs. Those are ambition figures, not installed capacity. The first Amsterdam facility will be much smaller, and the Rotterdam development still depends on land, power, permits, finance, customers and several years of construction.
Even so, an 800 MW AI campus would be exceptional by European standards. The planned Sines data-centre campus in Portugal has secured 1.2 GW of grid capacity, although only part of that capacity has so far been committed to specific AI deployments. Several other large European projects are also moving through various stages of announcement, permitting and construction.
Set against the largest American programmes, Rotterdam would be substantial rather than dominant. Projects associated with xAI, Microsoft, OpenAI and other hyperscalers are already being discussed at multi-gigawatt scale.
The more important comparison is continental. Estimates vary according to what exactly is counted, but Europe possesses only a small share of the world’s leading AI-compute capacity. The United States accounts for roughly three quarters of the measured performance of major AI clusters, with China a distant second and the European Union at around 5 per cent.
Rutger Bregman recently argued that this imbalance should concern Europeans almost more than any other political issue. Rotterdam would not close that gap. It is one proposed facility against a continental shortfall. But that share can only change if facilities are actually built, and Europe has spent years building relatively few of them.
Who owns the switch?
Bregman’s argument is that compute is power. Whoever controls it gains economic leverage that may eventually resemble political power. That makes ownership worth examining rather than treating “on Dutch soil” as a complete answer.
VOLT is a private Dutch company founded by Han de Groot. His family office is financing the first phase of the Dutch AI Cloud. Dell supplies the technical infrastructure, while NorthC will host the first AI factory in Amsterdam. Public announcements do not disclose VOLT’s full ownership structure, nor the eventual financing and ownership arrangements for the Rotterdam campus.
There is consequently a difference between Dutch location, Dutch operation and durable Dutch control.
NorthC illustrates why that distinction matters. The data-centre operator has changed hands between large infrastructure investors in recent years. Such transactions are normal, but they show that the nationality of digital infrastructure is not fixed merely because the buildings remain in the same country.
The Netherlands has screened investments in vital providers and sensitive technologies under the Vifo Act since 2023. From 1 January 2027, designated forms of artificial intelligence will also enter its sensitive-technology regime. Whether and how that framework would apply to the sale of a large AI-infrastructure operator will depend on the company’s activities and the eventual legal structure. The regulatory position is evolving, but it is not yet the same thing as a permanent guarantee of Dutch ownership.
Nor does ownership alone settle the sovereignty question. The chips are American, the core systems come from Dell and NVIDIA, and many potential customers and software platforms will be international. Sovereignty here can only be relative: greater jurisdictional and operational control, not complete technological independence.
France enters this race from an unusually strong position. Its nuclear fleet gives it access to large amounts of relatively affordable, low-carbon electricity, precisely the resource that AI infrastructure increasingly depends on. But this also raises a European question: will compute capacity remain a matter of national industrial strategy, with France, Germany and the Netherlands each building their own position, or will Europe eventually develop something closer to a compute union? I have added a recent Politico article on France’s energy advantage below.
Build first, govern seriously
None of this negates the significance of the announcement.
Uncertainty is normal when a new class of infrastructure is built ahead of the rules that will eventually govern it. Railways, electricity grids and the fibre routes on which the internet still depends were all developed before societies had fully decided who could own them, how access would be regulated or under what conditions they could be sold.
European AI infrastructure is entering a similar period. The answer cannot be to call every facility sovereign merely because it stands on European soil. But neither can Europe wait until every ownership and governance question has been resolved before it begins building.
The capacity proposed for Rotterdam does not exist anywhere in the Netherlands today. Every credible phase that is financed, connected and put into operation makes the European compute deficit slightly less severe.
That is reason for qualified optimism: welcome the capacity, examine the ownership, and build the legal framework with the same seriousness as the machines.