NVIDIA Tests Germany’s Retired Nuclear Sites as AI Factories

World Nuclear News reported that Germany’s PreussenElektra is exploring possible uses for former nuclear power plant sites in collaboration with NVIDIA. The dry wording matters. This is not a normal hunt for warehouses with fibre and cheap land. It is a search inside the remains of an energy system that a country chose to shut down.

NVIDIA aims to evaluate the technical requirements for AI factories on the decommissioned nuclear sites. That phrase, AI factories, is doing heavy work. It recasts computing as heavy industry, with power, cooling, permitting and local consent sitting alongside chips and software. The AI Infrastructure story is moving out of the cloud metaphor and into the fenced industrial estate.

The Site Becomes the Strategy

PreussenElektra is a subsidiary of EOn Group and is responsible for decommissioning eight nuclear power plants in Germany. All of PreussenElektra's nuclear power plants have been decommissioned and are in various stages of dismantling. That makes the company an unlikely landlord for the next wave of computing. It also makes it an unusually relevant one.

A retired nuclear site is not an empty field. It carries grid history, security habits, industrial zoning and a local workforce memory. Some of that may help. Some of it may hurt. The political point is sharper than the engineering point: a place built for baseload electricity could be asked to host machines that consume electricity at industrial scale.

Germany’s nuclear exit left behind physical assets and social arguments. The assets are easier to catalogue than the arguments. The collaboration places both back on the table without reopening the same debate in the same language. The subject is no longer whether reactors should run. The question becomes what kind of productive economy can occupy the land after they stop.

The former Biblis nuclear power plant site has been identified for the establishment of a laser-based nuclear fusion power plant by 2035. That example shows how these locations can become platforms for future energy narratives rather than memorials to past policy. It also shows the competition for credible industrial reuse. Computing is only one claimant.

There is a quiet irony here. The digital economy spent years presenting itself as weightless. Its next phase is visibly material. It needs land, transformers, cooling systems, permits, power contracts and communities willing to host large machines. The former nuclear site gives that materiality a name and a fence line.

Germany’s Exit Meets the Compute Boom

PreussenElektra has decommissioned eight nuclear power plants in Germany. The count matters because this is not a single brownfield curiosity. It suggests a portfolio of politically sensitive industrial sites moving through dismantling, each with a different local balance of constraint and possibility.

A decommissioned plant site can look like a problem for one ministry and an opportunity for another. Energy policy sees closure, waste handling and grid legacy. Industrial policy sees scarce serviced land. Digital policy sees the need for compute capacity. Local politics sees traffic, jobs, noise, tax base and suspicion. The same perimeter fence can contain all of those readings.

NVIDIA’s role also changes the feel of the story. This is not simply a utility looking for tenants. It is a leading computing infrastructure actor looking at places shaped by the old power system. That puts the bottleneck in a different place. The issue is not only chip supply or model demand. It is whether societies can build the physical estate that advanced computing now requires.

World Nuclear News reported that NVIDIA aims to evaluate the technical requirements for AI factories on the decommissioned nuclear sites. Technical requirements sound neutral. In practice, they are where policy becomes concrete. Power quality, redundancy, cooling, land remediation and connection capacity can turn a bold industrial slogan into a spreadsheet with hard vetoes.

The phrase also leaves room for disappointment. Many sites that look attractive in principle fail when the engineering survey begins. Others work technically but fail politically. A former nuclear location may be easier to secure than a new suburban site, yet harder to explain. The old stigma does not vanish because the tenant changes.

Nuclear Land Without Nuclear Power

The broader pattern is not confined to Germany. U.S. electric utilities operate approximately 98 gigawatts of nuclear generating capacity as of the time of writing. That figure describes an energy system where nuclear plants remain part of the operating base. It also provides a contrast with Germany’s decommissioning story.

Small modular reactors have a capacity of about 300 megawatts per unit or less. Microreactors generally have a capacity of 20 megawatts or less. Those scales sit far below the traditional image of a large nuclear station. They also explain why industrial users keep appearing in the conversation. Smaller units invite talk of campuses, bases and dedicated facilities, even where commercial deployment remains subject to difficult scrutiny.

The Department of the Army selected nine bases to potentially site microreactors as part of their Janus Program. That fact puts defence infrastructure into the same mental map as data infrastructure. Both involve critical loads, secure perimeters and a desire for resilient power. The analogy should not be stretched too far, but it is useful. High-compute facilities are becoming strategic assets, and strategic assets often pull energy planning closer to the customer.

In March 2025, the U.S. Department of Energy reissued a tender for $900 million to promote small modular reactor development. Public finance is part of the nuclear revival argument. It is also part of the computing buildout argument, even when the subsidy channels differ. The state may not always build the factory, but it often shapes the conditions under which the factory can be built.

The German case is distinct because the proposed sites no longer host operating nuclear power. That distinction is the story’s central tension. A country can reject one technology while preserving the industrial geography that technology created. The land remains useful because the energy system left marks on it.

Capital Costs Are Part of the Politics

The financing backdrop is not friendly to grand industrial plans. The fed funds rate was 3.75% in September 2026. The 10-year Treasury yield was 4.98% on 29 October 2026. The 10-year real yield was 2.63% on 29 October 2026. These are not site-specific figures, but they shape the cost of long-lived infrastructure.

A data centre on a former nuclear site is not a software launch. It is a capital project with civil works, electrical systems and a long permitting tail. Higher real yields make patience expensive. They also make reuse more attractive if it reduces uncertainty. Brownfield logic gains force when new-build assumptions become costly.

The 10-year breakeven inflation rate was 2.34% on 30 October 2026. The broad trade-weighted dollar index was 121.32 on 23 October 2026. Such macro conditions affect imported equipment, financing models and cross-border capital discipline. They do not decide whether a German nuclear site becomes an AI campus. They help decide how hard every assumption will be questioned.

This is where the story stops being a novelty item. The buildout of compute capacity is now colliding with the same constraints that govern ports, factories, grids and energy projects. It must fit into balance sheets. It must survive local review. It must find power without poisoning the politics around power. The cloud has acquired a planning department.

The U.S. CPI index was 336.3 in September 2026. U.S. M2 money supply was $23,413.9 billion in September 2026. Those measures belong to the wider monetary environment rather than the German site question. They are still part of the atmosphere in which infrastructure investors decide how much complexity they will tolerate.

The Industrial Afterlife of the Reactor

The former Biblis nuclear power plant site in Hesse has been identified for a laser-based nuclear fusion power plant to be established by 2035. That site now carries a different kind of future-facing promise. Fusion and AI factories are separate propositions, but they share a political need: both must persuade communities that advanced technology belongs on inherited industrial land.

There is a lesson in that pairing. Decommissioning does not end the argument over a site. It starts a new one. A place that once symbolised centralised power can be recoded as clean technology, national competitiveness, digital capacity or local burden. The winning label will depend less on slogans than on execution.

World Nuclear News reported that Germany’s PreussenElektra is exploring possible uses for former nuclear power plant sites in collaboration with NVIDIA. The word exploring is doing honest work. It keeps the story below the level of certainty. There is no need to treat a study as a revolution. The more interesting point is that the study is thinkable at all.

That shift says something about the next industrial cycle. Digital infrastructure is no longer just leasing space near users and fibre routes. It is entering the politics of energy transition, brownfield reuse and strategic capacity. The owners of old power assets may find themselves courted by the builders of new computing systems. The terms of that courtship will be technical, financial and deeply local.

The dry sceptical view is simple. A decommissioned nuclear site is not a magic socket. It may be constrained, contested and expensive. The bolder view is also simple. The sites that powered the old industrial economy may be among the few places with the scale and seriousness to host the next one.

What to Watch Now

What changed: World Nuclear News reported that Germany’s PreussenElektra is exploring possible uses for former nuclear power plant sites in collaboration with NVIDIA. The collaboration moves AI buildout into the terrain of energy legacy and industrial reuse.

Measurable implication: PreussenElektra is responsible for decommissioning eight nuclear power plants in Germany. That portfolio gives the concept more than a single test location. It turns site reuse into a repeatable question, even if each answer remains local.

Next dated milestone: The former Biblis nuclear power plant site has been identified for the establishment of a laser-based nuclear fusion power plant by 2035. That deadline will help show whether former nuclear locations can host credible next-generation industrial projects.

Strongest counterargument: All of PreussenElektra's nuclear power plants have been decommissioned and are in various stages of dismantling. Dismantling is not the same as readiness. The hardest work may sit between an attractive map and a buildable site.