NVIDIA and PreussenElektra, the German company responsible for dismantling nuclear power plants, will jointly study whether the sites of decommissioned nuclear plants could be converted into AI data centers, or "AI factories." PreussenElektra announced the plan on its official LinkedIn page in October 2026. The idea is to make use of large industrial sites and existing power infrastructure, but no specific location or investment amount has been decided yet.
Reusing former nuclear plants is drawing attention as a way to secure the land and infrastructure needed to build AI data centers. However, existing substations and cooling facilities cannot necessarily be used as they are. The property brochure for the Grafenrheinfeld nuclear plant, published by PreussenElektra, shows a wide gap between the presence of equipment and its actual usability.
NVIDIA and PreussenElektra to study converting former nuclear plants into AI factories
In its LinkedIn post, PreussenElektra said it will work with NVIDIA to investigate whether the sites of nuclear plants under its management are suited to AI factories. The company, a subsidiary of E.ON, is responsible for dismantling eight nuclear power plants in Germany, and the assessment of their suitability for AI data centers will be added to its ongoing efforts to reuse the sites.
The joint study will examine existing power infrastructure and land availability and compare them against the technical and operational requirements of AI data centers. According to the company's explanation as reported by German public broadcaster NDR, the study does not currently involve any specific investment or choice of construction site. The fact that eight plants are being dismantled does not mean there are plans to build AI facilities at all of them.
The initiative concerns reusing the land and infrastructure of nuclear plants that have ceased operation. It is not an announcement of restarting reactors or of supplying AI facilities with nuclear power. No concrete plans have been disclosed for the size of the data centers, the GPUs to be installed, or the power required.
A neighboring substation does not guarantee power supply
The Grafenrheinfeld site offers a useful reference for considering how former nuclear plants might be converted into AI data centers. According to the official property brochure published by PreussenElektra (PDF), 110 kV and 380 kV substations are adjacent to the site. However, there is currently no active grid connection, and a separate connection procedure would be required for a new facility to use power.
The brochure also describes conditions for land and cooling facilities that matter when considering reuse.
| Item | Current status | What must be confirmed for AI data center conversion |
|---|---|---|
| Power infrastructure | 110 kV and 380 kV substations are adjacent, but there is no active grid connection | Grid connection application, available power capacity, start date of supply |
| Land | 31 ha owned, 26 ha total usable area. Reuse of the commercial site expected from 2037 onward | Which parcels are actually usable, when construction is possible, regulatory conditions |
| Cooling facilities | Two cooling towers demolished in 2024. Cooling water intake and discharge channels remain | Whether existing channels can be reused, cooling capacity, required permits |
The key point is that having a substation next door is not the same as a data center being able to secure the power it needs. 110 kV and 380 kV indicate the voltage of transmission equipment, not how many megawatts an AI facility can receive. Actual receiving capacity and the start date of supply must be determined based on the state of the grid and the connection process.
There are also time constraints on land use. At Grafenrheinfeld, conventional demolition of buildings is expected to continue until the end of 2036, and full-scale reuse of the commercial site is said to come in 2037 or later, after the land is released from nuclear regulatory oversight. Some land may become usable earlier, but not all 26 ha listed in the brochure can be used for construction right away.
No announcement has been made that Grafenrheinfeld has been chosen as a candidate site for the joint study with NVIDIA. Nor are other sites managed by PreussenElektra necessarily in the same situation. The brochure should be taken as a concrete example of the technical and institutional challenges involved in reusing former nuclear plants.
Cooling towers are gone; can the existing cooling water channels be used?
Operating an AI data center requires not only large amounts of electricity but also a cooling system to remove the heat generated by GPUs and other computing equipment. Former nuclear sites may retain cooling-related facilities used by the power plant, but whether they suit a new purpose must be verified separately.
At Grafenrheinfeld, no nuclear fuel has been present on site since December 2020, and two cooling towers were demolished by blasting in 2024. The cooling water intake and discharge channels, on the other hand, remain unfilled. These channels are notable equipment for considering reuse of the site, but their existence alone does not mean they can be used to cool a data center. Required flow rates, water quality, and cooling performance must be confirmed, along with permits related to water use.
The same goes for communications infrastructure. The property brochure lists a figure of up to 10 Gbit/s for the fiber-optic line for wide-area communications. But the bandwidth an AI data center needs varies with the scale and mode of operation, so this figure alone does not show whether it is sufficient. If the site were used as a large-scale AI computing hub, communications infrastructure might need to be expanded to meet the facility's requirements.
Thus, while former nuclear sites retain equipment that may be reusable, some equipment has already been removed and some would need modification or permits for a new use. Whether using existing infrastructure can reduce construction costs and timelines can only be evaluated after examining the condition of each facility.
Power capacity and land availability timing are key to commercialization
For the joint study by NVIDIA and PreussenElektra to progress to actual construction plans, power supply and land-use conditions must be made concrete for each candidate site. What matters most is how many megawatts can be received from when, and which portion of the land can be used at that time. Even if sufficient power is secured, a facility cannot operate without usable land, and even if the land is ready, computing equipment cannot run until the grid connection is in place.
The situation for reusing nuclear plant sites also differs from site to site within Germany. NDR also contacted RWE, another nuclear plant operator, and a company spokesperson said there are currently no specific plans for reusing the Emsland plant site, as its dismantling will continue into the 2030s. This is a separate matter from the NVIDIA–PreussenElektra study, but it shows that dismantling progress and reuse prospects differ from plant to plant.
The announcement describes an early-stage effort to explore former nuclear plants as locations for AI data centers, and it is not yet clear how much construction cost and time could be cut by using existing infrastructure. Once specific candidate sites, power receiving capacity, and land availability dates are presented, and conditions such as cooling facilities, communications infrastructure, permits, and investors are in place, the project's viability can be assessed.
In building AI data centers, the timing of securing power and land is a key constraint, alongside procuring computing equipment. How far the assets remaining at former nuclear sites can help solve that problem will have to be judged by the concrete conditions that emerge from this joint study.