AI Datacenters: Why NVIDIA Is Assessing Former German Nuclear Sites

A quiet but symbolic announcement: NVIDIA and PreussenElektra, E.ON's decommissioning subsidiary, will jointly assess whether former German nuclear power plants can host AI factories. Eight sites being dismantled, sharing what the market desperately seeks: industrial land, grid connections sized for power plants, and infrastructure already in place.

The principle is not to revive nuclear power — Germany shut down its last reactor in April 2023 — but to reuse what took decades to build: high-voltage connections, cooling water networks, land and the acceptance of industrial use. This is brown-to-green applied to the country's most electro-intensive sites.

The approach remains cautious: an open-ended assessment, with no investment decision or site selection, and decommissioning timelines running into the 2030s. Yet the signal matters: the world's leading AI chip supplier now looks at energy brownfields as reserves of capacity.

Why Electrical Brownfields Are Becoming Strategic

The AI bottleneck is no longer just the GPU: it is years of waiting for grid connections and permitting delays. A site already connected at very high voltage, with an industrial history, eliminates the two main queues. This is exactly what operators seek when capacity commitments are measured in gigawatts.

Europe is full of such assets: decommissioned coal or nuclear plants, steelworks, refineries. Their common point: power was already there. In a market where the available megawatt is the scarce resource, these brownfields are reserves of time as much as of land.

What This Changes for Investors

The revaluation of electro-intensive brownfields changes how balance sheets are read: sites booked as decommissioning liabilities can become productive assets again. PreussenElektra itself frames it as responsible reuse and new economic activity for the regions.

But assessment must stay cold-headed: a former power plant is not a turnkey datacenter. Turbine halls were not designed for high-density racks; cooling, substations and fiber must be rebuilt, under specific regulatory constraints. Brownfield creates value when conversion cost stays below the time saved — not automatically.

How We Apply This at FINXIA

This announcement directly validates our brown-to-green bias in the TITAN DC AI strategy: we target existing sites already powered and connected, whose transformation costs less and above all less time than greenfield. Seeing NVIDIA apply this reading grid to former nuclear plants confirms our thesis.

We also retain the discipline: site-by-site assessment — connection quality, water, land, regulatory framework — is exactly our selection process. And we stay on sites where transformation is achievable in the short term, not on ten-year decommissioning promises.

Key Takeaways

NVIDIA and PreussenElektra will assess eight former German nuclear plants as candidates for AI factories. Beyond the symbol, it is the recognition that electro-intensive brownfields — connected, serviced, industrial — have become strategic assets for AI deployment.

Worth watching: the first sites selected after the assessment, the industrial structures that follow, and similar initiatives on former coal plants across Europe.