Summary

• RWE says two datacentre development sites are nearing agreement.
• The utility owns around 30 sites with substantial existing electricity infrastructure that could support compute projects.
• AI inference and other European workloads are bringing technology demand closer to the economics of power generation, grid access, and land.

The European datacentre market is drawing electricity companies deeper into digital infrastructure, as power connections that once served conventional generation sites acquire another commercial use in an AI economy struggling to secure enough grid capacity quickly.

RWE says two datacentre development sites are close to agreement, with chief executive Markus Krebber indicating during the German utility’s half year results that contracts could be settled in the near term. Neither the locations nor prospective customers have yet been disclosed, leaving both projects at the negotiation stage.

The company owns roughly 30 sites with substantial electricity infrastructure and is assessing how some of them could support datacentres alongside batteries, flexible generation, and other energy assets. Krebber put the attraction succinctly: “We own 30 sites with good electricity infrastructure. That is a real asset.”

Existing grid access can remove one of the longest lead times facing large compute projects. New datacentres may be technically straightforward to design compared with the surrounding energy work, yet projects can stall when local networks cannot supply the required load without new substations, transmission work, or years of reinforcement.

Compute is turning into an energy site selection problem

Fibre routes, customer proximity, planning, and land remain important to datacentre development, although dense AI infrastructure has increased the value of electricity capacity considerably. Large accelerator clusters can require hundreds of megawatts, while their power profile also places new demands on cooling, backup systems, and local networks.

For a utility such as RWE, that creates several possible revenue streams around the same customer. It can supply electricity through long term power purchase agreements, develop renewable or flexible generation, use existing sites as property for compute projects, and potentially add storage where the network benefits from greater flexibility.

Krebber has also pointed towards demand for operational AI infrastructure in Europe, where inference and customer facing workloads may need to run closer to users than the giant training clusters used to develop frontier models. Latency, jurisdiction, resilience, and data location can all influence where those systems operate, creating a case for more distributed capacity even if the largest training campuses remain concentrated in a smaller number of locations.

RWE’s first half results show the datacentre opportunity sitting inside a much larger investment programme. Adjusted EBITDA rose to €3 billion from €2.1 billion a year earlier, while new generation capacity continues to come online and further renewable and flexible assets remain under construction.

Power companies are becoming part of the compute supply chain

Technology companies have increasingly signed long term electricity contracts to support datacentre expansion, although annual renewable matching alone does not resolve every infrastructure constraint. Developers still need power at the right location and at the time their equipment consumes it, while local grids need enough capacity to serve other industrial, residential, and transport electrification demands.

That can produce political friction when a new datacentre appears to compete with factories, housing, or wider economic development for scarce connections. Energy companies that offer sites with existing infrastructure can shorten the development process, but the same advantage raises questions about how former or current power assets should be allocated when several sectors want the same network capacity.

Inference demand could make that competition more geographically dispersed than the first phase of the AI infrastructure boom. Training a frontier model can be concentrated where power and hardware are cheapest, whereas services used continuously by European companies and public bodies may place greater value on proximity and jurisdiction.

The two RWE projects are not final agreements, so their eventual scale and commercial importance remain unknown. The stronger structural signal lies in the asset base the company is bringing to negotiations: land, substations, grid connections, generation relationships, and expertise in supplying very large loads.

As AI computing expands, those physical assets are becoming part of the technology stack. Europe can purchase more accelerators, but the machines still need somewhere to operate, and the companies controlling viable electricity sites are acquiring an increasingly influential role in deciding where that capacity can be built.