Booming demand for artificial intelligence is encouraging big tech companies and their suppliers to explore converting old power stations and industrial sites into data centers.

Microsoft, Google, and Amazon are pouring billions of dollars into building data centers to power cloud computing and AI services, but it has become increasingly challenging to find suitable locations with sufficient power for the energy-hungry facilities.

Many data center markets are “heavily constrained when it comes to land availability and power,” which in turn fueled interest in smaller markets and “more complicated sites” such as old power stations, said Adam Cookson, head of land transactions for real estate firm Cushman & Wakefield’s Emea data center advisory group.

There are “increasing opportunities” for the owners of such assets, he added.

Daniel Thorpe, data center research lead at real estate group JLL, said the developers of large data center campuses were looking at locations including “infrastructure sites or power stations.”

“Usually, it’s a large ‘hyperscale’ facility that would be looking at a power station,” he added, referring to the largest cloud computing providers such as Microsoft, Amazon, and Google.

Coal power stations are being decommissioned in parts of the US and Europe, but may have attributes that a data center campus would need. Industrial sites will typically have been designed for high power usage, for example, and might come with power transmission infrastructure and be located close to a water source.

Microsoft intends to develop data centers on the sites of the old Eggborough and Skelton Grange power stations, near Leeds in northern England, with construction expected to start on the former in 2027. Amazon, meanwhile, is planning a campus on the site of the old Birchwood power station in the US state of Virginia.

At least one other similar power station deal in Europe is currently being brokered, according to one person familiar with the matter.

The tech industry has warned that electricity availability constraints threaten to hold back AI expansion, while other requirements, such as sufficient fiber connectivity, further shrink the pool of potential spots for a new data center.