Tech Giants Turn to Old Power Stations for AI Expansion
The insatiable appetite for artificial intelligence is prompting tech giants and their suppliers to explore repurposing decommissioned power stations and industrial sites into data centres.
Microsoft, Google, and Amazon are investing billions of pounds in building data centres to fuel their cloud computing and AI services. However, finding suitable locations with sufficient power for these energy-hungry facilities is becoming increasingly difficult.
Adam Cookson, head of land transactions for Cushman & Wakefield's EMEA data centre advisory group, highlighted that many data centre markets are "severely constrained in terms of land availability and power supply." This has led to a growing interest in smaller markets and "more complex sites" like former power stations.
Daniel Thorpe, data centre research lead at real estate group JLL, echoed this sentiment, saying that developers of large data centre campuses are actively looking at locations like "infrastructure sites or power stations." He added that "typically, it's a large 'hyperscale' facility that would be eyeing up a power station," referring to the leading cloud computing providers such as Microsoft, Amazon, and Google.
Coal-fired power stations are being decommissioned across the US and Europe. However, they often possess characteristics that make them ideal candidates for data centre campuses. Industrial sites are typically designed for high power usage, featuring existing power transmission infrastructure and often located near water sources.
Microsoft is planning to develop data centres on the sites of the former Eggborough and Skelton Grange power stations, near Leeds in northern England, with construction expected to begin on Eggborough in 2027. Meanwhile, Amazon is planning a campus on the site of the old Birchwood power station in Virginia, USA.
One person familiar with the matter revealed that at least one other similar power station deal in Europe is currently being negotiated.
The tech industry has warned that electricity availability constraints could impede AI expansion. Other requirements, such as sufficient fibre connectivity, further reduce the number of suitable locations for new data centres.
Analysts believe this situation encourages the exploration of less conventional options. The distinct demands of AI workloads present opportunities to place data centres in less central locations, further away from major computing hubs, as "latency" (the time it takes to send data and receive a response) is less crucial for training AI models.
Repurposing existing sites is also gaining traction. Rahul Mewawalla, CEO of Mawson Infrastructure Group, reported "an increased influx of inbound activity" from owners of industrial and power assets, including private equity groups, keen on partnering to convert these assets into data centres.
Virtus Data Centres, in which Macquarie Asset Management holds a minority stake, recently acquired two sites in Berlin, Germany, including a former solar farm and an old munitions factory in the UK. The plan is to transform these sites into data centre campuses by 2026.
Thor Equities Group recently acquired a former manufacturing plant in Georgia, USA. Joe Sitt, the company's chairman, described it as "equipped with transformers, water, sewer, and natural gas infrastructure" and "well-suited for data centre development."
This trend mirrors the actions of the energy-intensive bitcoin mining industry, which has actively sought to repurpose disused industrial sites, including old aluminium smelters.
However, some experts caution that converting these sites could be a lengthy, expensive, and bureaucratic process, potentially impractical if a power plant has been disconnected from the grid and excluded from the local operator's plans.
Mark Dyson, managing director of the carbon-free electricity programme at the Rocky Mountain Institute, warned that "it may not be easy for the utility to flip the switch and turn it back on." He added, "Those challenges have come up in our conversations with companies."
JLL's Daniel Thorpe emphasised that "a lot depends on the specifics of the site, how much it would cost for that adaptive reuse, land scarcity and land prices."
Research by the Rocky Mountain Institute suggests that renewable power could be co-located alongside existing fossil fuel generation, routed to the grid via the plant's existing connections when economically viable.
Theoretically, any surplus generation - grid connection facilities have limits on how much they can add to the system - could be used to power an on-site facility like a data centre.
