Teravolt, a London-based AI infrastructure firm, argues that meeting soaring AI power demand means repurposing older industrial assets, such as Bitcoin mining farms, aluminium smelters and disused thermal power plants, into datacentres rather than waiting for new grid capacity to be built. This matters because forecasters expect a substantial shortfall between electricity supply and AI compute demand within the next decade, and building new grid infrastructure takes far longer than constructing a datacentre, meaning existing energy assets with ready-made power connections, permits and staff offer the fastest route to closing the gap.

Gartner predicts global datacentre power demand will hit 132 GW in 2026 and 290 GW by 2030, while Teravolt itself forecasts a 240 GW shortfall by 2036. The firm says AI compute generates far higher revenue per megawatt than legacy industries, for example an aluminium smelter and is therefore economically incentivised to convert brownfield industrial sites, particularly in Eastern and Southern Europe where such sites remain comparatively cheap, into AI infrastructure, with executives predicting this "cannibalisation" of older industries will eventually spread to other regions.