Teravolt targets industrial 'brownfield' sites to bypass AI power bottlenecks
The London-based firm is repurposing aluminum smelters and power plants in Europe to secure high-voltage energy faster than grid expansion allows.
London-based AI infrastructure company Teravolt is repurposing "brownfield" industrial sites to solve the critical power shortage facing the AI industry. By converting old aluminum smelters, Bitcoin farms, and thermal power plants into data centers, the company aims to bypass the decade-long delays associated with building new electrical grid infrastructure.
Teravolt is focusing its current expansion on Eastern and Southern Europe, where industrial sites remain more affordable. This strategic pivot comes as prices for similar power-backed sites in the U.S. and Western Europe have surged four to five times over the last five years. The company notes that mid-sized AI labs, which often require capacity for 20,000 GPUs within 12 months, find such timelines virtually unachievable in Western markets.
The Grid Gap
The shift toward repurposing existing sites is driven by a massive disparity in construction speeds. According to Teravolt, while a data center can be built in one to three years, the planning and construction of new grid infrastructure typically takes between five and 15 years. This creates a dangerous bottleneck as AI demand scales.
Industry projections highlight the severity of the shortfall. Teravolt predicts AI power demand will hit 410 GW by 2036, leaving a projected gap of 240 GW relative to the 170 GW they anticipate will be available from the grid. These figures align with broader market trends; Gartner predicts global data center power demand will rise from 104 GW in 2025 to 290 GW by 2030. Similarly, Morgan Stanley expects U.S. data center demand to reach 74 GW by 2028, with a shortfall of approximately 49 GW.
The Economics of Cannibalization
This strategy represents a fundamental shift in industrial economics, where AI compute is becoming a more lucrative way to extract value from electricity than traditional heavy industry. Teravolt estimates that an aluminum smelter generates between $170 and $190 in gross revenue per MWh. In contrast, an AI data center can generate between $450 and $900 in revenue, with approximately $300 in EBITDA per MWh.
"The low marginal users of power will be cannibalized by the new form of extracting value from the power," said Denis Alkhazov, founder and general manager of Teravolt. Co-founder and managing partner Laert Karaashev added that while other layers of the AI stack may change, the bottom layer—power—is far more difficult to commoditize.
Future Implications
By "cannibalizing" legacy industries, AI infrastructure providers can accelerate the deployment of frontier models, but this may come at the cost of displacing traditional manufacturing in regions with limited power. The industry is now moving away from seeking empty land in favor of "power-backed" locations that already possess the necessary high-voltage connections, permits, and energy contracts. The success of this model will likely depend on the availability of affordable industrial sites across Europe as the race for compute capacity intensifies.