54 percent of data centre power demand is geographically flexible
BDF Decoder

54 percent of data centre power demand is geographically flexible

Zonal electricity prices would encourage data centres to move away from congested hubs, Banque de France blog authors argue. They estimate that 54 percent of projected data centre electricity demand in 2030 relates to uses that can relocate.

Growth crowds into congested hubs

Josef Nasr and Thaddée Toulouse cite projected increases in annual data centre electricity demand by 2030 of 45 TWh in Europe and 240 TWh in the United States.

EU renewable capacity is forecast to expand by 440 GW, but grid congestion limits delivery.

Frankfurt, London, Amsterdam, Paris and Dublin host three-quarters of Europe's compute capacity; an inventory of 118 construction projects shows continued concentration.

Dublin's data centres consume 20 percent of national electricity production, while grid connections take up to 10 years in Frankfurt and London.

Ireland plans 2 GW of additional gas-fired capacity for data centres by 2030.

The European Commission estimates grid expansion costs at €584 billion; secondary hubs account for only 30 percent of forecast growth in 2026.

One price conceals local scarcity

Electricity accounts for up to 60 percent of data centre operating costs, yet single national prices conceal local grid constraints.

Extremadura curtailed at least 402.6 GWh of renewable output in August 2025 as fossil generation compensated elsewhere.

German congestion-related redistribution cost €4.2 billion in 2022; Sweden has used four price zones since 2011.

McKinsey projects 2030 demand shares of 28 percent for AI training, 43 percent for inference and 29 percent for non-AI uses.

The authors estimate that uses accounting for 54 percent of demand can relocate.

Location carries a carbon cost

The authors identify grid congestion, rather than insufficient renewable supply, as the central transition risk.

Zonal prices would make location costs visible to developers.

They stress that water, latency, sovereignty and incomplete local data also constrain planning.

Report an error