A new BloombergNEF report predicts that US data centres will consume roughly one-fifth of the country's generated electricity by 2035, a fourfold increase from current levels, driven largely by surging demand for AI computing power. This matters because it signals a dramatic acceleration in energy demand forecasts, with data centre capacity expected to reach nearly 200 gigawatts within a decade, placing further strain on power grids that are already struggling to keep pace with connection requests.

BloombergNEF's new estimate is 83% higher than its own December forecast, and other bodies including EPRI and S&P have similarly revised their projections sharply upwards. Nearly half of the new capacity will go towards AI training and inference, with the US expected to host 64% of AI chips by power demand by 2033. Grids in the PJM Interconnection (Virginia to Illinois) and ERCOT (Texas) will face particular pressure, with data centres taking up 34% and 22% of capacity respectively; PJM has already seen electricity prices rise 76% over the past year amid the strain, prompting one utility to threaten withdrawal from the network. Globally, aggressive AI adoption could add 1,935 terawatt-hours of new demand by 2033 — nearly matching India's entire annual electricity use.