A new analysis from Oil Price estimates that the energy demand of AI data centers, expressed in barrels of oil equivalent, is now reaching levels comparable to the daily consumption of entire nations, underscoring the enormous energy footprint of the technology. The report builds on a methodology previously used to quantify Bitcoin's energy use, finding that AI's appetite for power is accelerating even faster.
Bitcoin mining currently consumes between 138 and 175 terawatt-hours annually, translating to roughly 500 to 600 barrels of oil equivalent per coin depending on efficiency estimates. AI data centers, the report warns, are on a trajectory to eclipse that figure as training and inference compute demands grow exponentially.
The analysis highlights that this energy is drawn largely from fossil fuel-heavy grids in regions like the US Southeast and parts of Asia, where natural gas and coal remain primary baseload sources. This creates a direct link between AI expansion and increased hydrocarbon demand.
This trend poses a counterpoint to decarbonization goals, as the incremental power demand from AI risks locking in fossil fuel infrastructure for years. The report notes that while some hyperscalers have signed renewable power purchase agreements, the sheer scale of new demand may outpace green energy additions.
A counter argument holds that efficiency gains in both hardware (more efficient accelerators) and software (model compression) could reduce AI's energy intensity over time. Additionally, grid-scale battery storage and nuclear partnerships (e.g., Microsoft's Three Mile Island restart) may eventually decouple AI growth from oil consumption.