A new study projects that electricity consumption by AI data centers will rise from approximately 118 terawatt-hours in 2024 to between 239 and 295 terawatt-hours by 2030, representing roughly 1% of global electricity demand. The study focuses on six firms, Amazon, Microsoft, Google, Meta, Oracle, and Apple, which collectively account for approximately 70 to 75% of global hyperscale and cloud-linked data center electricity demand. North America, Western Europe, and the Asia-Pacific together account for more than 90% of projected computing capacity, with regions such as Oregon, Ireland, and Iowa identified as facing concentrated grid pressure. The study combines large language model analysis of corporate, policy, and media sources with scenario-based projections of future electricity demand.
The findings frame AI infrastructure as a structural component of power-system dynamics rather than a passing demand surge, with GPU-based computation consuming up to six times more power than conventional racks. Concentrated deployment in a small number of regions means that localized grid stress and transmission constraints will intensify even as global averages remain modest.
A peer-reviewed Nature publication projecting AI data center electricity demand reaching 1% of global supply by 2030 is a high-credibility, quantified impact finding with broad industry relevance. The Nature journal source and the specific 2030 projection signal this as a major research milestone worth covering.