Study projects AI datacenters near 1% of world electricity by 2030

On September 21, 2026, the open-access journal Communications Sustainability published a study by Danbo Chen, Zijun Zhou, Yongyang Cai and co-authors on how much electricity AI datacenters may need.

The study models the datacenters of six companies: Amazon, Microsoft, Google, Meta, Oracle and Apple. The authors say these firms represent roughly 70 to 75% of global hyperscale and cloud-linked datacenter electricity demand.

For those operators, the authors estimate AI datacenter electricity use at about 118 terawatt-hours (TWh) in 2024. They project it reaching 239 TWh in a conservative scenario and 295 TWh in an optimistic one by 2030, which they put at about 1% of global electricity demand. One terawatt-hour is a billion kilowatt-hours.

The study finds that more than 90% of the projected computing capacity sits in North America, Western Europe and Asia-Pacific. It names Oregon, Ireland and Iowa as facing greater pressure on their power systems than larger systems such as Texas.

To build the estimate, the authors used a language model to analyze company reports, policy documents and news coverage, then ran projections under several growth scenarios.

Why it matters

Every AI answer is computed on chips in a datacenter, and those chips need electricity both to run and to be cooled. A global share of about 1% can sound small. The study’s regional findings add a second point: the same demand weighs more on some local power systems than on others. The What AI Costs course explains how electricity, chips and buildings feed into the price of using AI.