Datacentre Expansion May Face Power Constraints by 2030

The energy consumption of datacentres is projected to increase by 26% this year, primarily driven by artificial intelligence, with grid capacity potentially failing to meet demand by 2030, according to Gartner.

Gartner anticipates global datacentre electricity usage will reach 565 terawatt-hours (TWh) in 2026, with power demand escalating from 104 GW in 2025 to 132 GW this year.

These figures exceed the previous estimation of 500 TWh annually by 2027 for AI-optimised servers.

The escalating requirement for computational power to support AI workloads is the principal factor, as companies invest heavily in AI projects amidst a fear of missing out, often without substantial returns.

AI-optimised servers continue to drive the rise in datacentre power consumption, with significant investments from hyperscalers and other entities funnelling budgets into advanced systems for AI processing.

Currently, AI-optimised servers are expected to account for 31% of total datacentre power consumption this year, and they are projected to surpass the power usage of all conventional servers by next year.

This aligns with previous forecasts indicating that AI workloads are poised to overtake other server operations, such as databases and analytics, becoming the leading server deployment workload by 2027.

The persistent growth, however, suggests a concerning outlook. Gartner estimates that total datacentre electricity consumption will exceed 1,200 TWh by 2030, cautioning that grid capacity may not support additional datacentre developments.

Previous warnings have highlighted the potential for datacentre energy demands to exceed the grid's delivery capacity. Goldman Sachs predicted that their combined energy usage would more than double by the decade's end, but Gartner’s current figures suggest demand is already above their 2027 projections.

Schneider Electric previously outlined scenarios for future electricity consumption by AI datacentres, yet Gartner's latest figures for 2030 surpass even Schneider's most aggressive forecasts.

Power grid operators and datacentre planners in the United States, in particular, face challenges, with no straightforward solutions evident, as industry analysts have recently reported.