Global datacenter power consumption will more than triple by 2030, rising from 387 TWh in 2024 to a forecast 1,200 TWh — a 20.8% compound annual growth rate (CAGR) per Gartner's latest datacenter power forecast published this week.

That trajectory pushes datacenters to 3.3% of worldwide electricity consumption by 2030, up from 1.3% in 2024.

How does this compare to overall power growth?

Worldwide electricity consumption grows at just 3% CAGR over the same span, from 30,856 TWh in 2024 to roughly 36,500 TWh in 2030. Global generating capacity expands faster, at 5.9% CAGR. Power generation has historically outpaced datacenter demand, but hyperscaler growth is now pulling compute demand ahead of the global trend.

The disparity drives a structural reallocation of capacity toward compute infrastructure:

• Datacenter allocated capacity: 104 GW (2024) → 290 GW (2030), 18.56% CAGR
• Datacenter share of global capacity: 1.1% (2024) → 2.1% (2030)
• Datacenter share of total electricity consumption: 1.3% (2024) → 3.3% (2030)

Over six years, datacenter power demand expands 3.1x while total global electricity grows 1.18x.

Which workloads drive the curve?

Gartner separates traditional servers, AI servers, and ancillary datacenter equipment in its breakdown. AI servers carry most of the acceleration. Memory bottlenecks — high-bandwidth memory (HBM), DRAM, and flash — compete with grid power as binding constraints on the next wave of generative AI deployment.

Roughly 75% of all datacenter electricity flows through the largest facilities: hyperscalers, cloud builders, neoclouds, HPC centers, sovereign operators, and the largest enterprises. The remaining quarter spreads across approximately 7 million enterprise, sovereign, and closet-scale sites, alongside 11,500–12,000 "very large" datacenters worldwide.

Smaller facilities continue to consolidate as site sizes scale up. Concentration in fewer, denser builds amplifies the local grid impact — a key factor for hyperscaler clusters in the United States, where most AI capacity expansion is concentrated.

Where does compute sit against other electricity consumers?

Consumer demand and manufacturing still dwarf datacenter loads in absolute terms. The relative growth differential — not absolute volume — tilts the global share in datacenters' favor through 2030.

Is this a replay of the dot-com power scare?

Comparable concerns surfaced after the Dot-Com Boom. Jonathan Koomey, then a staff scientist at Lawrence Berkeley National Laboratory and Stanford professor, projected that datacenters could burn nearly 4% of global power by the early 2010s. The actual share peaked under 2% — still above the 0.5% recorded in 2000 and the 1% in 2010.

Two factors closed the gap between forecast and outcome. Server virtualization migrated mainframe-class efficiency onto x86 platforms, and the post-2008 recession throttled server footprint expansion. Market and engineering responses effectively engineered the doomsday scenario out of existence.

The current cycle differs structurally. Hyperscaler and cloud capacity grow exponentially, and datacenter power demand now expands faster than global generation. The efficiency cushion that absorbed the 2000s curve no longer covers the delta.

What could break the trend?

The same market forces that eased the 2000s trajectory could reappear. Architecture-level tuning for AI workloads already appears in the market, six years ahead of the 2030 endpoint. A second efficiency cycle would compress the 1,200 TWh forecast.

Fusion power remains speculative. Commercial reactors would relieve generation-side pressure, but no published timeline connects fusion deployment to the 2030 datacenter curve.

What is the data provenance?

Gartner supplied the datacenter consumption and capacity figures, with explicit forecasts for 2026–2027 and endpoints at 2030. The 2025 base year and intermediate figures came from stated growth rates. Global electricity and capacity context draws on the U.S. Energy Information Administration and Ember, an energy consultancy.