RESEARCH
AI Data Center Power Surge: 415 TWh in 2024, On Track to Double by 2030
Global data centers consumed approximately 415 TWh of electricity in 2024—about 1.5% of world electricity use—driven increasingly by AI training and inference workloads. The IEA projects this will nearly double to 945 TWh by 2030 in its baseline scenario, while Goldman Sachs forecasts a larger 165% rise from 2023 levels, reflecting genuine uncertainty in AI compute growth trajectories. The U.S. alone—accounting for roughly 45% of global consumption—faces DOE-funded projections of 325–580 TWh by 2028. All credible sources agree on the direction: AI infrastructure is reshaping global electricity demand at a pace not seen outside major industrialization events.
Findings
- Global data centers consumed ~415 TWh in 2024, equivalent to the combined electricity use of Germany and Austria, and up ~12% annually since 2017.
- AI accelerated servers (GPUs, TPUs) are growing at 30% per year within the broader data center footprint—four times faster than total global electricity consumption.
- The IEA 2030 baseline projection (945 TWh) and Goldman Sachs forecast (165% growth from 2023 levels) diverge by ~30–40%, reflecting genuine uncertainty in how quickly AI model training and inference scales.
- The U.S. hosted ~45% of global data center electricity in 2024; DOE/LBNL projects U.S. data center demand will reach 325–580 TWh by 2028—a wide band driven by AI adoption pace.
- Multiple independent sources agree on the 2024 ~1.5% global electricity baseline but diverge on 2030 upper bounds, making UVRN verification valuable for distinguishing central-case consensus from high-scenario disagreement.
Delta Engine result
↔ Divergence Detected — Δ 0.375 (threshold 0.05)
- IEA T5 95
- LBNL/DOE T5 82
- S&P Global T3 72
- Goldman Sachs T0 78
- World Economic Forum T0 65
Evidence quality
avg 62 · min 25 · max 95 · spread 70