The AI boom runs on electricity, and the scale of that demand has become one of the biggest stories in energy policy. Data centers are being built faster than some grids can supply them, raising real questions about cost, sustainability, and capacity.
These AI energy consumption statistics for 2026 lean heavily on the International Energy Agency (IEA), the most authoritative source on the topic, to separate the genuine trend from the alarmist and the dismissive takes alike.
Key takeaways
- Data center electricity is surging. Demand jumped sharply in 2025, far outpacing overall electricity growth.
- The IEA expects it to roughly double by 2030. Reaching around 3% of global electricity.
- AI is the main driver. AI-focused facilities are growing fastest within data centers.
- Water matters too. Cooling consumes significant water, prompting a shift to liquid cooling.
- Capital is enormous. Hyperscalers are spending hundreds of billions, mostly on AI infrastructure.
AI data center electricity statistics
According to the IEA’s Energy and AI analysis, global data center electricity consumption reached roughly 460–490 TWh in 2025, a surge of around 17% year over year, dramatically faster than the roughly 3% growth in overall global electricity demand. AI-focused facilities are driving a disproportionate share of that increase.
To put it in perspective: data centers already consume more electricity than some entire countries, and the growth rate, not just the level, is what concerns grid operators.
AI energy projections to 2030
The IEA projects data center electricity consumption roughly doubling from about 485 TWh in 2025 to around 950 TWh by 2030, which would represent close to 3% of global electricity demand. Power use from AI-focused data centers specifically is projected to grow even faster, potentially tripling over the same period.
| Metric (IEA) | 2025 | 2030 (projected) |
|---|---|---|
| Data center electricity | ~485 TWh | ~950 TWh |
| Share of global electricity | ~1.5% | ~3% |
| AI-specific power use | Base | Roughly triple |
These are projections, not certainties, efficiency gains, chip improvements, and policy could change the trajectory. But the central estimate from the most credible source is clear: a near-doubling this decade.
AI water usage statistics
Electricity isn’t the only resource. Data centers use water directly for cooling and indirectly through power generation. U.S. data centers used on the order of billions of gallons directly in recent years, with much larger volumes attributable indirectly to the electricity they consume.
The industry response is a rapid shift to liquid cooling, direct-to-chip and immersion, which can cut direct water use substantially and is increasingly necessary as AI chip racks reach power densities that air cooling cannot handle.
AI infrastructure investment statistics
The capital behind this is staggering. The largest hyperscalers have signaled combined capital expenditure in the range of hundreds of billions of dollars for 2026, with the large majority tied directly to AI infrastructure, chips, servers, and the data centers and power to run them. That spending is itself a leading indicator of where energy demand is heading. For the broader money picture, see our AI investment statistics.
What this means
For businesses and policymakers, three things follow. First, energy is now a strategic constraint on AI, compute growth is increasingly limited by power availability, not just chips. Second, efficiency is a competitive edge: more efficient models and cooling reduce cost and footprint. Third, sustainability claims need scrutiny, “AI for good” has to be weighed against real resource use, and transparency about energy and water is becoming an expectation.
Methodology and sources note
Figures here come primarily from the IEA’s Energy and AI work, supplemented by reporting from outlets such as Consumer Reports and analysis from Brookings. Projections involve significant uncertainty and depend on assumptions about efficiency and buildout. Treat 2030 figures as central estimates, not guarantees, and check the IEA’s latest data for updates.
Frequently asked questions
How much electricity do data centers use?
The IEA estimates roughly 460–490 TWh globally in 2025, up about 17% year over year, far faster than overall electricity demand growth. That’s more than some entire countries consume.
How much will AI energy use grow by 2030?
The IEA projects data center electricity roughly doubling to around 950 TWh by 2030, about 3% of global electricity, with AI-specific use growing even faster.
Does AI use a lot of water?
Yes, indirectly and directly, mainly for cooling. Data centers consume billions of gallons, which is driving a shift to liquid cooling that can cut direct water use significantly.
Is AI’s energy use a serious problem?
It’s a serious challenge, especially locally where grids are strained. It’s not an extinction-level issue, but power availability is becoming a real constraint on AI growth.
Can AI become more energy efficient?
Yes. More efficient chips, models, and cooling are improving rapidly, and efficiency is now a competitive and environmental priority. Whether efficiency outpaces demand growth is the open question.
Conclusion
The data is unambiguous on direction: AI is driving a steep rise in data center electricity and water use, with the IEA projecting a near-doubling of demand by 2030. Energy has become a strategic constraint on AI, making efficiency and transparency essential rather than optional.
