Data centers used about 1.5% of global electricity in 2024. 1
The International Energy Agency estimates roughly 415 terawatt-hours of data-center electricity consumption in 2024 and projects around 945 TWh by 2030 in its base case.
Dossier 03
Technology & economy
How should communities meet AI’s growing electricity demand?
AI’s energy footprint is rising quickly but unevenly. The central policy choice is not simply growth or restraint—it is who builds, who pays, which power sources respond, and how fast grids can adapt.
01
First, establish the record
What we know—and how well we know it
These tiers distinguish direct documentation, broad expert interpretation, and questions the evidence cannot yet settle.
Documented
Directly supported by law, measurement, or an authoritative record.
Demand accelerated in 2025. 3
The IEA reports that global data-center electricity use grew 17% in 2025, while electricity use at AI-focused facilities rose even faster.
New demand will be met by a mix of power sources. 2
IEA projections have renewables meeting nearly half of added data-center demand through 2030, with natural gas, coal, and later nuclear power also contributing.
Broadly supported
A synthesis supported across multiple credible sources, with stated limits.
Location matters as much as the global total. 14
Large facilities are geographically concentrated, so local grid constraints and generation choices can be substantial even when data centers remain a modest share of global demand.
Efficiency alone may not reduce total consumption. 3
Power use per AI task is falling rapidly, but expanding adoption and more energy-intensive uses can outweigh those savings in aggregate.
Still unresolved
An important question that current evidence or law does not settle.
How closely will projections match actual deployment? 13
Demand depends on model efficiency, hardware, utilization, investment, grid queues, supply chains, and how quickly AI use expands.
Who will bear grid and generation costs? 4
Rate design and utility regulation will determine whether infrastructure costs are absorbed by data-center operators, shared across customers, or offset by wider system benefits.
02
Where views diverge
Each side’s strongest case
These are argument families, not political identities. A person may agree with parts of more than one.
Position A
Build power and compute together
AI infrastructure is a strategic economic asset; regions should accelerate generation, transmission, and permitting so electricity constraints do not push investment elsewhere.
- Compute capacity can support productivity, research, national security, and high-value investment.
- The durable answer to new demand is greater energy abundance rather than rationing promising uses.
- Large, creditworthy customers can help anchor investment in grids and new generation.
PrioritizesInnovationAbundanceCompetitiveness
Grounded in 2 sources
Position B
Make growth pay its full costs
New data centers should proceed only with clear protections against higher household bills, dirtier generation, water stress, and infrastructure costs shifted to the public.
- A project can impose local costs even when its global footprint appears modest.
- Contracts and rate structures should place new infrastructure risk on the companies creating demand.
- Clean-energy and water commitments should be measurable in the places and hours where consumption occurs.
PrioritizesAccountabilityAffordabilityEnvironmental stewardship
Grounded in 2 sources
Position C
Use flexibility before adding supply
Efficiency, flexible computing schedules, better siting, and existing grid capacity should be exhausted before committing customers to long-lived generation and transmission.
- Some computing loads can shift across hours or locations more readily than traditional industrial demand.
- Fast-changing hardware and demand forecasts increase the risk of overbuilding durable infrastructure.
- Transparent grid queues and performance incentives can reward operators that reduce system peaks.
PrioritizesEfficiencyOptionalitySystem reliability
Grounded in 2 sources
03
Beneath the talking points
Why the disagreement persists
Facts matter, but many durable debates turn on different values, definitions, forecasts, and levels of institutional trust.
Forecasts
Small differences in adoption and efficiency assumptions produce very different infrastructure needs.
Who pays
The same project looks different depending on whether costs and risks fall on operators, utilities, or ratepayers.
Time horizon
Near-term grid constraints and long-term economic benefits are weighed differently.
System boundary
Global electricity shares can look modest while local grid, water, and emissions effects are concentrated.
What would move the debate?
Evidence worth watching
Not every disagreement can be measured away. These developments would narrow important factual uncertainties.
- Facility-level reporting of electricity, water, emissions, flexibility, and forecast-versus-actual demand.
- Transparent utility rate cases showing who pays for dedicated and shared infrastructure.
- Real-world demonstrations of flexible AI workloads and clean, reliable 24/7 power procurement.
Sources & review
Follow the evidence
Every evidence-bearing statement above links back here. External sources open in a new tab.
01
Independent research
02Energy demand from AI
International Energy Agency · April 10, 2025
Independent research
03Energy supply for AI
International Energy Agency · April 10, 2025
Independent research
04Data centre electricity use surged in 2025
International Energy Agency · April 16, 2026
Government analysis
DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers
U.S. Department of Energy · December 20, 2024
Review and corrections policy
This prototype was reviewed on August 15, 2026. We distinguish documentation, expert synthesis, and open questions; we revise claims when stronger evidence or legal developments change the record. Corrections can be sent to editorial@commonground.news.