How Much Electricity Do Data Centers Use? 2026 Statistics and Forecasts

Two engineers walking beside a data center and its electrical substation with transformers and power lines

Data centers now rank among the fastest-growing sources of electricity demand. AI servers account for much of the recent increase.

Our cloud computing statistics track the wider growth behind that demand. Global and US estimates use different periods and models. Current consumption should remain separate from future scenarios.

Data centers used about 485 terawatt-hours worldwide in 2025, according to the International Energy Agency. That equals roughly 1.6% of global electricity. The agency projects about 950 TWh in 2030. US data centers used 176 TWh in 2023 and could reach 521 to 843 TWh in 2030.

Data-center Electricity Use at a Glance

Diagram

Data-center Electricity Use at a Glance

  1. Global, 2024
  2. Global, 2025
  3. Global, 2030 projection
  4. United States, 2023
  5. United States, 2028 projection
  6. United States, 2030 projection
  7. United States, 2030 reference case
Geography and periodElectricity useShare or context
Global, 2024415 TWhAbout 1.5% of global electricity
Global, 2025485 TWhAnnual demand rose 17%
Global, 2030 projectionAbout 950 TWhAround 3% of global electricity
United States, 2023176 TWhAbout 4.4% of US electricity
United States, 2028 projection325 to 580 TWh6.7% to 12% of US electricity
United States, 2030 projection521 to 843 TWh9.5% to 15.3% of US electricity
United States, 2030 reference case649 TWh11.8% of US electricity

TWh measures annual energy. MW and GW measure power capacity at a moment in time.

How Much Electricity Do Data Centers Use Worldwide?

The International Energy Agency estimated global data-center demand at 485 TWh during 2025.

Consumption rose 17% from the prior year. Global electricity demand grew about 3% during the same period.

The IEA's 2024 estimate was 415 TWh, or 1.5% of worldwide electricity. The increase reflects AI, cloud services, storage, streaming, and other workloads.

AI-focused facilities grew faster than the wider sector. Their electricity use rose about 50% during 2025.

How Much Electricity Could Data Centers Use in 2030?

The IEA's central projection reaches 950 TWh in 2030. That is close to double the 2025 total.

The projected share rises to around 3% of global electricity. AI-focused data-center consumption could triple during the period.

This forecast is not a fixed outcome. Grid bottlenecks, equipment supply, efficiency, financing, and regulation can change delivery.

The agency still sees higher upside after 2030. Agents, reasoning models, and video generation may add more intensive workloads.

How Much Electricity Do US Data Centers Consume?

The LBNL 2024 report measured 176 TWh during 2023.

That represented about 4.4% of US electricity. Consumption had risen from 58 TWh in 2014.

The earlier model projected 325 to 580 TWh in 2028. That range represented 6.7% to 12% of US demand.

LBNL updated the forecast in June 2026. Its new 2030 reference case is 649 TWh, or 11.8% of US electricity.

Why Is the 2030 US Range So Wide?

LBNL's compounded range runs from 521 to 843 TWh. That equals 9.5% to 15.3% of US electricity.

The model changes several uncertain inputs:

  • data-center equipment installations;
  • specialized graphics-processor shipments;
  • AI chip operating life;
  • AI server utilization; and
  • idle server power.

The reference case uses expected equipment shipments and modeled annual electricity. Sensitivity cases test different assumptions.

This approach provides a range, rather than false precision.

What Is the Difference Between Energy and Power?

Energy measures electricity used across time. Power measures the rate of use at one moment.

A data center rated at 100 MW can draw up to 100 megawatts. If it ran at that level for one year, it would use 876,000 MWh.

The calculation is:

100 MW × 8,760 hours = 876,000 MWh, or 0.876 TWh

Real facilities do not always operate at full nameplate capacity. Utilization and maintenance change annual use.

This distinction prevents a common reporting error. GW of planned capacity cannot be compared directly with annual TWh.

Why Are AI Servers Increasing Demand?

AI accelerators draw more power than many conventional processors. New racks also pack more computing into the same space.

Training uses large clusters for fixed development runs. Inference adds continuous demand as people and software call deployed models.

Reasoning and agentic tasks may require several internal steps. Video generation can also demand more computing than short text.

The AI energy consumption guide explains why model-level estimates remain uncertain.

Does Efficiency Offset Growth?

Power per task can fall as processors, models, and serving systems improve. Facility overhead has also declined.

LBNL placed average power usage effectiveness near 1.4 in 2023. It was around 1.6 in 2014.

At 1.4 PUE, a facility uses 1.4 energy units for each unit reaching IT equipment.

Efficiency savings may be overtaken by more users and workloads. Lower computing cost can encourage higher total consumption.

Where Is Demand Concentrated?

Data centers cluster around fiber, power, customers, skilled workers, and available land. This concentration raises local grid effects.

The IEA reported that nearly half of US capacity sat within five regional clusters. National demand can therefore understate local pressure.

Utilities may need new substations, transmission, generation, and storage. Those projects often take longer than building the computing equipment.

The IEA said around 20% of planned projects could face delays without faster grid integration.

What Energy Sources Will Supply New Demand?

The IEA expects several sources to contribute through 2035. Renewables could supply about half of global growth.

Natural gas, nuclear, geothermal, and storage also appear in utility and operator plans. Availability differs by market.

Annual renewable contracts do not guarantee carbon-free power every hour. Grid mix and timing still affect emissions.

The cloud infrastructure hub covers the operating choices behind capacity and architecture.

How Can Operators Reduce Electricity Demand?

Operators can improve hardware efficiency, server utilization, cooling, and workload scheduling. Retiring inactive equipment also reduces idle power.

Software teams can choose smaller models for simpler work. Caching and batching may reduce repeated computing.

Cooling choices also shift data center water usage, so energy savings can raise water demand. Heat reuse can recover some thermal energy. Its value depends on nearby demand and usable temperatures.

These steps reduce demand per unit of work. They may not reduce total consumption when usage continues growing.

How Should Forecasts Be Read?

Analysts asking how much electricity do data centers use must keep projections separate from observed demand.

Use three checks before repeating a projection:

  1. Confirm the geography and base year.
  2. Separate observed use from forecast use.
  3. Read the scenario assumptions and uncertainty range.

The 649 TWh US figure is a reference case. It is not a guaranteed 2030 result.

Methodology

This article uses IEA, DOE, and LBNL evidence available through 8 November 2026. Observed totals and projections appear separately.

We use TWh for annual consumption and MW or GW for power. Percentages retain each source's national or global denominator.

FAQs

What Percentage of Global Electricity Do Data Centers Use?

The IEA estimated about 1.6% during 2025. Its central 2030 projection reaches around 3%.

What Percentage of US Electricity Do Data Centers Use?

LBNL measured about 4.4% during 2023. Its 2030 reference case reaches 11.8%, with a 9.5% to 15.3% range.

Are AI Data Centers Included in These Totals?

Yes. Sector totals include AI-focused facilities and conventional workloads. AI is the fastest-growing driver in current forecasts.

Is One Gigawatt the Same as One Terawatt-hour?

No. A gigawatt measures power. A terawatt-hour measures energy used over time.

Will Data-center Electricity Demand Double by 2030?

The IEA's central global projection nearly doubles between 2025 and 2030. Physical and financial constraints could change the result.

Resources