ENDE

Environment & Energy

How much energy do data centers consume every day?

Global data centres used ~485 TWh in 2025 — up 17% YoY (IEA, Apr 2026); projected to double to ~945 TWh by 2030

Roughly 513K households every second.

Equivalent household-days of electricity consumed by global data centres today (1 household-day ≈ 30 kWh)

~485 TWhannual consumption (2025)
+17%YoY growth (2025)
~945 TWhby 2030 (IEA Base Case)
AI's energy appetite: global data-centre electricity jumped 17% in 2025 to ~485 TWh; AI-focused facilities alone grew ~50% YoY (IEA Key Questions on Energy and AI, Apr 2026). Hyperscaler capex surpassed $400 billion in 2025.

Source: IEA - Key Questions on Energy and AI (April 2026), building on the Energy and AI special report (April 2025). View on dashboard →

The hidden energy cost of every click, stream, and AI query

Data centres used ~415 TWh of electricity globally in 2024 and roughly 485 TWh in 2025 — about 1.5-1.7% of total demand (IEA, Apr 2026). Demand rose 17% year-on-year in 2025, with AI-focused facilities growing ~50%. The IEA Base Case projects ~945 TWh by 2030, enough to power every US home for more than three years.

What data centre energy consumption actually means

485 TWh/year is roughly Germany's entire annual electricity consumption — or enough to power every US home for nearly four years, all consumed by data centres alone in 12 months

The 17% jump in 2025 alone added ~70 TWh, about as much electricity as Austria consumes in a full year

What this means for you

Data centres now consume approximately 1.5-1.7% of global electricity (~485 TWh in 2025, IEA Apr 2026), with AI-focused facilities growing at ~50% per year. The IEA Base Case has global data-centre consumption more than doubling to ~945 TWh by 2030, pushing the sector to roughly 3% of world electricity — comparable to Japan's entire current electricity consumption.

Streaming one hour of HD video uses roughly 0.036 kWh of data centre energy. Multiply that by your monthly viewing hours and you get a rough "internet carbon footprint." For comparison, the average US household uses 30 kWh per day for all purposes.

The decisions being made now about where data centres are built, how they are powered, and how AI workloads are designed will lock in energy infrastructure for decades. Several major cloud providers have committed to 100% renewable energy for their data centres, but the definition of "renewable" and the timing of those commitments varies widely.

Data centre energy vs. water use, today

Cooling is the largest operational expense for data centres, and it consumes enormous amounts of water alongside electricity.

Energy used today
- so far today- this year
equivalent household-days (~485 TWh/year, 2025)
+
Water used today (litres)
- so far today- this year
litres of water for cooling data centres

Energy demand: the key numbers

IEA (Apr 2026): global data centres used ~485 TWh in 2025, up 17% vs. 415 TWh in 2024 — more than five times the 3% global electricity demand growth rate

AI-focused data centres grew ~50% year-on-year in 2025 alone and are projected to triple from ~155 TWh (2025) to ~465 TWh by 2030

IEA Base Case: data-centre electricity demand reaches ~945-950 TWh by 2030, about 3% of global consumption — comparable to Japan's total electricity use today

Capex of the five largest cloud/AI players (Amazon, Google, Meta, Microsoft, Equinix) surged to over $400 billion in 2025 and is set to rise a further 75% in 2026

A single generative-AI query uses roughly 10× more energy than a standard web search, but per-task energy is declining by at least an order of magnitude annually

IEA (May 2026): data-centre electricity surged 17% in 2025; Big Tech CapEx set to rise a further 75% in 2026 after surpassing $400 billion in 2025; supply-chain bottlenecks (grids, transformers, skilled labour) increasingly cap expansion speed

Data centres and the AI energy reckoning

The AI energy shock, one year on

For most of the 2010s, data-centre energy consumption grew much more slowly than data traffic, thanks to extraordinary efficiency improvements in chips (following Moore's Law) and cooling (free air, liquid immersion). The shift to hyperscale cloud providers enabled massive economies of scale. The AI boom that began in late 2022 broke that pattern: large language models require huge amounts of compute for both training and inference, and unlike batch web workloads they cannot easily be time-shifted. The IEA's April 2026 update confirms the scale of the shift — global data-centre electricity jumped 17% in 2025, AI-focused facilities alone grew ~50%, and hyperscaler capex surpassed $400 billion.

What 485 TWh looks like

485 TWh is roughly equivalent to the total annual electricity consumption of Germany, and enough to power every home in the United States for nearly four years. The IEA Base-Case projection of ~945 TWh by 2030 would make global data centres the world's third-largest "country" by electricity consumption, behind only China and the US. The 2025 update also highlights new bottlenecks: grid connections, gas turbines, transformers and skilled labour are all in short supply, so infrastructure constraints may begin to cap growth more than demand itself. Water used for cooling (~660 billion litres/year globally) adds a further environmental dimension often overlooked in energy-only debates.

Global data centre energy consumption: 2010-2025

Global data-centre electricity consumption rose from ~415 TWh in 2024 to roughly 485 TWh in 2025 (IEA Key Questions on Energy and AI, April 2026) — a 17% year-on-year jump, more than five times the global electricity demand growth of 3%. AI-focused data centres grew even faster at ~50% YoY. The IEA Base Case projects a doubling to ~945-950 TWh by 2030, with AI-specific consumption roughly tripling to ~465 TWh.

2010
9.07 GWh/d
2015
10.08 GWh/d
2019
25.49 GWh/d
2024
37.87 GWh/d
2025
44.35 GWh/d
0.0025B50B74B99B201020152019202420252030ESTIMATED9B10B25B38B44B~86B
YearRate (kWh/s)GWh/dayContext
2010105K kWh/s9072.0 GWhPre-cloud era; moderate growth
2015117K kWh/s10080.0 GWhHyperscale growth with efficiency offset
2019295K kWh/s25488.0 GWhCloud boom; hyperscaler efficiency plateau
2024438K kWh/s37872.0 GWhAI demand driving ~12%/year growth
2025513K kWh/s44352.0 GWhAI-focused DCs +50% YoY; hyperscaler capex >$400B; grid/turbine bottlenecks emerge
2030 (IEA Base Case)997K kWh/s86112.0 GWhTotal demand ~doubles from 2024; AI-specific roughly triples to ~465 TWh

When data centre energy became a policy issue

  1. 2010IEA: global data centers use ~194 TWh; efficiency gains keep growth slow despite rapid traffic increase
  2. 2017Hyperscale era takes hold; Amazon, Microsoft, Google begin doubling electricity use every 4 years
  3. 2022ChatGPT launches (Nov 2022); AI inference demand begins exponential growth phase; 240-340 TWh baseline
  4. 2024IEA: ~415 TWh; AI responsible for >10% of data-centre electricity; ~12%/year growth rate
  5. 2025IEA (Apr 2026): ~485 TWh (+17% YoY); AI-focused DCs +50%; hyperscaler capex >$400B; grid/transformer/turbine bottlenecks emerge
  6. 2030IEA Base Case: ~945 TWh (~3% of global electricity); AI-specific consumption triples to ~465 TWh; comparable to Japan's total electricity use

What energy and emissions research shows

YearFindingValueSource
2010IEA: global data center energy ~194 TWh; growth modest; efficiency gains offsetting demand growth194 TWh/year (2010)International Energy Agency
2015IEA: global data center energy ~215 TWh; hyperscale era begins; efficiency improvements continue215 TWh/year (2015)International Energy Agency
2019IEA: ~280 TWh; hyperscale operators double electricity use 2015-2019; cloud adoption accelerates280 TWh/year (2019)International Energy Agency
2022IEA: ~240-340 TWh (midpoint ~290); range reflects different measurement methodologies290 TWh/year (2022)International Energy Agency
2024IEA: ~415 TWh; AI workloads driving acceleration; 1.5% of global electricity415 TWh/year (2024)International Energy Agency
2025IEA (Apr 2026): ~485 TWh, +17% YoY; AI-focused DCs grew ~50%; hyperscaler capex surpasses $400B; physical bottlenecks (grids, gas turbines, transformers) begin limiting expansion485 TWh/year (2025)International Energy Agency
2030IEA Base Case: ~945-950 TWh by 2030 (~3% of global consumption); AI-specific demand triples to ~465 TWh; renewables account for ~40% of corporate PPAs signed in 2025945 TWh/year forecast (2030)International Energy Agency

How the number is calculated

IEA (Apr 2026): global data-centre electricity ≈ 485 TWh in 2025 (up 17% vs. 415 TWh in 2024). The counter shows equivalent household-days of consumption (1 household-day ≈ 30 kWh, US average). 485 TWh ÷ 525,600 min/year × 10^9 kWh/TWh ÷ 30 kWh/household-day ≈ 30.8 million household-days per minute. AI-focused facilities alone contributed an estimated ~155 TWh in 2025 (IEA Base Case), heading toward ~465 TWh by 2030.

Sources: IEA - Data Centres and Data Transmission Networks (archive, superseded by Energy and AI 2025/2026) - IEA - Energy and AI (April 2025 landmark report) - IEA - Key Questions on Energy and AI (April 2026 update): data-centre electricity demand 2024-2030 - IEA - Data Centre Electricity Use Surged in 2025 (news, May 2026): global data-centre demand +17%, AI-focused +50%; Big Tech CapEx >$400 bn; doubles by 2030 to >1,000 TWh. Methodology →

Frequently asked questions

How much electricity do data centres consume globally?
The IEA's April 2026 update estimates global data-centre electricity consumption at roughly 485 TWh in 2025, up from 415 TWh in 2024 — about 1.5-1.7% of global electricity demand. This excludes cryptocurrency mining (~110 TWh/year), which uses similar infrastructure but is tracked separately.
How fast is data-centre energy consumption growing?
IEA data shows roughly 12% annual growth since 2017, accelerating to 17% in 2025. AI-focused data centres grew ~50% year-on-year in 2025 alone — both well above the 3% global electricity demand growth rate. Hyperscalers' capex hit over $400 billion in 2025 and is set to rise another 75% in 2026. The IEA Base Case has consumption more than doubling to ~945-950 TWh by 2030, with AI-specific demand roughly tripling.
Does AI use significantly more energy than regular computing?
Yes, but efficiency is catching up. A single generative-AI query uses roughly 10× more energy than a traditional web search; training a frontier large language model consumes hundreds of GWh. However, the IEA notes that power consumption per AI task is declining by at least an order of magnitude annually thanks to hardware and software advances. Aggregate demand still rises because more people and agents use AI: AI-focused data-centre consumption is projected to triple from ~155 TWh in 2025 to ~465 TWh by 2030.

How the data centre energy estimate is built

Primary source: the IEA's April 2026 update 'Key Questions on Energy and AI', which builds on the landmark April 2025 'Energy and AI' special report. IEA data centre electricity figures are used by governments, investors, and grid operators worldwide as the most rigorous independent benchmark. The 17% year-on-year jump in 2025 is corroborated by hyperscaler capital-expenditure disclosures (Amazon, Google, Meta, Microsoft, Equinix together surpassed $400 billion in data-centre capex in 2025, set to rise another 75% in 2026).