01 · The year in electricity

World electricity doubled in a quarter century — and coal still leads.

Seven fuels, 2000–2025, compiled from national statistics. Scrub the timeline or switch the scale — the shape of the transition changes with how you look at it.

World generation, 2025: 31,734 TWh
Clean share, 2025: 42.6%up from 35% in 2000
Coal, 2025: 10,490 TWhjust below its 2024 record of 10,540 TWh

Electricity is not all energy — it is less than one-fifth of the energy humanity uses.

Year2025 · 31,734 TWh

Ember's figures compile national statistics and fill gaps with its own estimates, including distributed solar and off-grid generation missing from official data.

Labeled "compiled": national statistics as compiled by Ember — not meter-by-meter measurements.

Oil is grouped into "Other Fossil" — it has no standalone band in this chart.

Source: Ember Electricity Data Explorer (2026-09-21 vintage), compiled. Method

02 · Trends

Coal still generates more than wind and solar combined — 10,490 TWh against 5,491 TWh.

Seven fuels, seven panels, one shared vertical scale (0–11,000 TWh): a panel that looks flat is genuinely small next to coal, not rescaled. Solar’s curve is the only one bending sharply upward.

Coal · 10,490 TWh in 2025 — still the largest single source

Gas · 6,896 TWh in 2025 — the steady climber

Nuclear · 2,812 TWh in 2025 — roughly flat for a decade

Hydro · 4,432 TWh in 2025 — wind+solar passed it in 2024

Wind · 2,712 TWh in 2025 — from 31 TWh in 2000

Solar · 2,779 TWh in 2025 — passed wind for the first time

Other · 1,613 TWh in 2025 — bioenergy, other renewables, other fossil

Source: Ember Electricity Data Explorer (2026-09-21 vintage), compiled.Method

03 · Sources of generation

In 2025, coal generated more electricity than wind and solar combined — 10,490 TWh against 5,491 TWh.

Each band is a fuel’s measured flow into total world generation — 31,734 TWh in 2025. Wind and solar have grown 171-fold since 2000, and solar edged past wind for the first time; drag the slider back to see how thin those threads began.

Year2025

Drag to re-derive the flows for any year, 2000–2025, from the same measured dataset.

Demand is not shown: at world level it equals generation by construction.

Source: Ember Electricity Data Explorer (2026-09-21 vintage) · measured.Method

04 · Country comparison

China generated 10,578 TWh of electricity in 2025 — more than twice the United States’ 4,520 TWh.

The same twenty grids, ranked three ways: sheer scale, per-capita generation, or how much already runs clean. Switch the measure — the pecking order flips.

Source: Ember Electricity Data Explorer (2026-09-21 vintage); population Our World in Data — Energy Data (population from UN World Population Prospects).Method

05 · Data centers — estimated

No one has ever measured global data-center electricity — every figure here is an estimate.

Three modeling teams estimate data-center electricity on three different scopes: the IEA models the whole world, LBNL and EPRI model only the United States. The chart keeps them in separate groups — and EPRI's range includes crypto mining, which the IEA excludes.

†Second-hand figure quoted from the cited report — seeMethod.

National snapshots — measured and estimated

Ireland

6.969 TWh 2024

Measured

Germany

17 TWh†~2022

Estimated

Singapore

3.71 TWh 2020

Estimated

Vendor context — not demand:Equinix 8.87 TWh · Digital Realty 11.65 TWh (2024) — TOTAL company energy, not data-center-only; not demand anchors.

Source: Based on IEA data (Key Questions on Energy and AI, Apr 2026 update) · LBNL · EPRI · CSO Ireland · Bitkom via Clean Energy Wire · MTI Singapore. Method

Method notes — plain words

IEA (global). Modeled from device shipments and market data — no measured global total exists. Excludes crypto mining. 415 TWh†in 2024 (≈1.5% of world electricity) → 485 TWh†in 2025 → 950 TWh†in the 2030 base case (≈3%).
LBNL (United States). Chip and IT-equipment shipment model. 2024: 192 TWh. 2030 reference scenario 649 TWh with a 521–843 TWh uncertainty band (11.8% of US electricity).
EPRI (United States). 2026 Powering Intelligence outlook. 2030 range 380–790 TWh (9–17% of US electricity) — its scope includes small data centers and crypto mining, so it cannot be compared with the IEA series.
Ireland (measured). CSO official metered-electricity statistics: 6,969 GWh in 2024 (~22% of national metered electricity), data centers identified post-hoc from metered demand.
Germany (~2022, estimated). Industry estimate via Bitkom: 17 TWh†, ≈3% of national electricity.
Singapore (2020, estimated). MTI's official 7% share; the TWh figure is derived by applying that share to Ember's Singapore 2020 total (53.07 TWh) → 3.71 TWh. Not an official TWh figure.
Vendor totals. Equinix 8.87 TWh and Digital Realty 11.65 TWh are total company energy (all fuels), calendar 2024 — not data-center-only. Never demand anchors, never ranked.
What we cut. The UK anchor (source URL unverifiable), S&P/451 figures (source broken), and Big Tech company totals (boundaries not comparable) were removed rather than decorated with badges.

06 · Methodology

Every number on this page can be traced to its source.

Six datasets, two build scripts, one rulebook. Here is what each figure is made of, how sure we are of it — and what we refused to publish.

How confidence is marked

measured: Solid rendering. Reported or compiled by an authority — Ember's national-statistics aggregates, Ireland's metered data.
estimated: Dashed + translucent + badged ESTIMATED. Derived from partial data, e.g. Singapore's TWh from the official 7% share × Ember's total.
modeled: Same dashed/translucent/badge encoding on charts — a modeled figure is never shown as solid "measured" data. Built from models where no measured figure exists (IEA/LBNL/EPRI data-center demand).

Data sources

Dataset	Source	License	Tag
Hero figures	Ember — Yearly Electricity Data	CC BY 4.0	measured
Global fuel mix · 2000–2025	Ember — Yearly Electricity Data	CC BY 4.0	measured
Country mix · top 20 · 2015–2025	Ember — Yearly Electricity Data	CC BY 4.0	measured
Flow sankey · 2025	Ember — Yearly Electricity Data	CC BY 4.0	measured
Population (per-capita denominator)	Our World in Data / UN-WPP	CC BY 4.0	measured
Data-center demand	IEA ·LBNL ·EPRI ·CSO (Ireland) ·Bitkom ·MTI (Singapore) ·Equinix ·Digital Realty	Mixed — IEA data is Non-CC (see note)	modeled (per-anchor tags vary)

Ember data built by scripts/fetch-ember.mjs from the long-format CSV release; population patched byscripts/fetch-population.mjs. Data pulled 2026-09-21.

How each module was built

Hero: Ember's World Total Generation row, 2025; YoY from Ember's own YoY column, cross-checked against (latest − prev)/prev.
Trends: Annual global generation by fuel, 2000–2025. "Clean" = Ember's aggregate (renewables + nuclear); there is no separate Oil category — oil sits inside "Other Fossil".
Sankey: 2025 fuels → total generation → total demand. Demand equals generation by construction (world net imports = 0); Ember "demand" is supply-side accounting, not end-use consumption. No sector split — no honest open source exists for it.
Country: Top 20 countries by 2025 generation (ranking pinned at build time); fuel mix 2015–2025; per-capita = generation ÷ OWID/UN-WPP mid-year population.
Data centers: No measured global total exists — global anchors are modeled, never "measured". IEA excludes crypto mining; EPRIincludes mining and small data centers — never compared side by side without the scope notes. Vendor figures (Equinix, Digital Realty) are total company energy, boundary-labeled, never demand anchors.

Units

All energy figures are in terawatt-hours (TWh). Per-capita figures use kWh/person per IEA convention — the sole exemption, recorded here — and TWh and kWh/person never share a chart, title, or lede.

IEA licensing note

IEA "free product" data is Non-CC periea.org/terms — free to download, not openly licensed. This site quotes IEA only as small conclusion figures (≤5 numerical points: the 415 / 485 / 950 TWh anchors and the 1.5% / 3% shares), each attributed"Based on IEA data" and presented graphically. No downloadable IEA datasets; nothing a reader can reverse-engineer back into the underlying data.

What we cut and why

UK DSIT 2.3 TWh — source URL unpinnable in the research notes; audit-flagged Tier-2. Cut until the original source is pinned.
S&P Global / 451 Research 860→1587 TWh — single-vendor modeled series with incompatible scope; adds nothing beyond the IEA anchors.
"2024 mining ≈160 TWh" — year-misaligned; IEA treats mining separately from its data-center series anyway.
Google / Microsoft / Apple single-company figures— company self-reports are not comparable demand anchors.
IEA country shares (45/25/15) — cut from on-page figures; quote-footprint discipline under the Non-CC license.
Singapore TWh — kept, not cut, but flagged: the 7% share is official MTI (2020); the 3.71 TWh is derived (7% × Ember's Singapore 2020 total of 53.07 TWh). Never presented as an official figure.

Source: src/data/provenance.json (2026-09-21) · full dictionary: src/data/DATA_DICTIONARY.md