How much electricity does the world use, and how much solar would cover it?

The answer is smaller than most people expect: the world generates about 31,700 TWh of electricity a year, and covering all of it with solar would take roughly 281,000 km² of land — about the size of New Zealand, or 0.19% of the world's land area (matching annual generation, before storage and overbuild). But land area is the least important part of this problem. The real constraints are storage and matching supply to demand across the day. The panel below lets you change the assumptions; after it come the sources, the derivation, and what is wrong with framing the question this way at all.

What the world actually uses today

The first thing to get straight: "electricity use" and "total energy use" differ by almost a factor of six. What people usually call global electricity demand is only the electricity row.

A few things from 2025 are worth remembering: global electricity demand grew by 849 TWh (+2.8%), and solar alone covered 75% of that net increase. Solar generation rose 636 TWh (+30%), passing wind for the first time. Renewables reached 33.8% of the mix, overtaking coal's 33.0% for the first time in a century.

The model: how much a square metre produces in a year

The whole estimate turns on one multiplication. Irradiance sets the ceiling, efficiency and system losses set the conversion, and ground cover ratio sets how much land the same modules have to be spread across:

Land area (km²) = annual demand (TWh) x 1000 / [irradiance x efficiency x PR x GCR]

One counter-intuitive but correct consequence: annual output per kW of installed capacity is independent of module efficiency (it equals irradiance x PR). Higher efficiency shrinks the area but does not change output per unit of capacity. So efficiency is irrelevant when you are talking about installed capacity, and only matters when you are talking about land.

An important calibration: the real capacity factor is only 12%

Working backwards from 2025 actuals: 2,778 TWh divided by an average installed base of about 2.64 TW over the year (about 2.31 TW at the start, about 2.97 TW at the end) gives 1,051 kWh per kW per year, a capacity factor of 12.0%.

That is about 18% below the 14.6% implied by the "global average" assumptions above. The reason is that new capacity has been concentrated in China and Europe, where irradiance is moderate at best, and that curtailment is rising. So the default assumptions are optimistic. The "Calibrated to 2025 actuals" preset sets irradiance to 1,310, which drops land yield to about 92 kWh/m² and pushes every area figure up by roughly 22%.

This 12.0% is derived here from published generation and capacity figures; it is not an officially published number.

How much area each scenario needs

"Share of world land" uses 149 million km² (including Antarctica). Excluding Antarctica, about 134 million km², every percentage rises by roughly 11%.

Why the "fully electrified" row is the meaningful comparison

The 172,000 TWh figure is thermal content, and most of it is thrown away as waste heat during combustion. Electric vehicles are three to four times as efficient as internal combustion engines, and heat pumps three to four times as efficient as boilers, so delivering the same services after full electrification needs 40-45% less energy. Many arguments that solar "cannot possibly replace fossil fuels" skip this step and divide 172,000 TWh directly, which overstates the area by nearly a factor of two.

The 95,000 TWh figure comes from applying those efficiency multipliers generally; it is not a specific institution's scenario value, and carries about ±20% uncertainty.

A sense of scale

• Today's global electricity: 281,000 km² = 3.1% of the Sahara.
• All energy after full electrification: 843,000 km² = 9.2% of the Sahara, or 5.4% of global cropland (about 15.6 million km²).
• For comparison: the technical potential of existing rooftops worldwide alone is estimated at about 27,000 TWh per year, nearly equal to today's entire global generation. Land has never been the binding constraint.

The new curve: AI data centres

Solid bars are actuals, hatched bars are forecasts. The figure on the right is the solar land needed to match annual generation.

Some structural points worth noting: AI-optimised servers already account for about 31% of data centre power draw and are expected to pass conventional servers in 2027. Inference accounts for 60-70% of the compute power draw at large AI companies, with training only 30-40%. Global data centre IT capacity grows from about 103 GW in 2025 to about 200 GW in 2030.

The single most decision-relevant number: matching 945 TWh in 2030 with solar on an annual-energy basis needs about 738 GW of capacity — and in 2025 alone the world added roughly 608-698 GW. In other words, about 1.1-1.2 extra years' worth of global solar installation would cover every data centre on the planet in 2030 (at the global-average 14.6% capacity factor; at the 12.0% actual figure it takes about 899 GW, roughly 1.3-1.5 years' worth). In aggregate terms, AI is nowhere near the main event in the energy system; its impact is concentrated in regional grids and interconnection queues.

What is wrong with the "area" framing

This page concludes that land is not the bottleneck. In fairness, here are the constraints that actually bind:

1. Area was never the bottleneck; storage and grid are. A 100% solar system needs inter-seasonal storage that may cost several times the panels themselves. In practice, once solar passes roughly 40% of supply, the marginal cost of each additional unit rises steeply (more curtailment, lower effective capacity factor, which in turn pushes the area back up). Phrasing it as "a few countries' worth of land" creates the illusion that building it is the whole job.
2. Land is not uniformly available. You can compute 843,000 km², but that does not mean 843,000 km² exists that is simultaneously high-irradiance, low-conflict with ecology and agriculture, close to demand centres, and has grid access. Land near demand centres is the scarce commodity.
3. The 12% real capacity factor is bad news, not noise. It reflects both geography and curtailment, and curtailment gets worse as penetration rises. Using 14.6% or higher for long-range estimates assumes future additions shift towards high-irradiance regions and that curtailment does not worsen. Neither is guaranteed.
4. Demand forecasts have been systematically low. The IEA has underestimated solar deployment for years running, and has been conservative on electricity demand too, especially AI load. The 72,000 TWh for 2050 is more likely a floor than a central estimate.

How much to trust these numbers

• High confidence: 2025 generation, solar generation and installed capacity, cross-checked across several sources.
• Moderate: data centre consumption for 2024-2026. The IEA and Gartner use slightly different boundaries (485 vs 447 TWh for 2025). The 945 TWh for 2030 is the IEA Base Case; Deloitte puts it at 1,065 TWh, so treat the range as 900-1,100.
• Lower: the roughly 1,300 TWh for 2035 is derived, not quoted — inferred from the IEA's statement that its High Efficiency Case reaches about 1,100 TWh, "more than 15% below" the Base Case. The 60,000-90,000 TWh range for 2050 is a wide scenario extrapolation.
• Not included in the model: land for storage, transmission corridors and substations; module degradation (about 0.5%/yr) and the rebuild area implied by a 25-30 year replacement cycle; the wider spacing (and so lower real GCR) needed at high latitudes.

Sources

1. Ember, Global Electricity Review 2026 — 2025 global generation mix. Source
2. Carbon Brief, "Clean energy pushes fossil-fuel power into reverse for 'first time ever'" — 2025 figures by source. Source
3. IEA, Energy and AI (2025) — data centres at 415 TWh in 2024, 945 TWh in the 2030 Base Case, and the scenario definitions. Source
4. Gartner press release (2026-06-10) — 565 TWh in 2026, up 26%. Source
5. S&P Global — global data centre capacity from 103 GW to 200 GW. Source
6. pv magazine — cumulative installed solar approaching 3 TW at the end of 2025. Source
7. Data Center Frontier — Stargate site capacities and how they are powered. Source
8. The global rooftop solar technical potential of about 27,000 TWh/yr comes from a 2021 study in Joule. That is technical, not economically realisable, potential, and this site has not read the original paper directly.

Page generated 2026-10-24 (UTC+8). Energy and electricity figures were verified in August 2026 against 2025 full-year actuals, which were still the latest annual data when this page was written. The AI data centre forecasts were verified at the same time; that field moves quickly, so more recent figures may exist.