Watt It Takes

How much of the world do we have to use in order to save it?

Pick how you would clean up a grid, anywhere on Earth. See what it takes: the land, the materials, the money, the years. Every figure traces to peer-reviewed work or real operating data.

Eight guesses about where you live, then the real numbers. About three minutes.

Simple viewFull calculator
1

Pick your region

shift-click to combine placeslong-press to combine places
↓ Pick a region to start
2

What would you build?

Also replace fossil fuels beyond the grid Cars, furnaces, factories. Their oil, gas and coal are swapped for clean power and clean fuels, so the build gets much bigger.
2

What are we replacing?

↓ Choose one
2030·2040·2050·2060
3

Build your mix

clean vs renewable ⓘ · sliders set the ratio
↓ Pick a starting mix, or move any slider
Firm it: storage and backup for round-the-clock supply Prices keeping the lights on through nights, calms and winter. Variable sources pay in storage and overbuild; dispatchable ones pay by running below full output to follow demand. It is an estimate from published weather studies, not an hour-by-hour simulation or a formal reliability standard, and the worst stretches across many years can need more. Off shows the fair-weather version.
4

What else does the new grid have to carry?

optional; each one adds demand
Capture the carbon this grid still emits Direct air capture, sized to the new grid's own lifecycle emissions. The capture plants run on that same grid, so they have to clean up after themselves too.
End energy poverty Size the build to give everyone here enough electricity for a decent standard of living.
Target per person: decent living with cooling, 4,000 kWh/yr
Also replace fossil fuels beyond the grid Cars, furnaces, factories. Half gets directly electrified by EVs and heat pumps, which are far more efficient; the rest becomes carbon-neutral synfuels, with real conversion losses. Adjust the split.
Power-to-fuel efficiency: 45%

What it takes

What this scenario asks of the physical world, every number live.

⚡The grid

TODAY
AFTER THE BUILD
Today’s clean electricity, and the carbon to 2100

🏗The build

The machines and batteries, counted against what this place runs today, how often each is rebuilt, and how long it all takes.

How the units are sized

How long each source would take on its own, if this place matched the record pace ⓘ for it.

🗺The land

The ground the plants, farms and reservoirs take, measured at real sites and drawn inside this place.

Change how land is counted

⛏The materials

Everything it takes to build, once: concrete, steel, copper and more.

Everything this build uses, melted into one block and set beside the Golden Gate Bridge, to scale. The ruler is marked in tower heights.
Every material, by tonnage and against what the world mines

👷The jobs

Who builds it and who runs it, by source, against the fossil jobs that end.

🧾The bill

What it costs, and how this mix ranks against the reference mixes on six measures.

Ranks are against the reference mixes:

Rank every mix on one measure
Rank them all on
How each mix holds up under stress
Where the money goes
Your share, drawn to scale
Split the whole build evenly among everyone in this place. Left: your share of the materials as one block, next to a person. Right: your share of the land, on a court or field you know.

📣Share

A card built for screenshots and group chats, a link, a one-page PDF, or the data.

🧮Challenge a number Think one is wrong? Change it here

Every objection below is a control, not an argument. Your version travels in the share link.

Advanced mode: unit sizes, capacity factors, land assumptions
💲 PRICES
Batteries cover the first 12 hours. Longer stretches of storage use cheaper long-duration technology.
Money spent over a long build earns interest until the plant runs. It costs a 7-year reactor about 16% more and a 2-year solar farm about 2.5% more.
⚛️ NUCLEAR
%
🌬 WIND
%
☀️ SOLAR
%
💧 HYDRO & GEOTHERMAL
🔥 BIOMASS
🔋 GRID & FIRMING
$/kWh installed
🏗 BUILD
TWh/yr of new clean supply, ignoring the scenario buttons
For educators

A classroom guide is coming soon. It will be a free download with ready-to-run exercises, discussion questions and a teacher's answer key, built around the scenarios in this calculator.

In the meantime, every number here traces to a linked source in Methodology & sources below, and any scenario prints as a one-page handout from the Share section.

Methodology & sources

The gist. These assumptions drive most of the results:

• Build speed is the fastest any country has grown each source, measured against the size of its own grid. Each record is measured against the country’s grid as it grew, and applied to this place’s grid as it grows, so a build many times today’s grid takes correspondingly longer. For nuclear that is 4.1% of the grid a year, set by South Korea from 1985 to 2000. It shows what a determined country has done, not what is politically likely.
• Costs are matched to the place you pick. Every source pays a mature price. A large reactor costs $4,700 per kW in the US (the US Department of Energy’s estimate once 10 to 20 have been built), $4,540 in South Korea, $4,000 in China and $3,550 in India. Wind, solar and batteries get cheaper each year of the build, as NREL projects. Vogtle’s $17,500 and other real costs are in the reactor cost menu, and the cost basis can price every source at today’s cost.
• Land counts wind’s pads and roads and nuclear’s fenced site; the farmland between turbines and under power lines is stated separately. The Land section can count it too.
• Reliability is priced. The more wind and solar, the more storage and overbuild a grid needs. Grids with a lot of nuclear either throttle their reactors or store the surplus, whichever is cheaper.
• Deaths per unit of power include accidents, Chernobyl and Fukushima among them, and air pollution at each country’s own rate.
• Everything counts electricity generated in the place you pick, not electricity used there.

What changes by region, by default:

WhatAdjusted for the place?How
Today's electricity mixYesEach place's own generation (Energy Institute, EIA, Statistics Canada)
Demand growthYesEach place's population projection; Canadian provinces use Canada's rate
Wind and solar output per turbine or panelYesEach place's measured fleet, lifted to new-project levels
Land per plantYesScales with the local capacity factor
Build paceYesThe best grid-share records, applied to this place's own grid
Hydro and geothermal limitsYesEach place's resource; geothermal for US states and assessed countries
Carbon from fuel burned off the gridYesGlobal Carbon Project for each country, minus its power plants; states and provinces share their country's
Carbon intensity todayYesEmber's figure for each country; US states use the US figure
Air-pollution deathsYesMcDuffie et al. for each country; US states use the US rate
Retail power priceYesEach place's own price
Nuclear bansYesEach place's laws, flagged
Which large reactor is builtYesThe 925 MW CANDU in Canada, the 700 MW CANDU 6 in Romania and Argentina, India’s own 700 MW heavy-water reactor in India, the 1.1 GW reactor elsewhere
Solar, wind, battery and large-reactor costsYes, unless switched offThe price table under Prices by region, below
SMR, microreactor, custom reactor, Korean, Chinese and Indian design costsNoEach design’s own figure everywhere
How costs change over the buildNoNREL’s projected declines, applied to each place’s price
Offshore wind costNoNo regional figures
Geothermal, hydro, biomass and gas-with-capture costsNoOne figure everywhere
Materials and jobs per plantNoThe same machine is the same machine
Reactor build time and plant lifetimesNoSet per technology
Storage and backup needsNoSet by the mix, not the place

Where the thumb could be. Read this before arguing with the results: Defaults that favor nuclear: large reactors are priced at the Department of Energy’s Nth-of-a-kind estimate ($4,700/kW, the cost once 10 to 20 are built), though the NREL baseline used for every other source never brings them below $6,950/kW, and following it for reactors too raises the cost per metric ton of every reactor-heavy mix, as does charging first units first-build prices (learning by building), most for small builds; build paces are all-time records; nuclear's site land is the measured mean of operating stations rather than a padded round number; nuclear's 80-year life is a renewed license term, longer than the 60 years today's large reactors are designed for; SMRs and microreactors inherit the 90% capacity factor of mature large reactors, though no small reactor yet has a commercial operating record; the CANDUs borrow the large reactor’s price, though no Monark has been built and neither has a mature-fleet estimate; and the default renewables-led mix is the IEA’s world mix, 46% solar, which costs more and takes more land in cloudy, northern places than a mix chosen for them would. Defaults that work against nuclear: the mortality rate includes Chernobyl and the Fukushima evacuation; wind, solar and batteries take IRENA’s and BloombergNEF’s 2025 market prices, below NEA and EPRI’s engineering estimates (US solar $1,150 against $1,420/kW, batteries about $270 against $390 per kWh); once reactors and wind and solar together pass about 70% of the grid, the grid pays for the daily swing, by load-following or storage, whichever costs less; uranium is assumed 100% freshly mined, no recycling; wind and solar are lifted to new-project performance everywhere, including regions whose grids curtail heavily today; above 80% wind and solar, storage and overbuild follow a least-cost study that assumed perfect transmission across the whole US, the most favorable geography there is; and the storage curve treats the grid as mostly inflexible, which is the assumption Jacobson's roadmap disputes hardest, so his own storage and overbuild figures are selectable in Advanced and are documented above. All but two of these are one click away from their opposite in Advanced. The two that are not, the death rates and the least-cost storage top, both work against nuclear.

Assumptions worth knowing: Each source builds at a record set by a country that was mostly building that one source, and all sources build in parallel, which no country has done for every source at once. Build years divide everything built, including the overbuild that will be curtailed, by a pace measured on delivered generation. Nuclear has a minimum build time of seven years; other sources have none. Today’s reactors are kept to the target year and each is charged one long-term-operation refurbishment, about $900 per kW: the median of what NEA and EPRI’s Costs of Generating Electricity 2025 reports for 20-year programs ($391 in the US, $848 in France, $946 in Sweden, $1,267 in Switzerland), spread over the 20 years it buys. Today’s wind and solar are charged for rebuilding as they wear out. That work runs alongside the new build and draws on the same industry, so it counts toward the build time by the share of that industry it takes: a wind repower, which keeps the tower and foundation, about 85% of a new turbine’s work, and a solar repower about 120%, the same ratios as its cost. The wind figure is NREL’s estimate for new rotors and drivetrains on old towers (Lantz et al., 2013), which PacifiCorp’s actual repowering cost matched; the solar figure is the middle of the only published estimate, 10 to 30% above a new farm, since old panels have to come off and old racking be refitted. Today’s hydro, geothermal and biomass plants are assumed to keep running with no refurbishment charge, which flatters grids rich in them a little. Gas with capture pays to pipe and store its CO₂; the tool checks for rock that can hold it in US states only. Above 80% wind and solar, storage and overbuild follow the least-cost mix from a study that assumed perfect transmission across the whole US, the most favorable geography there is; smaller grids would need more of both. The energy-poverty floor compares generation per person with a consumption floor, so grid losses make it slightly generous.

Rounding: Modeled results are shown to three significant figures at most, since no input here supports more. Measured figures, such as today’s generation and published prices, are shown as their sources give them.

What this model leaves out: Taxes, subsidies, and any cost of money above the real rate used for interest during construction. The headline investment is capital plus interest during construction. Fuel and upkeep stay out of it and appear in the money breakdown and in the cost per metric ton.

What the tool counts, and where its data comes from

Inputs carry their own dates. Electricity data: Energy Institute Statistical Review 2026 (full-year 2025). Solar and wind costs: IRENA, 2025 report (July 2026). Gas prices: World Bank, October 2026 forecast. NREL cost projections: ATB 2025. Built by Generation Atomic, a 501(c)(3) growing the movement for the atomic energy of today and tomorrow. Questions or corrections: gena@generationatomic.org.

Today’s grid, calibrated to Ember: The lifecycle figures above apply to the plants a scenario builds. Today’s plants vary: 490 g/kWh for gas describes a modern combined cycle, and US coal burns at about 1.04 metric tons of CO₂ per MWh. With one factor per fuel the World read 403 g/kWh against Ember’s published 458, which understated the carbon every build avoids by about 14%. So each country’s fossil terms are scaled by one factor until its grid matches Ember’s published 2025 lifecycle intensity (2024 where 2025 is not out), the method of Zeke Hausfather’s wind-and-solar counterfactual. The factor is held between 0.75 and 1.6, his plausibility band. Grids under 5% fossil, Canadian provinces (Canada’s factor is driven by prairie coal) and Ukraine keep the global factors; US states take the US factor of 1.10. As a check, the EIA’s plant data put the US grid at 349 g/kWh at the smokestack, and the calibrated lifecycle figure, 384, sits 10% above it, about what upstream emissions, chiefly methane leaking from gas supply, should add.

Geothermal in today's grid: Existing generation now separates geothermal from biomass. For countries this uses OWID's non-biofuel other-renewables series, which is over 98% geothermal worldwide (the remainder is wave and tidal); for US states it uses EIA's utility-scale geothermal. This matters because the two are nothing alike: geothermal is firm, low-carbon (19 g CO₂/kWh) and safe, while biomass carries roughly 230 g and far more land. It matters most for grids like Iceland, Kenya, New Zealand and the Philippines, where what looks like biomass in aggregated data is overwhelmingly geothermal.

Generation, not consumption: Every figure here is built on electricity generated inside a region, not electricity consumed there. Regions that import much of their power look small: the District of Columbia generates almost nothing because it buys nearly all its electricity from neighboring states, and any net importer shows a grid smaller than what it actually uses. The model asks what it would take to build clean generation equal to what a region produces from fossil fuels today, which is the honest question for a builder. Cross-border imports and exports, and the transmission that carries them, are not modeled.

Future demand: by default this scales with the finish year on a curve anchored at ×1.45 by 2035 and ×2.00 by 2050, interpolated linearly. The 2035 anchor sits inside IEA WEO 2025's finding that peak electricity demand rises around 40% by 2035 under Stated Policies; the 2050 doubling is the range deep-electrification studies land in once transport, heat and industry move onto the grid. That curve is for the world. Each place’s figure scales it by that place’s own population growth relative to the world’s, since demand per person rises everywhere but people do not: Germany gets about 1.6× by 2050, the US about 1.9× and Nigeria about 2.6×. That curve is a default, not a claim: Advanced → Future demand basis switches it to flat at today's level, or to any growth figure you want, and the implied compound annual rate is shown next to it. Demand growth is applied to the whole grid before the mix is sized, so it scales land, materials, capital and build time together.

Population at the target year: The build finishes in the year you pick and has to serve the people alive then, so the floor is sized on each country's projected population rather than today's. Region populations are scaled by UN World Population Prospects 2024 medium-variant ratios (via Our World in Data), interpolated between 2025, 2035, 2050 and 2060. This matters more than it looks, because the growth is not evenly spread: the world gains 17% by 2050 while Nigeria gains 51%, Tanzania 84% and DR Congo 93%, and those are precisely the countries sitting below the floor. Sizing on today's head-count understated the global energy-poverty uplift by about a third, 12,258 TWh against 16,365. Two boundaries worth stating. The projection deliberately does not touch the electricity demand multiplier, because that curve is calibrated to IEA total-demand outlooks which already carry population growth inside them, and applying it twice would double-count. And every US state rides the national projection rather than its own: state-level projections exist, but they diverge from each other far less than the 2050 uncertainty band does, and carrying 51 separate curves would imply a precision this does not have. Advanced can hold population at today's level, and a shared link says when that has been done.

Finishing the build by a date: The slider in Step 2 sets a completion year, and the year sets how much demand the grid has to serve when it opens. The curve runs through three published points: today at the 2025 end, 1.45× by 2035, which is the IEA World Energy Outlook range of 40 to 50% global electricity demand growth, and 2× by 2050, where full-electrification scenarios converge. Years in between are interpolated, which is close to the shape the underlying projections draw anyway. Data-center and AI load is already inside those trajectories, since the IEA builds its demand outlook on its own AI forecast, so there is no separate data-center toggle to add on top and double-count. Growth lands on the electric build target; the fuels-beyond-the-grid target stays tied to today’s fuel use. Picking a date cuts both ways, which is the point of making it one control. A later year buys build time and costs you a bigger grid. An earlier one shrinks the grid and shortens the runway, and the note under the slider says plainly whether the mix you have chosen can be built in the years remaining, at the pace that mix assumes.

Plain-language glossary: A TWh (terawatt-hour) is a billion kWh, enough electricity for about 90,000 US homes for a year. Capacity factor is the share of the time a plant effectively runs at full power (nuclear ~90%, solar ~20%). Overbuild means building more wind and solar than demand needs so there's enough on weak days; the surplus on good days is curtailed (thrown away). Firming is everything that turns variable power into reliable power: storage, overbuild, flexible demand. Load-following is a plant ramping down when demand drops. France's reactors do it nightly. Overnight cost is construction cost excluding financing, as if built overnight. NOAK (Nth-of-a-kind) is the cost after a design has been built enough times to get good at it. gCO₂/kWh measures how dirty each unit of electricity is, over the full life cycle.

The mixes: presets, statutes and resource ceilings

The IEA Net Zero 2050 preset (the default) is the world electricity mix in 2050 under the International Energy Agency’s Net Zero Emissions by 2050 Scenario, from the World Energy Outlook 2025: 9% nuclear, 29% wind, 46% solar, 9% hydro, 1% geothermal, 4% bioenergy and 2% fossil fuels with carbon capture and hydrogen. It is mapped to the sources this tool models, with three substitutions: bioenergy with carbon capture is shown as ordinary biomass, hydrogen and fossil fuels with capture as gas with capture, and the scenario’s 0.3% of unabated gas is rounded away. It is a published, global pathway built to hold warming to 1.5°C, not the cheapest mix for any one grid, and it leans on wind and solar more than most national studies do. The Sweden, Ontario and France presets are real grids that already run mostly clean, at 28%, 49% and 70% nuclear. The all-nuclear, all-wind and all-solar presets are bounding thought experiments, not proposals; they exist to show where the edges are.

The Jacobson 100% renewables-only preset: Mark Z. Jacobson at Stanford is the most prominent researcher arguing the world can run on wind, water and sunlight with no nuclear at all, so his roadmap sits in the mix list as he published it rather than as a strawman. The shares come from Table 7 of the 150-country study (Jacobson, Sambor, Fan, Mühlbauer & DiBari, 2025), which reports the share of all-sector end-use demand each generator meets across a LOADMATCH run that matches supply to demand every 30 seconds through 2050 to 2052. Onshore wind at 33.19% and offshore at 11.53% become the 45% wind here. Rooftop PV 15.16%, utility PV 33.63%, concentrated solar 0.035% and solar thermal heat 0.416% become 49% solar. Hydro 4.52% with wave and tidal folded in becomes 5%, and geothermal electricity plus geothermal heat becomes 1%. Nuclear, bioenergy, carbon capture and blue hydrogen are excluded, which he states openly in every paper and defends on grounds of cost, build time, weapons proliferation and meltdown risk. The preset shows nuclear at zero because that is his position, not because this tool decided it for him.

Where Jacobson and this tool disagree: Storage and overbuild, and the gap is enormous. Run his mix across the whole world grid under this tool's default curves and it asks for roughly … TWh of storage and …× overbuild. His own simulation of essentially the same mix reports 40.4 TWh of electricity storage (32.3 TWh of batteries, 8.0 of pumped hydro, 0.12 of concentrated-solar phase-change) and 16.6% curtailment, which is 1.20× overbuild. Advanced carries a switch, “Storage & overbuild basis”, that swaps this tool's curves for his figures so the same mix can be read both ways. Three assumptions do most of the work. He treats 5,356 GW of his 8,962 GW of average demand as flexible, mostly hydrogen production, industrial process heat and building heating and cooling, so well over half the load can chase the weather instead of being chased by it. He builds about 770 TWh of heat and cold storage, underground thermal stores, hot water tanks and firebrick heat batteries, roughly twenty times his battery fleet, and this tool neither sizes nor prices any of it. He also counts existing hydro reservoirs, 1,588 TWh of them, as dispatchable storage discharging at up to 1,262 GW. This tool's curves come from studies that model a mostly inflexible grid buying reliability with batteries and overbuild, the conservative reading. Both positions are defensible. The useful summary is that his low storage number is a claim about demand flexibility as much as a claim about supply.

What Jacobson assumes about demand and land: His 2050 world uses 54.2% less end-use energy than business as usual. The breakdown is 37.0 points from the higher work-to-energy ratio of electricity over combustion, 10.6 from no longer mining, shipping and refining fuels (uranium included), and 6.6 from efficiency beyond the reference case. Electricity demand still rises 1.85×. Set this tool to rebuild the whole grid, hold demand at today’s level and electrify everything beyond the grid directly, and it sizes about 82,400 TWh a year against his 78,500, roughly 5% apart. The two are not measuring the same year, since his figure follows population and economic growth out to 2050 while this tool electrifies today's energy, so read that as a sanity check rather than as agreement. Land is where the two genuinely part. He reports 0.18% of the 150 countries' land as footprint and 0.39% as spacing between turbines, which he treats as farmable and multi-use. That spacing works out at about 15 km² per TWh a year of new onshore wind. This tool counts wind’s pads and roads by default and states the spacing separately; counting shared land in the Land section uses 99 km²/TWh, the roughly 3 W/m² array density in the wind-resource literature. One convention is worth a factor of six on the headline land number, so the choice sits in front of you rather than buried in the model. His capacity factors are well above this tool’s measured ones: onshore wind 38.1% against a world fleet that actually delivers 25.6%, blended PV 20.5% against 15.0%. Some of that gap is real, since his is a 2050 fleet of new machines on chosen sites while the measured figure carries today’s hardware and today’s curtailment. Even so, IRENA puts projects newly commissioned in 2025 at 36% onshore wind and 16.5% solar, so his solar assumption sits above what the best current year of building achieves. Geothermal is the outlier at 61% against 85%.

The Ontario-style preset mirrors the clean half of the grid Ontario actually runs: roughly half nuclear, then wind, solar, about a quarter hydro, no geothermal and a little biomass. Ontario decarbonized to over 90% clean and phased out coal entirely by 2014, largely on the strength of its CANDU fleet. One honest gap: Ontario still burns natural gas for a single-digit share of its electricity, mostly for peaking, and the mix builder here only offers clean sources, so that gas is renormalized across the clean ones rather than shown. Read the preset as “Ontario’s clean mix, scaled to carry the whole grid” rather than as a literal copy of its dispatch.

Two presets built for the region: Grow today’s mix scales up the clean sources a region already runs, nuclear, wind, solar, hydro and geothermal, all by the same factor. That is the mix that keeps the largest growth multiple any one source needs as small as possible. Biomass is held at today’s level, because fuel and land run out long before turbines or panels do, and hydro and geothermal still stop at their resource ceilings. Fastest to build searches nuclear, wind, solar, hydro and geothermal shares, the last two only as far as the place has rivers and hot rock to build on, for the soonest finish at record national build pace, the tool’s own measure of what has been shown possible; where mixes tie within 0.05 years, which happens when several hit the floor of one reactor’s construction time, it takes the cheaper. The two answer different questions. Growth multiples are not comparable across technologies: solar has grown fivefold in many countries, nuclear almost never, and a source a region does not run today can never appear in the first preset however fast it could be built. So the first shows what more of the same looks like, and the second what the build-pace records allow. In Germany the first takes about … years and the second, about …% nuclear, about …. Both rebuild themselves when the region changes, and the first is unavailable where a region has almost no clean power to grow.

Unabated gas in the mix: Gas without carbon capture is a slider because almost every real grid that calls itself clean still leans on it. Choosing it does what it does in reality: emissions rise (490 g CO₂/kWh against nuclear’s 6 and solar’s 29) and the battery bill falls, because storage and overbuild are sized on the wind-and-solar share net of everything that can follow demand. That credit goes to every dispatchable source, hydro, biomass and gas with capture included, not to fossil gas alone; hydro counts as fully dispatchable, which is generous to run-of-river grids. Nuclear and geothermal run steadily here, and their load-following is costed separately. The treatment is generous to gas in crediting firming by energy share rather than available capacity, and harsh in charging its full lifecycle carbon. A mix that is a fifth gas is a mostly-clean grid with a fossil backstop, and the carbon number says so.

Where new nuclear is illegal: Eight US states hold statewide moratoriums as of July 2026: California, Hawaii, Maine, Massachusetts, Minnesota, Oregon, Rhode Island and Vermont. Some are outright bans, some need a public vote, some wait on a federal waste repository. Connecticut and New York restrict only certain sites, so the tool names the limit without calling the state closed to nuclear; New York’s Master Plan (November 2026) aims for 5 GW of new nuclear on top of the 3.4 GW running. Abroad the tool flags bans in Australia, Austria, Denmark, Ireland, New Zealand and Switzerland, and phase-outs in Germany, Spain and Taiwan. The tool flags the restriction and builds the scenario anyway, because these laws keep moving: six US states have repealed theirs since 2016, Illinois and New Jersey in 2026, and Italy’s parliament passed the law ending its ban in September 2026, nearly four decades after the 1987 referendum. To see a scenario without nuclear, set it to zero in the mix. This is the most perishable data in the tool, so check the date before quoting it.

Hydro potential ceiling: Because hydro cannot simply be scaled up anywhere the way solar panels or reactors can, the tool checks each scenario's hydro against a per-country feasibility ceiling: the sum of a nation's existing hydro generation plus its remaining economically feasible potential (from Gernaat et al., Nature Energy 2017, a high-resolution assessment of 3.8 million potential sites, cross-checked against IHA and IEA current-generation figures). When a mix asks for more hydro than that ceiling allows, a warning appears in the results. Hydro is held to its ceiling by default, and when geothermal is held to its own, its surplus moves to the other sources and to hydro only as far as hydro’s ceiling allows; whatever no source can take is left unbuilt, and the results say how much of the grid the build covers. The ceiling is a physical-and-economic limit, not an environmental endorsement; many feasible sites carry serious ecological and displacement costs. For US states the ceiling is each state’s current hydro plus two Oak Ridge National Laboratory assessments for the Department of Energy. That is a technical ceiling rather than an economic one, so US states get a looser limit than countries do, which cuts against this tool’s framing rather than for it; the method is under Sources and reference figures. One thing the tool does not model: hydro's seasonal and drought variability. Real hydro output swings with rainfall and snowmelt, and dry years (recent examples in Brazil, Zambia, and China) have forced grids back onto fossil backup. The tool treats hydro as steady annual generation, so it modestly overstates hydro's firmness in drought-prone regions.

Geothermal resource ceiling: Conventional geothermal needs heat, water and permeable rock in the same place, so like hydro it cannot be built wherever a scenario asks. US states take the US Geological Survey’s national assessment, so only thirteen states, nearly all in the West, have any conventional resource; countries take their own national assessments, a country with no geothermal plants today gets no conventional resource beyond what it runs, the World takes Stefánsson’s world assessment, and countries that run geothermal but have no assessment here are left uncapped. The figures are under Sources and reference figures. Above a ceiling, the surplus moves to the other sources in the mix; where geothermal is the only source, the tool builds only the share that exists, or says none can be built. Enhanced geothermal systems, which create permeability in hot dry rock, could reach far more, 517,800 MWe on USGS’s provisional western-US estimate, but NREL’s Annual Technology Baseline (2024 edition) prices them well above conventional plants and their cost is not modeled here. The cap can be switched off, which treats the excess as engineered geothermal at a conventional plant’s cost and is flagged as a changed assumption. The zero rule may understate a few countries with volcanic resources not yet developed, in the Andes and the East African Rift, and switching the cap off restores them.

How long it takes to build

Whole-grid growth record: for any grid that has to at least double, the results add a yardstick beside the per-source paces: the most any country of 5 million people or more has multiplied its whole electricity supply, from a grid of at least 10 TWh, over the same span (Our World in Data from the Energy Institute and Ember, 1985 to 2025). Laos holds the five- and ten-year records and Vietnam the longer ones, 26 times over from 1993 to 2023. It answers a different question from the per-source records, how fast a whole grid has ever grown rather than how fast each source has, and it changes no result.

Lead times: Nothing delivers power until its first plants are built, so each source’s build time is its lead time plus its years at the record pace: 2 years for solar, 3 for wind and gas, 4 for hydro, geothermal and biomass, which are the EIA lead times the interest during construction already uses, and for reactors one unit’s build, 7 years for large reactors, 4 for an SMR and 3 for a microreactor. The time bars show the lead time as the darker start of each bar.

Build pace, the default: How fast has any country grown each source relative to the size of its own grid? That is what an advocate means by “if we scaled up like Korea or France did.” The records come from a per-grid record book anyone can rerun (gridbook.py, on Our World in Data’s 1985–2025 generation series): the largest rise in each source over a window, divided across its years and by the logarithmic mean of the country’s grid at the start and end, for countries of at least 5 million people with an average grid of at least 37 TWh. The best sustained 15 years are nuclear 4.1% of the grid a year (South Korea, 1985–2000), wind 2.1% (Denmark, 2000–15), solar 2.0% (Chile, 2010–25) and hydro 5.2% (Vietnam, 1998–2013); the five-year sprints are nuclear 8.3% (South Korea, 1985–90), wind 3.8% (Finland, 2019–24), solar 3.8% (Pakistan, 2020–25) and hydro 8.8% (Laos, 2018–23). The logarithmic mean is exactly what a share-of-grid rate works out to while the grid grows, so records are measured and applied the same way, to the log-mean of the region’s grid today and at the finish. Dividing by the average grid would flatter records set on booming grids, and applying records to the finished grid would let a huge build earn a proportionally faster pace. A build of five years or less uses the sprint and one of fifteen or more the sustained rate; in between, the total a country can add grows in a straight line from one to the other, so a bigger build never finishes sooner than a smaller one. For nuclear the sprint adds 41% of the grid in five years and the sustained record 62% in fifteen, so past a sprint’s worth each extra year adds only about 2% of the grid. A share of the grid measures a scale-up against the system being transformed and applies one rule to every technology; what it leaves open is whether an industry can grow that far, and where a build would add a source faster than the whole world ever has, the results say so. Geothermal and biomass have no series of their own and keep the per-person record. Advanced keeps three alternatives: the same records counted per person, the median of every serious national effort, and the world industry’s best years.

Build pace per person (in Advanced): The same national records can be counted per person instead of as a share of the grid: the largest 5-year rise in a source’s annual generation, divided across those years and by the country’s population at the time, or the best sustained 15 years for builds longer than five. Records come from the 2026 EI Statistical Review (full-year 2025), for countries of at least 5 million people, with 3-year smoothing on hydro so a wet year does not score as construction; geothermal and biomass come from Our World in Data’s composite, since EI reports them together. The 5-million floor hides one record: Norway built hydro at 945 kWh per person a year in 1978–83 with 4.1 million people, so the book’s 399 is the best it managed once large enough; Iceland’s 3,189 at 0.3 million people shows why the floor exists. Per-person pace gives bigger regions bigger workforces automatically. Sources build in parallel, and the years to build are the slowest lane. The record book, generated from the same table the model reads:

Two guards on the per-person basis: Per-person records break at both ends of the scale. A very populous country would outbuild the world: at India’s 1.43 billion people they had it adding about 600 TWh of wind a year, more than the whole world installs. So each country’s yearly build is capped at the largest share of the world’s additions any one country has taken: nuclear 41.7% (the US, 1986–91), wind 59.6% and solar 47.5% (China, 2020–25), hydro 88.8% (China, 2010–15), at most about 72 TWh of nuclear a year, 133 of wind, 185 of solar and 114 of hydro, scaled by the pace multiplier. The World itself is not capped, and geothermal and biomass have no industry series to cap against. At the other end, a grid bigger per person than any record holder’s would be paced absurdly slowly: export hubs like Wyoming, North Dakota and Quebec, the Gulf states, Iceland and Norway, and for solar the US itself. So each region builds at whichever is faster, the record per person or the same holder’s record as a share of the region’s grid. Keeping one holder either way keeps this conservative; mixing the fastest per-person and fastest per-grid holders would let every region pick whichever flatters it.

Pacing by the world's industry instead of a national record: The record book above answers "how fast did one country go, per its own people?" Scaled to 8.1 billion, that question implies a supply chain far larger than any that has existed: nuclear at Sweden's pace means the world adds 6,205 TWh a year, and the most it has ever added is 171. The fourth pace basis in Advanced asks the other question. Measured the same way, as the rise in annual generation divided across five years, from the 2026 EI Statistical Review, the world's best five years are 171 TWh/yr of nuclear (1983–88), 224 of wind and 390 of solar (2020–25), and 129 of hydro (2009–14). Those peaks sit in different eras, so they do not add: the best five-year window of the combined series is 669 TWh/yr, in 2020–25, and that is the figure any "most the world has ever built" comparison has to use. Applied to a single country, this basis hands that country its population share of the world industry, which understates a rich country commanding more than its head-count share of the supply chain and overstates a poor one. For the World it is exact. Geothermal and biomass stay on their national records because EI reports them in one combined column with no usable split.

How fast the industry grows, per source: Freezing the world's industry at today's throughput for a forty-year build is pessimistic, and the fix has a trap in it. All low-carbon additions together grew 13.25% a year across the decade, from 193 TWh/yr in 2010–15 to 669 in 2020–25. That figure is overwhelmingly solar. Of the +476 TWh/yr of extra annual build between those windows, solar is +345 of it, 73%, and wind +126. Hydro went backwards. So each lane now carries its own measured growth rather than the fleet's: nuclear flat, wind 8.7% a year, solar 24.2% and hydro flat. Nuclear's 2010–15 additions were negative after Fukushima, so there is no growth rate to measure off that base, and its +6% a year since 2020 is a rebound out of that trough rather than a trend. Hydro's additions are genuinely falling, at −12.3% a year, but running an industry backwards for forty years is no more defensible than running it up, so it is held at its record. Advanced can still apply one uniform rate to every source, including the composite 13.25%, and a shared link says when that has been done. The results also state how large the clock-setting industry ends up under whatever rate it carries, because at 24% a year an industry doubles every three years and compounding is how a long build quietly becomes a claim about a supply chain nobody has costed.

Why geothermal's build pace is so low: At 100 kWh per person per year, geothermal has the weakest record of any source in the table, roughly a seventh of nuclear's. That is a fact about demand, not about difficulty. Installed geothermal worldwide has been small enough that, as the Society of Petroleum Engineers puts it, few drilling contractors or service companies could sustain themselves on geothermal work alone, so no country has ever mounted a crash program. Two things suggest the ceiling is much higher. Iceland added new geothermal generation at roughly 1,970 kWh per person per year between 2004 and 2009, about twenty times the figure used here and more than double nuclear's record, though from a population of 300,000 that does not straightforwardly scale. And the constraint is drilling capacity, which is fungible with oil and gas: a typical geothermal well supports 6 to 10 MW, and enhanced-geothermal developers are now completing deep, 400 °F-plus wells in about three weeks using standard oilfield rigs and shale-basin drill bits, on a demonstrated 35% learning curve. Redirecting something like a third of global oil-and-gas drilling to geothermal would sustain roughly the pace assumed here, so 100 is best read as "a serious but not wartime effort" rather than a physical wall. The tool keeps the historical figure because every other technology's pace is derived the same way, and hand-raising one source on a promissory argument would undermine the comparison. Use the mobilization multiplier in Advanced to explore faster.

Reactor size and build time: Unless you pick a reactor in Advanced, the tool sizes it to the grid, because a grid has to survive losing its largest unit. Islanded grids such as the territories, Hawaii and Alaska keep any one unit to 15% of peak demand. Regions inside a larger interconnection allow up to half their own peak. Peak is the grown grid at a 60% load factor, and the choice is among a large reactor, a 300 MW SMR and a 15 MW microreactor, or a custom reactor whose size, capital cost and build time you set yourself (its materials, land and operating costs are interpolated between the three standard sizes). The large reactor is the 1.1 GW light-water unit, except in Canada, where it is the 925 MW CANDU, Canada’s own design and the one its utilities are weighing, in Romania, where it is the 700 MW CANDU 6 Romania is completing at Cernavoda, and in Argentina, which runs a CANDU 6 at Embalse and has planned another; in India it is India’s own 700 MW heavy-water reactor. A selection that mixes these places with others gets the 1.1 GW unit. Picked by hand, the menu also offers a 1.4 GW Korean and a 1.1 GW Chinese design, each at its own country’s cost (both CANDUs: see What it costs). Build time for reactors has one limit the per-person records cannot see: it never exceeds the number of units times one unit's build, as if only one were under construction at a time. One unit is taken as 7 years for large reactors, the CANDUs included, 4 for an SMR after OPG's Darlington project, and 3 for a microreactor, which is a vendor claim. Provincial wind and solar capacity factors are Statistics Canada's 2024 generation over installed capacity; Ontario's solar figure is withheld because the two tables imply 9.3%, which no PV fleet at that latitude delivers.

What it costs

Cost basis: Every capital cost is entered as an overnight cost in 2024 US dollars per kilowatt, the cost to build as if done in a day, before grid connection and transmission, and quoted the same way for every technology. The tool then adds interest during construction by default (below), which weighs most on slow builds like reactors. By default every source pays a mature price. For a large reactor that is the Nth-of-a-kind cost reached after 10 to 20 units: the DOE Liftoff figure of $4,700/kW, within 2% of the $4,625/kW from MIT’s Center for Advanced Nuclear Energy Systems (Shirvan, 2024). Both are built down from Vogtle 3 and 4’s actual costs, so they share a starting point. Wind, solar and batteries are built by the thousand and still getting cheaper, so their mature price follows NREL’s Annual Technology Baseline 2025, Moderate scenario, averaged over each year from now to the target year, because the money is spent across the build rather than on its first day. Over the years to 2050 that puts onshore wind about 27% below today’s price on average, utility solar about 32%, rooftop solar about 38% and batteries about 27%. Offshore wind, geothermal, biomass and gas follow the same baseline; new hydro is flat in it, so hydro does not move. The reactor price is still the friendlier of the two. Vogtle 3 and 4 realized roughly $17,500/kW as a genuine first-of-a-kind after a supply chain and workforce that had stopped building reactors. MIT puts the next AP1000 at $8,300 to $10,375/kW, and Idaho National Laboratory (2025) models the learning curve at $10,000, $7,800 and $6,200/kW for the second, third and fourth two-unit plants, with build times falling from 7 to 5.5 years. The same NREL baseline that prices wind and solar here puts large reactors at $10,800/kW in 2030, falling to $6,950 by 2050, in 2024 dollars. If you are arguing about the plant being built now rather than the fiftieth, the cost basis setting prices every source at today’s cost, with large reactors at MIT’s $8,300/kW wherever the tool otherwise uses the US figure, scaled to local building costs as on every basis; China, South Korea and India keep their own, which are current costs from plants built in series. Another setting follows NREL for every source, reactors included: large reactors then average about $9,300/kW over a build to 2050, and SMRs and microreactors follow NREL’s paths for them. Every path is averaged from November 2026, the date all results are computed from, so they change only with a new version of the tool and never with the reader’s calendar. A price you set yourself is used as set on every basis. On today’s prices SMRs and microreactors, whether sized for a small grid or picked by hand, pay their first units’ prices, $14,700 and $25,000/kW, as under learning by building; NEA and EPRI’s median for a near-term SMR is $13,021. Custom reactors and the Korean, Chinese and Indian designs keep their own costs on every basis, and the CANDUs take the large reactor’s price on each.

Learning by building: Reactors, gas plants with carbon capture and carbon capture plants have few or no units built in the West, so another cost basis charges the first ones more. The first unit pays a first-build price and each later one costs the same percentage less, reaching the mature price at the tenth unit and staying there. DOE puts that point at 5 to 10 standard units (2024) or 10 to 20 (2023), and MIT’s mature AP1000 is about the twentieth. Units are counted within the build, as one program, because learning stays mostly with the builder: a state building two reactors pays close to first-build prices, while a national fleet averages near the mature price. First-build prices: a large reactor starts at its next-plant price, $8,300/kW where the US figure applies (MIT), since Vogtle 3 and 4 were the AP1000’s first. The 925 MW CANDU has never been built, so its first two units pay what Vogtle’s did, about $15,000/kW overnight in 2024 dollars (MIT), scaled to each country’s building costs like every large reactor, and then follow the AP1000’s path; the CANDU 6, built eleven times, starts at the next-plant price like the AP1000. An SMR starts at OPG’s first BWRX-300, C$6.1 billion plus a quarter of the C$1.6 billion shared works, about $14,700/kW with interest taken out; a microreactor at $25,000, the top of its range, since none runs commercially. Gas with capture starts at about $6,900/kW, the £4 billion of contracts for Net Zero Teesside Power’s 742 MW, the first commercial plant; some reports count part of the shared CO₂ network in that figure. Carbon capture plants start at Stratos’s $2,600 per metric ton a year, in plants of a million metric tons a year. China, South Korea and India keep their own reactor costs, which already come from series builds, as do the Korean, Chinese and Indian designs wherever they are picked. Wind, solar, batteries and every other source with a mature fleet are priced as on mature prices, and a price you set yourself, or one global price, is used as set. Build times do not change, because reactors already take the first plants’ seven years on every basis. Two of these choices lean: the large reactor carries Vogtle’s learning everywhere the US figure applies, which favors nuclear outside the US, and the SMR gets cheaper more slowly than OPG’s own estimates, which fall 33% by its fourth unit, which works against it.

Prices by region: By default each technology’s installed cost is matched to the place you pick. Batteries: BloombergNEF’s 2025 four-hour system averages (about $219 per kWh in the US, $177 in Europe, $73 in China) plus about $50 for installation and grid connection, Ember’s split of its all-in figure, giving $270, $225 and $125; elsewhere, Ember’s all-in $125 for markets outside the US and China. Solar and onshore wind: IRENA’s Renewable Power Generation Costs in 2025, the weighted average for projects commissioned that year in each of its regions, brought from 2025 to 2024 dollars like every other cost here. Each place takes its IRENA region’s 2025 average, or its own country’s figure where IRENA reports one; the full lists are under Sources and reference figures, at the end. Large reactors cost $4,700 per kW in the US, $4,540 in South Korea, $4,000 in China and $3,550 in India, applied to large reactors only. Elsewhere the US figure is scaled to local building costs by the ratio NEA and EPRI’s Costs of Generating Electricity 2025 reports between that country’s large reactor and the US one, the mean of its near-term and mature ratios where it gives both: Australia 1.48, Belgium 1.36, Bulgaria 1.14, Canada 0.89, the Czech Republic 1.04, Finland 1.10, France 1.12, Japan 0.93, Poland 1.35, Romania 1.14, Slovakia 1.42, Slovenia 1.40, Sweden 1.33 and Switzerland 1.02. Other European countries take the middle of the European ratios, 1.14, and places elsewhere keep the US figure. That puts reactors on the same footing as wind, solar and batteries, which carry each place’s own price. South Korea’s own submission to that report, a mature $2,487 per kW, is a projection, so the tool keeps Barakah’s realized cost for the Korean design. The Indian figure is Kudankulam 3 and 4, two 1,000 MW Russian-designed light-water reactors, at their revised cost of ₹688.9 billion (India’s parliament, the Lok Sabha, July 2026), converted at 2024’s average exchange rate, with interest during construction taken out because the tool adds its own. India’s own 700 MW heavy-water reactors cost about half as much, and they are the reactor the tool builds in India (see What it costs); the $3,550 applies to a light-water reactor picked by hand or to a selection that mixes India with other places, and a CANDU picked there takes it too, so it is priced high. A selection of several places takes each one’s price weighted by the size of its grid. US states take US prices, and Canadian provinces take Canada’s, which follow the US for batteries and, scaled by Canada’s ratio, for reactors. To use your own price for one source, pick another price in its menu in Advanced; picking the “Set for” entry hands it back to the region, and one global price sets all of them at once. The 1.1 GW reactor is priced for the region whether the tool picks it or you do, and the CANDUs, which have no price of their own, take the large reactor’s; the Korean, Chinese and Indian designs keep their own country’s cost, and SMRs, microreactors and custom reactors have no regional figures. With one global price, batteries default to $150 per kWh installed, close to the roughly $140 IRENA reports as 2025’s global installed cost for a four-hour system. All of these are today’s prices; on the default cost basis, wind, solar and batteries then fall from them along NREL’s projection.

Capital cost: Overnight costs per kW at today’s prices, matched to each place by default; the cost basis above sets how they change over the build. Without regional matching, solar is $650/kW and wind $951/kW, IRENA’s 2025 global averages, and batteries $150/kWh, all used as set. The reactor menu in Advanced lists realized and projected costs for each design: Rooftop solar costs 1.93 times the utility-scale price per kW (NREL residential and commercial blend), about $1,250/kW at the global price and offshore wind to about $4,000/kW with a 45% capacity factor and no land footprint. Geothermal ($7,490/kW) and reservoir hydro ($5,488/kW) are the medians across NEA member countries in NEA and EPRI’s Costs of Generating Electricity 2025, replacing EIA’s cheapest-site figures ($3,200 and $3,300/kW), since a region-wide build does not stay on its best site; Advanced restores them. Gas plants are priced after the turbine shortage that roughly doubled what a new combined cycle costs between 2024 and 2025: $2,340/kW, IRENA’s figure for a new US plant in 2025, against the $1,100 the EIA estimated before the shortage. BloombergNEF ($2,157 in early 2026), NextEra ($2,400) and Lazard ($2,400 to $2,600 for plants after 2028) agree. Gas with carbon capture is $3,490/kW: the new combined cycle plus about $1,150 for the capture plant, which NETL finds adds 100 to 104% to a combined cycle’s cost at normal prices. The capture plant has no turbines in it, so its share did not ride the shortage. No such plant runs commercially at scale, so treat it as an engineering estimate. Both prices apply everywhere, which overstates them where China builds. Gas with capture also pays to store what it catches, 0.34 metric tons per MWh at NETL’s $12.50 a metric ton, about $4.25 per MWh. Investment figures include interest during construction by default and leave out fuel and upkeep, which appear in the money breakdown and the cost per metric ton. Storage rock: for US states the tool checks the Department of Energy’s storage atlas (2012 edition, its low estimate of deep saline formations, oil and gas fields and unmineable coal, as tabulated by state in the EPA’s 2015 analysis of where carbon storage is available). That is a screening estimate of what the rock could hold, not of sites studied, permitted and ready to inject. Where a state’s rock holds less than its gas plants with capture would inject over their lives, or the atlas did not assess it, as with Minnesota and New England, which the EPA also counted as having none, the plants are charged as remote plants with 200 km of power line, as if built where the rock is and wired in, which is how the EPA expected such states to be served. NETL’s storage price already pays for a 100 km pipeline. Countries and provinces are not checked.

Interest during construction: Quoted plant costs are overnight costs, as if each plant were built in a day. Real money is spent over the build and earns interest until the plant runs, which matters far more for a reactor that takes seven years than for a solar farm that takes two. The tool adds that interest by default, at 5% a year in real terms, assuming spending is spread evenly over the build: total cost rises by ((1+r)L − 1)/(rL) for a rate r over L years. Reactors use their own build time; other plants use the EIA’s lead-time assumptions for new plants (wind 3 years, solar 2, hydro, geothermal and biomass 4, gas plants 3, batteries 1). At 5% that adds about 16% to a large reactor, 5% to wind and 2.5% to solar. Advanced offers 3%, 7%, or leaving it out.

Cost per metric ton, and fuel: The cost per metric ton of CO₂ avoided is levelized: each part of the build’s capital becomes a yearly payment over its own life at the same real rate used for interest during construction (5% by default), plus yearly upkeep and decommissioning, divided by the metric tons avoided each year. It is the cost of the new system alone: the fuel and upkeep the retired fossil plants no longer need are not subtracted, so it is not a net cost of cutting carbon. Every mix in a place retires the same fossil plants and saves the same dollars a year, and subtracting them would lower every mix’s figure, most where that fuel is imported. Upkeep is the EIA’s operations and maintenance for new plants (Annual Energy Outlook 2026 assumptions, Table 3, brought from 2025 to 2024 dollars), with batteries as four-hour systems; as in NREL’s Annual Technology Baseline, battery upkeep covers the augmentation that keeps a battery at its rated capacity, so battery fade adds nothing on top. NEA and EPRI’s figures for the same upkeep run higher for most sources, most for hydro. Decommissioning follows NEA and EPRI: 15% of a reactor’s overnight cost, spread over the 10 years after it closes, and 5% over two years for every other plant, for batteries and for carbon capture plants, discounted to the build. That is about 1% of the capital for a 30-year plant and 0.2% for an 80-year reactor. Fuel is included: nuclear at $9.06 per MWh for a light-water reactor and $5.69 for the CANDUs and India’s heavy-water reactor (below), NEA and EPRI’s full fuel-cycle figures, which include storing and disposing of spent fuel (the US fleet’s own $5.32 in 2023 carries no disposal cost, since the federal disposal fee was suspended in 2014); gas at the regional price times EIA heat rates for new plants; and biomass at $73 per MWh, from the $70 per dry short ton at which the DOE’s Billion-Ton Report counts supplies. Regional gas prices come from the World Bank’s October 2026 forecasts, brought to 2024 dollars. A gas plant burns fuel for 25 to 30 years, so on the mature and NREL bases it pays the forecast for 2028, by which the World Bank expects the Strait of Hormuz closure and the strikes on Qatar’s LNG plants to have mostly worked through: $3.63 per MMBtu for North American pipeline gas (the US, Canada and Mexico), $11.64 in Europe, $12.10 where the next gas plant would burn imported LNG (Japan, South Korea, Taiwan, most of South and Southeast Asia, Kuwait, Bahrain, Jordan, Türkiye, Chile and the LNG importers of Central America and the Caribbean), and $7.87 elsewhere, halfway between pipeline gas and LNG. China and India take that middle price, since most of the gas they burn is their own or piped in. On today’s prices, gas pays the 2026 to 2028 average, crisis included: $3.44, $15.21, $15.83 and $9.64. The middle price is high for countries that burn their own gas.

Full system cost, and the stress test: The UN Economic Commission for Europe’s Protocol for a Full System Value and Cost Approach (2026) asks that electricity be judged on its whole system cost rather than each plant’s: the plants, the cost of matching variable output to demand, balancing, grid stability services, the grid itself, flexibility and the costs a system puts on others. This tool counts the plants, the storage and overbuild that carry wind and solar through their swings, the power lines, deaths from air pollution and accidents, land, materials and carbon. It does not cost balancing reserves, inertia and the other stability services, distribution networks or grid connections, and it does not model hour by hour, so the protocol’s multi-week weather droughts and its value of lost load are out of reach. From the protocol’s step 4 it runs two stress tests in the comparison, on every reference mix: interest rates 2 points higher, applied to both the interest during construction and the yearly payments (with interest during construction switched off, the test switches it on, so that column also counts interest the base leaves out), and every fuel, uranium included, 50% more expensive. Cross-checked against NEA and EPRI’s Costs of Generating Electricity 2025, which uses the same 5% real rate, the tool’s mature large reactor ($4,700/kW) matches their US figure ($4,760), hydro and geothermal are their medians, and the SMR’s $7,000 matches their mature median ($7,092). Their US solar ($1,420/kW) and four-hour batteries (about $390 per kWh) cost more than the IRENA and BloombergNEF prices used here ($1,150 and $270), which are newer market figures, and their gas plant ($1,580) less than the $2,340 used here, which prices the turbine shortage. Their hydrogen turbines assume hydrogen at about $1.30 a kilogram, so the tool does not use them.

CANDU: The 925 MW CANDU is AtkinsRéalis’s Monark, licensed at 925 MW net as a close copy of the Darlington units, with room to uprate to about 1,000 MW. None has been built and no one has estimated what a mature Monark fleet costs, so it takes the large reactor’s price for the place and cost basis: in the US, $4,700/kW on mature prices and $8,300 on today’s. The one like-for-like comparison supports treating the two alike. Canada offered Poland a CANDU plant at C$45 to 50 billion, about PLN117 to 130 billion, against the PLN178 billion budgeted before financing for the three AP1000s Poland is building: 15 to 25% less per kW if the CANDU plant is three units too. That is an offer set against an approved budget, so the tool reads it as support for equal prices rather than a discount. The CANDU menu in Advanced lists realized and quoted costs, from Qinshan III’s $3,400/kW to about $12,000/kW all-in for the Poland offer. Two costs differ from a light-water reactor. A CANDU’s pressure tubes and feeders are replaced about halfway through its life; Darlington’s four units cost C$12.8 billion, about $2,700/kW, finished in 2026. A light-water reactor’s long-term operation program costs about $900/kW, which the tool does not charge new reactors, so the CANDU pays the $1,800/kW difference for every 40 years it runs, once on the default 80-year life, discounted and spread over the life like the capital: about $13/kW a year at 5%. Its fuel is natural uranium, which needs no enrichment. NEA and EPRI put a heavy-water reactor’s full fuel cycle at $5.69 per MWh against $9.06 for a light-water reactor, 1.6 times less, the low end of the 1.6 to 1.8 Hu and Tang (Canadian Nuclear Society) find, which saves about twice what the retube costs. The 700 MW CANDU 6 is the design Romania is completing at Cernavoda 3 and 4, built eleven times in five countries and offered today as the Enhanced CANDU 6. It follows the same price rule and burns the same fuel; its retube follows CANDU 6 refurbishments, about $3,000/kW (Embalse about $3,280, Cernavoda 1 about $2,850), so it pays $2,100/kW beyond a light-water reactor’s life extension. It is also the reactor the tool builds in Argentina, which runs one at Embalse. India’s heavy-water reactor is NPCIL’s own 700 MW design, descended from the CANDUs India built in the 1970s and now built in series, and it keeps India’s cost wherever it is picked, like the Korean and Chinese designs. RAPP 7 and 8, finished in April 2026, cost ₹229.2 billion for 1,400 MW: about $1,690/kW at 2024’s exchange rate with interest during construction taken out, the same conversion as Kudankulam. Kakrapar 3 and 4 came to about $1,660, and Mahi Banswara’s four units, started in 2026, are budgeted at about $1,550 (₹420 billion, NTPC; later reports say ₹500 billion, about $1,840). It burns the CANDU’s natural-uranium fuel, and its coolant channel replacement, about $200/kW at Kakrapar 1 and 2’s cost, is charged in full every 40 years.

SMR & microreactor economics: Switching reactor size also switches the capital-cost menu, because small reactors sacrifice the economies of scale that make large plants cheaper per kW. The evidence so far is sobering: NuScale's canceled Idaho project reached roughly $21,600/kW, higher per kW than Vogtle. Vendor NOAK targets are lower (IEA 2025 puts EU SMRs near $10,000/kW; independent bottom-up work by Asuega et al. 2023 spans $4,300–6,400/kW), and the tool offers that full range with a mid-estimate default. Microreactors are costlier and more speculative still, with credible figures from about $6,700/kW (Oklo's NRC filing) up to $25,000/kW. The theory that factory production drives costs down is sound but unproven; only three SMRs were operating worldwide as of 2024, too few to show the learning effects proponents expect. Treat every small-reactor cost here as a scenario input, not a market price.

Jobs: Operations jobs per MW are literature-anchored estimates in the tradition of Rutovitz, Dominish & Downes (2015, for Greenpeace/ISF), the most widely-cited global source for energy employment factors. The nuclear figure (0.7 jobs/MW, generation plus fuel cycle) reproduces the US Energy & Employment Report 2025's actual combined total for nuclear generation and fuel (67,900 jobs) applied to the current US fleet almost exactly, the closest independent check available. Solar and wind use lower permanent-operations factors than some advocacy sources cite, because utility-scale plants need very little ongoing staff once built: a large solar farm runs with a few technicians on site. Per unit of energy the three land closer together than per MW, and the order depends on the place, because a weaker wind or sun needs more MW for the same energy: about 90 permanent jobs per TWh for nuclear, 55 to 100 for wind and 55 to 150 for solar. Much of solar and wind employment is one-time construction and factory work, which is counted separately as job-years, not permanent jobs. The other technologies' factors follow the same literature but haven't each been re-verified against USEER line items; treat them as reasonable estimates. Job-years to build each MW count the work on site, from planning through commissioning, plus the factory work behind it, wherever the factory is, which need not be the country building the plants. A job-year is a full-time year of work, 2,000 hours. For nuclear, wind and solar the site work is measured on real builds. A reactor’s site work depends on how mature the build is, so the tool reads it between two measured builds by the reactor’s price. Korea’s operator, KHNP, put Shin Kori 5 and 6, built in series, at 8 million person-days for both units (as reported in 2019), about 11 per MW, and that is the figure for a large reactor at the mature $4,700. Vogtle 3 and 4 in Georgia, the first of their kind in the US, took at least 165 million hours, about 39 per MW, by the safety records reported to the Georgia Public Service Commission, at about $17,500 per kW. A reactor priced between the two takes a straight line between them, and one priced outside takes the nearer figure. Labor isn’t a fixed share of a reactor’s cost: most of what made Western builds expensive was labor and supervision (Eash-Gates and others, Joule, 2020; the NEA), so hours are read off builds rather than scaled with price. Barakah’s four reactors in the UAE had taken more than 325 million hours for 5,600 MW by 2020 with the project 94% done (ENEC), about 31 per MW once finished, nearly three times Shin Kori’s at a similar price; low-wage crews there did work that machines and prefabrication do where labor costs more. At US wages that much site work would cost about as much as the whole mature reactor. Large reactors take the US price on your basis wherever they are built, since a cheaper local price mostly buys cheaper hours rather than fewer; SMRs, microreactors and a price you set take that price. Onshore wind’s site work is about 2.5: the US Energy & Employment Report counted 42,000 wind construction workers in 2020 (as tabulated by NREL), a year the US built 16.8 GW of new wind and 17.5 GW counting repowered turbines, 2.4 to 2.5 per MW, and IRENA’s study of a 50 MW wind farm puts installing and connecting it at 2.8. Utility-scale solar’s is about 1.25, from IREC’s count of 29,708 utility-scale installation and development jobs in 2023 against about 31 GW of panels built, 1.0 per MW of panels and 1.25 per MW of the inverter capacity the tool sizes solar by. IREC counts everyone at firms whose main business is installing and developing projects, so office staff are in it along with crews. Factory work is 1.3 for nuclear, from Rutovitz, and 1.8 for wind, from IRENA’s wind study, with planning and transport included. Solar’s is 1.7, from IEA’s study of solar supply chains: 1,300 jobs for each gigawatt a year of capacity from polysilicon to finished modules, at 1.3 MW of panels per MW; transport adds 0.36, from its study of a 50 MW solar plant. Like wind’s, it counts the main components only. That makes about 13 job-years per MW for a large reactor at $4,700, 20 at the $8,300 MIT expects the next US reactor to cost, 4.3 for wind and 3.3 for solar, of which 1.3, 1.8 and 1.7 are factory work. Wind and solar keep their figures on every cost basis: their prices fall along NREL’s path mostly because turbines and panels get cheaper, not because less work is done on site. Vogtle’s and Barakah’s figures are hours turned into full-time years; wind’s and solar’s are counts of people, taken as full-time because their hours aren’t recorded. Big builds often run long weeks, which spreads the same hours over fewer people. Wind and solar crews also work long days in season, and some only part of the year, so their counts could miss in either direction. The construction counts in the US Energy & Employment Report can’t be divided by the year’s build: wind’s was 42,000 in 2020, when the US built 17 GW, and 44,000 in 2024, when it built 5, most of solar’s are rooftop installers, and nuclear’s showed about 2,100 workers in 2023 while Vogtle was still being built. Hydro, geothermal, biomass and both gas plants use Rutovitz. Ranked on jobs, each mix shows its permanent jobs beside the workers building it, on site and in factories, the job-years of the build spread over the years it takes, with the average hourly pay across its sources, each weighted by its jobs and at local buying power. The ranking carries no green or red: more jobs for the same power is more work and more cost at once. Solar’s permanent jobs, 0.10 per MW, are checked against national counts as nuclear’s are: about 0.11 per MW at US solar operations and maintenance firms (IREC’s Solar Jobs Census, 2024) and 0.09 at solar generation companies (BLS QCEW, 2024), against 0.05 to 0.07 counting only the staff on site at large plants (NREL, 2013).

Pay & public revenue: Wages anchored to BLS OEWS May 2024 medians (nuclear reactor operators $123,380, nuclear technicians $104,240, power plant operators $103,600, wind techs $62,580, electricians $62,350, the pay of the technicians who keep a solar farm running); each technology gets a plant-wide average. Against what US power plants report paying (BLS QCEW, 2024), the payroll per MW matches for nuclear ($73,000 per MW in the records, $74,000 here) and wind ($10,600 and $10,500). Solar’s records run higher: $19,000 per MW of utility-scale capacity, or $13,000 counting rooftop panels, against $6,200 here, since most solar staff in the records work in offices. Pay then follows the place, in US dollars, and each place is compared with the US through the same kind of source. US states and Puerto Rico use the average pay in electric power from those same records, against the US average of $137,821. Germany, the United Kingdom, China, Brazil, South Africa and Canada use their statistics offices’ employer records for the electricity sector (Destatis, ONS, NBS, IBGE, Stats SA, Statistics Canada) against the same $137,821, at 2024 average exchange rates. Statistics Canada publishes utility pay for Alberta, British Columbia, Ontario and Quebec; the other provinces and territories take Canada’s figure. Thirty other European countries use Eurostat’s 2022 Structure of Earnings Survey, also employer records, as a ratio to Germany’s pay in the same survey. Fifty-nine countries use the International Labour Organization’s average pay in the electricity sector, from 2018 or later, against the ILO’s US figure from the same kind of household survey, with US pay moved to the same year at 4% a year. That US figure runs 26% below the employer records, so these countries sit on a different footing from the employer-record ones. Ten countries with no ILO electricity figure, Japan and Australia among them, use the ILO’s average pay across the whole economy, which assumes electricity work carries the same premium as in the US. The remaining 28 scale by income per person (World Bank GDP per capita, 2023) to the power 0.92, the fit through the ILO countries. Its typical miss is a factor of 1.7, and the payroll line says when a place uses it. “World” weights each country by the size of its grid, and several places by their share of the build. Jobs per MW are the same US-validated factors everywhere. Plants in lower-wage countries often employ more people per MW, so there the job counts run low and payroll with them. Household surveys also run below employer records in poorer countries (Brazil’s employer records sit 43% above its ILO survey figure), so payroll there runs low again. A place’s figure covers every source, so each source keeps the gap to the others that it has in the US. Where there is evidence for one source on its own, that source gets its own figure, by the same rule for every source. In China, India and South Korea the companies that run each kind of plant publish what they pay, and each is set against what US plants of the same kind pay (QCEW, 2024: nuclear $167,135, wind $136,300, solar $127,486; hydro, coal and gas, geothermal and biomass from the 2013 records, raised by the 43% that nuclear and wind pay each grew since), with employer pension and social contributions taken out where the published figure includes them. China: CNNP for nuclear; Huaneng Hydropower, Gansu Energy and Mindong Electric for hydro; Huaneng Power International, Huadian Power International, Guangdong Electric Power and Shenzhen Energy for coal and gas, taking the middle one where several report. India: NPCIL, NTPC and NHPC, pooled into one figure for nuclear, hydro, coal and gas (the reason closes this section), and ReNew for wind and solar. South Korea: KHNP for nuclear and hydro, and the five KEPCO generating companies for coal and gas, at their 2019 ratio to KHNP’s pay. A source with no operator figure takes the middle of the country’s others. In India, NTPC pays its staff the most, then NHPC, then NPCIL, and ReNew pays under a third of what NPCIL does. In China, CNNP pays more than the middle hydro company, which pays more than the middle coal and gas company. In South Korea, the generating companies pay about 4% less than KHNP. These operators pay 1.6 times China’s sector figure, 3 times India’s and 0.83 times South Korea’s, so no single correction fits the countries whose figures come from surveys, and those stand as they are. ČEZ, Nuclearelectrica and Energoatom also publish their pay, and it lands within 5% of their countries’ figures, so the Czech Republic, Romania and Ukraine keep one figure for every source. US states and Puerto Rico use what their own plants of each kind pay, where the records cover 250 or more jobs: nuclear in Alabama, Nebraska, Pennsylvania, South Carolina and Tennessee, solar in 15 places and wind in 9. The hourly pay beside each source is at local buying power, so it compares directly with a US wage: local pay in dollars times how much more a dollar buys there, from the ILO’s own conversions to purchasing-power dollars, or a fit to income per person where it has none. In India a dollar buys 4.4 times what it buys in the US. The payroll stays at exchange rates, since it sits beside costs in dollars. The tax estimate covers US places only, since it uses US tax rates, and is deliberately rough: 25% of payroll (combined income and payroll taxes) plus 0.5% of capital cost per year as a property-tax proxy, consistent with reported single-plant figures like nuclear stations paying tens of millions a year to local counties. Where a country’s pay comes from ILO’s wage survey, which covers the whole electricity sector or the whole economy, it is raised to what power plants pay: in seven countries with both figures, operators earn 0.8 times the survey in South Korea, 1.6 in China, 2.4 in Thailand (EGAT), 4.1 in India (its three state generators), 5.3 in Nigeria (Geregu Power), 5.5 in Kenya (KenGen) and 5.7 in Ghana (the Volta River Authority), a gap that widens as income falls. Fitted on the other six, the curve predicted Thailand within 9% before EGAT’s accounts were added. The multiple is fitted to income per person and held at Nigeria’s level below it; Nigeria, Kenya, Ghana and Thailand use their operators’ own pay. India’s three state generators, NTPC, NPCIL and NHPC, are pooled into one figure for every large plant, since their pay per head differs mostly in how much of the plant work goes to contract staff; private wind and solar developers, which pay far less, keep their own. A company’s pay per head covers only its own staff, and in India contractors’ employees do much of the plant work, so each figure blends them in. NPCIL’s operating stations had 10,349 contractors’ employees in September 2024 (India’s parliament, December 2024) beside NPCIL’s 10,576 staff, about half the workforce, paid at least the central minimum wage for skilled work, about ₹2.98 lakh a year, a seventh of the state companies’ pay per head; that brings the state companies’ figure down by 43%. NTPC lists far more contract workers, but its count includes its construction sites. ReNew lists 201 of 4,586, so its figure falls 2%.

Fossil jobs displaced: ~127 jobs per TWh of fossil electricity displaced (USEER 2025 generation jobs plus the power-sector share of gas, coal and oil fuel-supply jobs, divided by US fossil generation) and ~38 jobs per thermal TWh of non-electric fossil fuel displaced by synfuels. US-derived factors applied globally as a rough proxy. Synfuel-plant jobs created are not counted, so the net-jobs figure is conservative for synfuel scenarios.

Keeping the lights on: firming, storage and transmission

Why firming is charged at all: The premise is not ours. Sepulveda, Jenkins, de Sisternes & Lester (2018, Joule) evaluated nearly 1,000 decarbonization cases and found that firm low-carbon resources, nuclear among them, cut the cost of a fully decarbonized grid by 10 to 62%, and that every least-cost path to clean electricity included at least one; batteries and demand flexibility did not remove the value of firm capacity. MIT’s Future of Nuclear Energy in a Carbon-Constrained World (Buongiorno et al., 2018) reaches the same place from the cost side. Sepulveda et al. (2021, Nature Energy) then asked what storage would have to cost to replace firm generation: $20 per kWh of energy capacity to cut system costs by 10%, and $1 per kWh to displace firm low-carbon generation entirely, at durations beyond 100 hours. Utility batteries cost $125 to $270 per kWh installed today. That gap is why wind and solar are charged for storage and overbuild here rather than having the problem assumed away.

Firm grids: load-following or storage: Every grid has to meet the daily swing in demand, and no source gets a free pass on it: the swing is charged whenever nuclear, wind, solar and geothermal together pass 70% of the grid. A firm grid can load-follow, throttling reactors at night so the fleet’s capacity factor slides from 90% toward the grid’s load factor (about 62% at 100% nuclear) and more nameplate is built to cover peaks, as France does. Or it can run flat out and store the surplus, which is why several pumped-hydro stations were built beside nuclear plants in the 1970s and 80s. Both paths are costed with the figures the cost per metric ton uses: capital with interest during construction, spread over each asset’s life at the same real rate (5% by default), plus the EIA’s fixed upkeep, about $137 per kW a year for a large reactor and $46 per kW a year for a four-hour battery that lasts 20 years. The load-following penalty grows steeply as plants throttle further while storage costs rise in a straight line, so the default searches for the cheapest mix, checking each point where a small fleet could build one reactor fewer. On the defaults an all-nuclear United States stores a little over half of the daily swing and a France-style one nearly two-fifths; costlier reactors push toward storage and costlier batteries toward load-following, and Advanced can force either. A firm grid needs only a daily shift of a few hours, an order of magnitude less storage than a wind-and-solar grid riding out multi-day weather. Electrolysis for synthetic fuel eases the swing in proportion to how much of it it can absorb, since it can run only when there is power to spare; electric cars, heat pumps and industrial heat count as ordinary demand that has to be firmed.

Wind and solar: storage and overbuild: When “Firm it” is on, storage and wind and solar overbuild rise with the variable share of delivered power, along anchor points from Tong et al. (2021, Nature Communications), which extends Shaner et al. (2018) from the US to 42 countries. Wind and solar sized to annual demand with no storage meet 72 to 91% of hours (mean 83%); with 12 hours of storage 83 to 94% (mean 90%); at 1.5 times generation and 12 hours, 89 to 100% (mean 98%). The two levers trade: 10% more generation is worth about 3.9 hours of storage, and Shaner found 3.4 times generation with no storage about as reliable as 1 times with 12 hours. Above 80% wind and solar the curve follows the least-cost mix Fan Tong et al. (2020, iScience) found for meeting every hour of US demand: at battery or long-duration prices, building more wind and solar beats storage for the rare worst hours. At US default prices that comes to about 2.0 times overbuild and 17 hours of storage at 100% wind and solar; cheaper storage shifts the mix toward storage. Storage never falls below the 12 hours charged at 80% wind and solar, and overbuild never below the 1.5 times charged at 85%, so neither drops as wind and solar rise; when storage is dear, as with lithium-ion for every hour, that costs a little more than the least-cost mix. Read the result as a firming estimate. Meeting a year’s energy, meeting every hour of a modeled weather period, and meeting a reliability standard such as a loss-of-load expectation are three different achievements, and these curves come from the second kind of study. That least-cost study assumed perfect transmission across the whole US and one weather year at a time, so the real need is higher on smaller grids and over many years: Ruhnau & Qvist (2022) optimized a 100% renewable Germany across 35 weather years and found about 24 days of average load in storage, with one weather year understating it by half. The curve is deliberately not stretched to match, because their answer is mostly hydrogen at chemical-storage prices that this tool does not model, so treat its storage as a floor. Cost-optimal studies that accept some unserved energy and lean on transmission find shorter durations, 6 to 10 hours in solar-led systems and 10 to 20 in wind-led ones; they answer a different question.

Storage prices: Batteries are matched to the region by default: $270 per kWh installed in the US and Canada, $225 in Europe and $125 elsewhere, or $150 with one global price (NREL’s 2025 four-hour benchmark is about $334; leading projects about $125 to $150). Lithium-ion is built for a few hours, and pricing days of storage at battery rates would be a strawman, so only the first 12 hours of duration are priced as batteries. Everything beyond is long-duration storage, by default $100 per kWh, the middle of the $50 to $150 range for compressed air in salt caverns, which has run commercially for decades (US DOE, 2023). Advanced offers $50, $20 (Form Energy’s target for iron-air, not yet commercial), or lithium-ion for everything. Long-duration storage is not counted in battery units or battery materials. NREL found hydrogen worthwhile for seasonal storage once power equipment falls below about $1,500 per kW and underground storage below about $1.80 per kWh (Guerra et al., 2020), so this tier could get cheaper still.

Transmission: Remote, diffuse generation needs new high-voltage corridors; plants at existing grid nodes, often retiring coal sites with the interconnection already built, need only a short spur. The model counts 500 kV circuits at one per 3.5 GW of corridor capacity (Princeton Net-Zero America’s convention), at roughly $2.67 million per km (MISO’s 2024 Transmission Cost Estimation Guide), over 200 km for remote wind, solar and hydro and 25 km for plants at existing nodes. Corridor length is the honest uncertainty, so Advanced can halve it, double it or switch transmission off. Each corridor’s 61 m right-of-way is shared land, like the space between wind turbines: crops and grazing continue under the towers, so it is stated beside the land total and joins it only when the Land section counts shared land; through forest a corridor is cleared and kept low, which the tool does not separate out. The result matches the literature in direction and rough scale: a wind-and-solar grid needs tens of times the new line a nuclear one does, and transmission comes to 5 to 6% of capital for variable-heavy mixes against well under 1% for firm ones; Net-Zero America’s renewables-led pathway roughly triples US high-voltage capacity by 2050, and its nuclear-heavy pathway needs much less. Treat the kilometers as indicative and the ratio between mixes as the point. Not modeled: distribution upgrades, interconnection queues and the politics of siting a line.

Capacity factors: what each plant delivers, by place

Capacity factor by place: A solar panel in Las Vegas delivers nearly twice what the same panel does in Newark, so every place uses capacity factors measured for it. US states come from the EIA’s State Energy Data System (Table F53, 2024), which already corrects for plants that ran part of the year: utility solar runs 29.8% in Nevada and 15.5% in New Jersey against a national 23.2%, onshore wind 43.1% in New Mexico and 18.9% in Massachusetts against 34.3%. Countries are computed the same way from different inputs: 2025 wind and solar generation from the Energy Institute, divided by the mean of end-2024 and end-2025 capacity from IRENA. That method puts the United States at 23.1% solar and 34.5% wind, within a fifth of a point of the EIA, and a cross-check against Ember’s own figures for 60 countries finds a median difference of zero. Solar runs 29.4% in the UAE and 10.5% in Germany; wind runs 60.2% in Kenya (Lake Turkana, the best wind site in commercial operation) and 16.3% in Russia. The world figure, 15.0% solar and 25.6% wind, reads low mostly because China holds about half of both fleets and runs 12.8% and 22.2%.

Where a place has no fleet to measure: Fifty-four countries with almost no solar take it from the Global Solar Atlas (World Bank and Solargis): the modeled yearly output of a fixed-mount kilowatt at the country’s geographic center, divided by the hours in a year. Nigeria reads 17.3%, Egypt 22.7%, Norway 11.3%. Tested against the 24 countries that have both, the raw conversion needs no correction factor and cuts the error against the flat world figure from 5.1 points to 3.0; rooftop-heavy fleets land below their resource and tracking fleets above it. Wind gets no such stand-in, because latitude predicts it no better than chance, so unmeasured countries keep the world fleet figure until a wind atlas is added. A place’s own figure counts only if its fleet is a sample rather than an anecdote, at least 0.15 TWh a year of solar or 0.3 TWh of wind, and only if it is physically possible (solar 7 to 31%, wind 10 to 62%), which rejects reporting mismatches like DR Congo’s computed 992% solar. A US state or Canadian province without a reading takes its country’s; solar north of 60°N takes the average of countries at that latitude; anything else falls to the world figure, and the screen says which is in use. Several places at once are weighted by how much of the build lands in each, not by the size of their grids today, and the measured places must carry at least 60% of the build or the whole selection uses the world figure.

New projects against new projects: A fleet average includes twenty-year-old turbines, early fixed-tilt solar and real curtailment. This tool models new construction, and it sizes nuclear on a new plant’s 90%, so sizing wind and solar on their fleets compared new reactors with old turbines. They are sized on new projects too. IRENA puts projects newly commissioned in 2025 at 16.5% solar and 36% onshore wind worldwide, against the world fleet’s 15.0% and 25.6%. Each region keeps its measured fleet as the starting point, because that carries how windy and sunny it is and how hard its grid curtails, and is lifted by the world’s proportion: ×1.10 for solar, ×1.41 for wind. The lift stops at the world’s newest-project average unless a region’s own fleet already runs higher, so no region is scaled to a good-site figure that a region-wide build would never average. Texas, measured at 34% wind, is lifted only to 36% rather than 48%. The alternative, sizing nuclear on its measured fleet too (83% worldwide in 2024, World Nuclear Association), was considered and rejected: it is symmetric on paper, but this tool models building plants, and new plants of every kind outperform old fleets. Advanced switches wind and solar back to the measured fleet, pins any figure over it, or uses one world figure everywhere. New solar is sized on its output averaged over its life: panels lose about 0.5% of their output a year (NEA and EPRI’s default; countries report 0.3 to 0.5%), 7% on average over 30 years, so a new farm is sized at 93% of its first-year figure, never below the measured fleet, which already carries its own panels’ aging. NEA and EPRI find aging small enough to leave out for every other plant except batteries, whose upkeep covers it. A figure you pin is used as set.

Land moves with it: Every land figure is km² per TWh delivered, measured by Lovering et al. on US sites, so each carries the capacity factor of the US plants behind it. A megawatt of utility solar occupies roughly the same ground wherever it stands, and weaker sun needs more megawatts per TWh, so each land figure is scaled by the local capacity factor against the US one. Picking the United States reproduces the published figures exactly; Germany’s solar land comes out about 2.2 times the US figure per TWh and its wind about 1.7 times. That holds land per megawatt constant when Germany’s fixed-tilt arrays actually pack tighter than US trackers, so its true figure sits somewhat below the scaled one. Advanced can switch the scaling off.

What is left out: Nuclear, hydro, geothermal and biomass capacity factors are not taken from fleets, because for a plant that can be dispatched the figure records how its owner chose to run it rather than what the place allows. Offshore wind keeps its own 45% everywhere. State figures are utility-scale while country figures include rooftops, so the rooftop share set in the Land section is reflected in neither.

Land

Land: Land use of real operating sites, from Lovering, Swain, Blomqvist & Hernandez (2022, PLOS ONE), in km² per TWh a year: nuclear 0.15 for the fenced site (the paper’s mean rather than its lower median, as the multi-unit note explains), or the site plus its emergency planning zone if switched on; onshore wind 0.4 for pads and roads, or 99 counting the spacing between turbines; utility solar 19, blended with rooftop solar at 0.5 by the rooftop-share slider (40% by default, roughly today’s world split); hydro 14; geothermal 1.4; gas with capture 1.3. Biomass depends on what it burns. Most biomass power runs on leftovers (REN21: in 2020 about 70% solid biomass such as pellets, chips and bagasse, 16% waste, 13% biogas). New biomass draws first on unused crop residues, from the DOE’s 2023 Billion-Ton Report ratio of 0.33 metric tons sustainably available per metric ton of grain, applied to each country’s cereal harvest (FAO) at about 1.04 MWh per dry metric ton with half kept for liquid fuels, and in a full rebuild on today’s biomass output as well. Those leftovers count at 0.15 km²/TWh/yr, and anything beyond them as dedicated energy crops at 580, the most land-hungry source there is. Advanced can count all biomass as crops or all as residue. The Chernobyl and Fukushima exclusion zones, spread over every TWh nuclear has generated, add 0.039 km²/TWh/yr.

Nuclear land & multi-unit sites: The 0.15 km²/TWh/yr nuclear figure is Lovering’s mean across 59 operating US stations. It covers the fenced plant site, not the wider emergency planning zone, which is a regulatory radius rather than occupied land. People live, farm and work inside EPZs today. That paper’s median is lower still at 0.071, and its authors call US plants an upper bound because most other nuclear countries put more reactors on a single site. This tool takes the mean because the mean is what the authors recommend for scale-up questions: building more of a source means building its land-hungry examples alongside its efficient ones. Because a reactor’s site scales sublinearly with capacity, multi-unit stations use far less land per TWh than single units, and siting four reactors together barely enlarges the footprint of one. Real fleets cluster their units, so the effective number is often below even this.

Emergency planning zones: The default counts nuclear’s fenced site, because an emergency planning zone is a planning boundary rather than occupied land: people live, farm and work inside it. Because critics cite it anyway, the Land section can count it, sized to the reactor selected. The familiar 10-mile ring comes from NUREG-0396 (1978), sized around the large reactors of that era. Modern assessments are much smaller, about 7 km for an AP1000 (Huang et al., 2011), and the NRC’s 2023 rule, 10 CFR 50.160 with Regulatory Guide 1.242, lets a design whose accident consequences stay onsite end its zone at the site boundary. The tool uses about 9 km² per TWh a year for a large modern reactor and effectively nothing for SMRs and microreactors, with the 1978 ring in Advanced for anyone who wants the number usually quoted. Even counting the modern zone in full, a nuclear grid occupies less land than an equivalent solar one.

SMR land: Selecting SMRs or microreactors also shrinks the fenced-site land per unit (×0.5 and ×0.3 of the large-reactor figure), reflecting the smaller emergency planning zones these designs pursue on the strength of passive safety, smaller radioactive source terms, and below-grade siting. The direction is what vendors and the NRC's now-final SMR emergency-planning rule point toward; the exact factors are illustrative, not certified, since few of these plants are built and licensed yet.

Land comparisons: Land is compared with places that appear on maps: US states and national parks for US places, Canadian provinces for Canada, and countries and islands everywhere else. City limits are not used: they never appear on maps and vary wildly with annexation rules (Jacksonville is 15 times the size of Boston). The nearest-sized place is chosen, never the place being studied itself. The simple view uses one shared reference across its three columns, sized so the multiples stay small, and adds a similar-sized place where one fits.

Why land differs so much between mixes: on your current region and land conventions, France style (70% nuclear) comes to …, against … for the IEA Net Zero 2050 default. Two things drive the gap. Wind’s figure depends on what is counted: its pads and roads are small, but its whole project area, 99 km²/TWh, would make it the biggest land user in the tool. Hydro at 14 km²/TWh is also geographically capped, so a mix leaning on it can hit a country's river ceiling before it hits a land ceiling. The default counts pads and roads, because most of the spacing between turbines stays farmed, and the spacing is stated separately beside it. The toggle in the Land section counts the whole project area instead, which moves wind-heavy mixes further than anything else in the tool.

Materials and fuel

Materials: Metric tons per GWh of lifetime generation, anchored to US DOE Quadrennial Technology Review (2015) as updated by Wang et al., Breakthrough Institute (2023). Fleet build mass = intensity × annual generation × the reference lifetime each published intensity was normalized to. Those references are not the operating lives used elsewhere in the tool; they are the denominators baked into the source figures, and they are fixed on purpose. For nuclear that denominator is 60 years even though the tool assumes an 80-year operating life, because changing how long a reactor runs alters how often it must be rebuilt, not how much steel and concrete went into building it once. Uranium fuel shown as annual demand versus world mine production.

Critical minerals: The materials section has two lists. The first is the whole build by tonnage, with every critical mineral collapsed into one row, so concrete and steel appear as the bulk they are. The second opens that row up in years of today’s world mine output, because critical-mineral tonnages span four orders of magnitude. Years of output measure scale, not a delivery schedule: a build needing three years of today’s silver would lean on new mines, recycling and stockpiles, so a large figure shows where supply would strain. Intensities come from the IEA (The Role of Critical Minerals in Clean Energy Transitions, 2021, maintained in its Global Critical Minerals Outlook), converted from kilograms per megawatt with each technology’s capacity factor and life, which reproduces the IEA’s totals closely (nuclear 5,000 to 6,000 kg/MW, onshore wind about 10,000, solar about 7,000). Battery minerals follow installed storage rather than generation, on an LFP-leaning blend of about 0.1 kg of lithium per kWh. Copper is counted once. Hydro, geothermal, biomass and gas with capture use the nearest documented analog and are estimates. World output figures are USGS-order annual production. Chromium, nickel, manganese and molybdenum are alloying metals already inside the steel tonnage, so the two lists are a drill-down, not a sum.

Breeders, thorium and seawater uranium: All three are off by default and are what-ifs, not deployed baselines. Today’s reactors burn only the ~0.7% of natural uranium that is U-235. Fast breeders turn the abundant U-238 into fuel, getting on the order of 60 times more energy per metric ton mined and able to burn down existing spent fuel and depleted uranium; the tool cuts uranium demand accordingly and raises nuclear capital cost about 20%. Breeders work (Russia’s BN-800 runs commercially) but have a thin, costly record (Superphénix, the US IFR). A thorium molten-salt reactor breeds uranium-233 from thorium inside the reactor, using about one metric ton of thorium per gigawatt-year against about 200 of mined uranium today; known thorium resources are about 6.4 million metric tons (World Nuclear Association), already mined as a byproduct of rare-earth sands. The tool counts that thorium and no uranium, leaves out each reactor’s one-time start-up charge, and applies the breeders’ 20% capital premium as a placeholder. No thorium power reactor runs commercially; China’s 2 MW experimental reactor reported converting thorium to uranium-233 in November 2025, with a 100 MW demonstration targeted for 2035. Uranium from seawater has been shown to work and would make supply effectively unlimited; its higher cost (about $400–1,000/kg against roughly $130 today) barely moves the price of power, because fuel is a small share of nuclear’s cost.

SMR materials: Large-reactor material intensities are the default. Selecting SMRs applies ×1.6 to structural materials and microreactors ×3, reflecting the surface-to-volume penalty of small cores (direction supported by Krall, Macfarlane & Ewing 2022, PNAS, and UC Berkeley's per-MW commodity comparisons), with genuinely large error bars. Some designs like the BWRX-300 target less concrete per MW than large plants. Fuel scales with generation, not with the multiplier.

Carbon, warming and carbon capture

Carbon: Lifecycle medians, g CO₂/kWh: nuclear 6, wind 14, solar 29, hydro 15, geothermal 19, gas w/ CCS 130, unabated gas 490, coal 820, oil 650. Nuclear’s figure comes from the UN Economic Commission for Europe’s Carbon Neutrality in the UNECE Region: Integrated Life-cycle Assessment of Electricity Sources, whose final March 2022 edition puts nuclear at 5.1 to 6.4 g CO₂e/kWh, the lowest of any technology it assessed. The rest are IPCC AR5 medians, cross-checked against the same report. UNECE issued a corrigendum in July 2022 correcting the land-use figures in that report, where biomass in the background electricity mix had inflated silicon PV; the greenhouse-gas results were untouched, and the land numbers here come from Lovering rather than UNECE in any case. Biomass grown as a dedicated crop is 230, the IPCC AR5 lifecycle median that reflects cultivation, fertilizer N₂O and land-use change; waste and residue is 34. The same split applies as for land: biomass up to a place’s current output counts as residue, anything beyond it as crops. Biomass carbon is genuinely contested, and counting it as low-carbon at all assumes the harvested carbon regrows, which can take decades. These per-kWh figures assume a plant’s normal output, so the new build’s carbon is counted on the fleet actually installed: an overbuilt wind and solar fleet that curtails part of its output still built every turbine and panel, and reactors that follow load built more nameplate than they run. Gas and biomass plants follow their output, since burning fuel is most of their footprint. Deaths per TWh for the new build are counted the same way.

How much carbon a build really avoids: The avoided-CO₂ figure compares the grid before and after, as if each clean kilowatt-hour removes a fossil one. Hausfather’s analysis of the real world finds that cycling losses, a demand rebound when power gets cheaper, and fossil-fuel market rebound trim that: his central case puts real avoided CO₂ at about 83% of a simple displacement count, with scenarios from about 45% to 94%. This tool reports the displacement count, so read its avoided-CO₂ and its cars-zeroed-out figures as the ceiling of that range.

Carbon beyond the grid: Grid carbon is shown in grams per kWh. All-energy carbon adds every other ton of CO₂ from burning gas, coal and oil, in cars, homes and industry, and is shown in metric tons per person a year, so the two never get mixed up. For each country it comes from the Global Carbon Project’s fossil CO₂ by fuel for 2024 (via Our World in Data), minus what that country’s power plants emit (this tool’s generation by fuel times the direct rates the EIA data confirmed: 0.42 t/MWh for gas, 1.04 for coal, 0.97 for oil). US states and Canadian provinces take their country’s figure in proportion to the fuel they burn off the grid. The totals land close to published figures: about 14 metric tons per person in the US, 7 in Germany, 2 in India and 4.5 for the world. The grid part is lifecycle, as everywhere in this tool, and the off-grid part is direct. Replacing fuels beyond the grid removes the off-grid part and adds the emissions of the clean power that replaces it. This measured figure replaced a single factor of 180 g per kWh of off-grid fuel, which was close for the world (the data gives 173) but off place by place.

Carbon kept out of the air, and warming: The tool adds up the CO₂ a build keeps out of the air between now and 2100. It assumes nothing is avoided until the build’s first plants come online, after the shortest lead time in the mix, then a straight-line ramp to the finish, and then its full yearly amount until 2100, compared with the same grown grid served at today’s carbon intensity. That comparison assumes no other cuts would have happened, so read it as an upper bound. It is set against the world’s fossil CO₂ in a year, about 36 billion metric tons in 2024 (Global Carbon Project). For the whole world, it is turned into avoided warming with the IPCC’s central estimate of 0.45°C per 1,000 billion metric tons of CO₂, with its likely range of 0.27 to 0.63, from the Sixth Assessment Report. That covers CO₂ from energy only, not methane, other gases or land use. For a single country the avoided warming is a tiny fraction of a degree, so it is not shown. The tool does not use a remaining carbon budget: the latest estimate gave 130 billion metric tons for a 50% chance of staying under 1.5°C from the start of 2025, which at today’s emissions runs out within about three years (Forster et al., Indicators of Global Climate Change 2024). Human activity has warmed the planet by about 1.36°C so far (same source, for 2024).

What avoided means here: Avoided carbon and lives saved compare the grid after the build with the same grown grid served at today’s mix: growth and energy-poverty uplift would otherwise be met by the plants running now. Comparing absolute emissions instead would score clean power serving new demand as added emissions. The all-energy total after the build is still the actual emissions, so a fast-growing grid can show a large avoided figure and a rising total at the same time. A rebuild retires today’s clean plants along with the fossil ones, so their deaths stop too, and rebuilding a grid with the mix it already has saves roughly nothing, as it should. Where a resource ceiling leaves part of the build with nowhere to go, carbon and lives are counted only for the part actually built; the rest of the demand stays on today’s mix.

Capturing the grid’s own carbon: Solid-sorbent direct air capture uses about 2 MWh per metric ton of CO₂ once the heat to regenerate the sorbent is counted, the middle of the published range for commercial designs, so one megaton removed costs two terawatt-hours. The toggle sizes capture to the lifecycle emissions of the grid being built, so a mix cleans up after itself. Capture plants draw from the grid they are cleaning, so removing a metric ton means removing a little more: capture equals gross emissions divided by (1 − 2 MWh/t × the grid’s intensity in metric tons per MWh). That converges only below 500 g/kWh; above it every capture plant emits more than it removes, and the tool refuses to print a number. Above 250 g/kWh the capture plants would draw more power than the rest of the grid combined, so the tool declines there too and shows the mix without capture. Capture is sized on the mix the build actually ends up with, after resource ceilings have moved shares, and on the fleet actually installed. The plants are costed on top of the power they draw: about $1,260 per metric ton a year of capacity, the midpoint of the National Academies’ estimate for a generic million-ton plant ($675 million to $1.26 billion in 2016 dollars), running 90% of the time for 30 years, with upkeep of 3.9% of capital a year plus $8 a metric ton for chemicals and waste. Piping the CO₂ away and storing it in a deep saline formation adds $12.50 a metric ton, the charge NETL uses in its power-plant studies. On today’s prices the plants cost $2,600 per metric ton a year, the reported price of Occidental’s Stratos plant in Texas. Whether a region has the rock to store what air capture catches is not checked; gas plants with capture are checked in US states (see Capital cost).

Lives saved

Lives saved: Deaths per TWh of electricity from Our World in Data (Ritchie), building on Markandya & Wilkinson (2007, The Lancet) and Sovacool et al., covering accidents and air pollution across each fuel chain: nuclear 0.03, wind 0.04, solar 0.02, hydro 1.3, biomass 4.6, gas 2.8, coal 24.6, oil 18.4. Those are the world defaults; biomass, gas, coal and oil get each country’s own rate, as the next paragraph explains. The nuclear figure includes Chernobyl and Fukushima, where nearly all deaths came from the evacuation rather than radiation, so 0.03 is conservative against nuclear. New plants’ deaths are counted on the fleet actually installed, overbuild included, like their lifecycle carbon. The headline separates lives saved on the grid from lives saved by replacing fuels burned off it.

Air pollution, country by country: One global rate per fuel meant a TWh of coal killed as many people in Ohio as in Uttar Pradesh, and the real difference is more than tenfold. Coal’s rate now comes from McDuffie et al. (2021, Nature Communications, “Source sector and fuel contributions to ambient PM2.5 and attributable mortality across multiple spatial scales”), which combined an atmospheric chemistry model, satellite exposure estimates and the Global Burden of Disease risk curves to attribute every country’s 2017 deaths from outdoor fine particles to their sources. Each country’s deaths from coal burned for power and heat, divided by its 2017 coal generation, gives deaths per TWh that reflect both how dirty its plants are and how many people breathe the result. Emissions alone cannot do that: Chinese coal plants now emit less sulfur dioxide per MWh than American ones, after China’s ultra-low-emission retrofits, but they sit among far more people. Each 2017 rate is moved to today by the change in coal-power SO₂ and NOx per MWh since, from the CEDS emissions inventory as compiled in Hausfather’s counterfactual repository. A country gets its own rate only if its 2017 coal fleet was at least 5 TWh and at least half of today’s; smaller fleets give noise, and a fleet mostly built since 2017 is not the one the rate describes. Those take the world rate, 15.4 deaths per TWh, the average of the 42 country rates weighted by today’s coal power. The United States comes to 2.4, which matches independent work putting US coal-plant sulfur deaths near 1,600 in 2020 (Henneman et al., Science, 2023); China to 8.8; India to 42.8. Biomass, gas and oil keep their global rates, scaled by the same ratio of the country’s coal rate to the world’s, so that gas never reads deadlier per TWh than coal in the same country. The study’s separate figure for non-coal energy was not used for them, because worldwide it equals coal’s and includes refineries and fuel production. These are deaths from fine particles only; ozone is not included.

Fuels burned off the grid, and why this figure is a floor: Fuel burned outside the power sector has no electricity to divide by, so when “replace fuels beyond the grid” is on it is counted at a blended 12 deaths per thermal TWh of fuel energy. That sits above every power-sector fuel on a common basis (gas power is about 1.4 per thermal TWh, coal 9.4 at 38% plant efficiency, oil 7.0), deliberately: a tailpipe or a home boiler has no tall stack and no scrubbers, and emits in the street. Transport, heating, industry and cooking burn several times the fossil energy power stations do, so at full global replacement this term supplies about 1.5 million of roughly 1.8 million lives a year. It is still a floor. For fossil fuels specifically, Lelieveld et al. (2019) estimate 3.6 million deaths a year and Vohra et al. (2021) 8.7 million, a spread driven by the concentration-response function each chooses; matching them would take about 26 and 67 deaths per thermal TWh. The low number is kept on purpose, since a tool that reaches for the largest estimate is easy to dismiss. Synthetic fuel is not harmless: in Transport & Environment’s tests of synthetic gasoline in a Mercedes A180, particle number fell 81–97% but NOx and particulate mass were unchanged and ammonia doubled on some blends, and those are what form most fine-particle harm. So the directly electrified share (50% by default) is credited in full and the synfuel share at half by default, adjustable to none or all under synfuel air quality; the gains are mostly sulfur, large for marine fuel and heating oil and small for road diesel.

Beyond the grid: fuels, water and energy poverty

Beyond the grid: Fossil fuel burned outside the power sector, in transport, heat and industry, is replaced two ways. A share (50% by default) is electrified directly, which delivers the same service on far less energy: an electric car uses about 30% of the fuel energy a gasoline car burns, a heat pump about 30% of a furnace’s. The tool applies a service-efficiency factor of 0.40 by default, selectable from 0.30 for transport-and-heat-heavy regions to 0.55 for heavy industry, where high-temperature heat gains little. The rest becomes synthetic fuel, at fuel energy divided by a power-to-fuel efficiency of 45% by default (published power-to-liquid chains run 40 to 50%, ICCT about 52%, pure hydrogen 65 to 70%). Direct electrification, wherever possible, beats making synthetic fuel; high-temperature industrial heat and some heavy transport are genuinely hard to electrify, which is what the industrial setting is for.

Ending energy poverty: When on, the build is sized so everyone in the region reaches a target yearly electricity use, on top of demand growth; places already above it see no change. North Americans use about 12,400 kWh a person, Europeans about 6,000, much of the global South a few hundred. The targets run from the Modern Energy Minimum (1,000 kWh, Rockefeller Foundation and Energy for Growth Hub, household plus productive use) through a decent-living floor (3,500 kWh, built on temperate climates) and that floor with cooling (4,000 kWh, the default) to European (6,000) and US (12,385) use. The default carries cooling because Europe is no guide: EU households used about 50 kWh a person for space cooling in 2024, while global cooling is heading for 5,000 TWh by 2050, about 500 kWh a person, and an Indian household cooling to 24–26°C uses about 700 kWh a year. Adding 400 to 600 kWh to 3,500 lands at 4,000, almost exactly what the average person on earth already uses (3,912 kWh in 2025): everyone reaching today’s world average, in the climate they actually live in. Decarbonizing while lifting billions out of energy poverty is a far bigger build than decarbonizing today’s grid alone.

Desalination: The toggle adds the electricity to desalinate 100 liters per person a day at 4.0 kWh per cubic meter, about 0.146 TWh a year per million people: the middle of the 3.5 to 4.5 kWh/m³ that full-scale seawater reverse-osmosis plants use with pre- and post-treatment (a survey of 39 plants averaged 4.13; the best large plants reach about 2.8). At world scale that is about 1,035 TWh, near 3% of today’s generation. It appears only where there is saltwater, which includes the Caspian countries (Kazakhstan’s Aktau plant ran on the BN-350 fast reactor from 1973). It is a control because water scarcity is one of the clearest cases for firm power sited anywhere: MIT’s Buongiorno group notes that a reverse-osmosis plant beside a reactor gets cheap electricity and can share its seawater intake and outfall, and retrofits onto plants already built. It covers drinking and household water only, not irrigation, and adds to the build rather than reshuffling it.

How long plants last

Plant lifetimes: Reactors run 80 years, the subsequent-renewal term US reactors are now being cleared for; wind farms 30 years, what US owners now assume (Lawrence Berkeley National Laboratory found an average of 29.6 years, up from about 20 in the early 2000s); solar farms 30 years, NREL’s project life for utility-scale PV, with modules now warrantied for 30. All three are planning lives rather than design minimums: 80 years is longer than the 60 today’s large reactors are designed for, and wind’s 20-year IEC 61400 design life is a certification floor, not a retirement age. Lifetime sets how much of a fleet must be rebuilt each year to hold its output, a little over 1% for nuclear and 3.3% for wind and solar at these defaults. The units table gives that pace for each source as machines a year once its first units wear out, and the time bars mark the year a source’s first units come due with ↺. It does not change build materials, because each source’s material intensity is converted at the life its own study assumed. Advanced offers 40 to 100 years for reactors, 20 to 35 for wind and 25 to 40 for solar.

Guess Watt It Takes

The quiz: Guess Watt It Takes asks eight questions about one place and draws each answer over the guess. Every answer is this calculator’s own result for that place at its standard settings, the ones a first visit opens with, taken from the same kind of run as the simple view’s three-column table, so at those settings the two always agree. If you have changed settings, the quiz still uses the standard ones, so everyone who plays a place answers the same questions, and the result screen says when the calculator would show you other figures. Four answers depend on what gets built: land, finish year, materials and the monthly cost. For those the answer is the range across the simple view’s three mixes (wind, water and solar only; the IEA net-zero plan; mostly nuclear), and any guess inside the range counts as right.

Scoring: A guess inside the range, or within 25% of its nearer end, is right (green). Within double or half is close (gold). Anything further is way off (orange). Four questions use fixed margins instead: the finish year allows 3 years for right and 10 for close, today’s clean share 5 and 15 percentage points, the grid’s share of carbon 7 and 20, and grid growth 0.2 and 0.6 times. For the biggest clean source, the runner-up counts as close if it makes at least half as much as the leader, and sources within half a percentage point of the leader tie for first. Places where clean sources make under 1% of the electricity skip that question. Land, materials and the monthly cost use sliders that move by multiples, from well below the answers to well above them, so a guess can land as close in a small place as in a big one.

Today’s grid: Clean electricity counts nuclear, hydro, wind, solar and geothermal, as the today bar does; biomass is left out. Shares are of the electricity generated in the place, rooftop solar included: US state figures add EIA’s estimate of it. The grid’s share of carbon is the power sector’s CO₂ as a share of all the CO₂ from burning fuel there, the same two figures as the simple view’s carbon bars.

The maps: Outlines are Natural Earth 1:50m boundaries, each drawn in an equal-area projection fitted to the place (Albers, or cylindrical equal-area near the equator), so a square of land drawn on it is to scale. The United States draws the lower 48, and small islands far from the mainland are left off. The world, and Morocco and Cyprus, whose borders are disputed, are drawn as a square of their area instead. Where the biggest answer would not fit beside the outline, as in Singapore, the picture zooms out until it does. The guess square can grow past the outline, and rings mark squares too small to see at the place’s scale.

Sources and reference figures

Papers and reports, with links: Sepulveda, Jenkins, de Sisternes & Lester 2018, Joule, firm low-carbon resources; Sepulveda et al. 2021, Nature Energy, long-duration storage; Buongiorno et al. 2018, MIT, The Future of Nuclear Energy in a Carbon-Constrained World; US Department of Energy, Pathways to Commercial Liftoff: Advanced Nuclear; Tong et al. 2021, Nature Communications, reliability of wind and solar worldwide; Shaner et al. 2018, Energy & Environmental Science, US wind and solar reliability; Fan Tong et al. 2020, iScience, least-cost storage and overbuild; Ruhnau & Qvist 2022, Environmental Research Letters, multi-year storage needs; Gernaat et al. 2017, Nature Energy, hydropower potential; Lovering et al. 2022, PLOS ONE, land use of electricity; Markandya & Wilkinson 2007, The Lancet, electricity and health; Our World in Data, deaths per TWh; McDuffie et al. 2021, Nature Communications, air-pollution deaths by source and country; Lelieveld et al. 2019, PNAS, fossil-fuel air pollution; Vohra et al. 2021, Environmental Research, fossil-fuel air pollution; National Academies 2019, Negative Emissions Technologies and Reliable Sequestration; NETL 2022, direct air capture case study; NEA 2012, The Economics of Long-term Operation of Nuclear Power Plants; IEA, World Energy Outlook 2025; IRENA, Renewable Power Generation Costs in 2025; NREL, Annual Technology Baseline 2025; World Bank, Commodity Markets Outlook, October 2026; Energy Institute, Statistical Review of World Energy 2026; Ember, electricity data; AtkinsRéalis 2026, CANDU Monark power rating; Nuclear Engineering International 2026, CANDU offer to Poland; NucNet 2026, Darlington refurbishment completed; Hu & Tang, Canadian Nuclear Society, CANDU and PWR fueling costs; Shirvan 2024, MIT CANES, total cost projection of the next AP1000; NEA and EPRI 2026, The Costs of Generating Electricity 2025; UNECE 2026, Protocol for a Full System Value and Cost Approach for Resilient and Carbon Neutral Electricity Systems, ECE Energy Series No. 83; EIA 2026, Assumptions to the Annual Energy Outlook 2026: Electricity Market Module; NREL ATB, utility-scale battery storage (upkeep includes augmentation); World Nuclear News 2025, the budget for OPG’s BWRX-300s; Net Zero Teesside Power 2024, financial close; AtkinsRéalis 2026, Enhanced CANDU 6 notice to the US NRC; World Nuclear Association, nuclear power in Argentina (Embalse refurbishment); World Nuclear News 2024, Kakrapar 3 and 4; NTPC 2025, Mahi Banswara; Business Standard 2017, Kakrapar coolant channel replacement; Gutiérrez-Negrín 2024, Geothermal Energy, worldwide geothermal power and national potential estimates; Stefánsson 2005, World Geothermal Congress, world geothermal assessment; Muraoka and others 2008, AIST, geothermal resources of Japan; MBIE 2020, Future Geothermal Generation Stack, New Zealand’s geothermal resources (after Lawless 2005); Unione Geotermica Italiana 2016, Stime di crescita della geotermia in Italia; Iceland GeoSurvey (ÍSOR) 2025, updated assessment of Iceland’s geothermal resources; Liu, Wang, Zhu & Li 2021, Energy Exploration & Exploitation, China’s geothermal resources; US Department of Energy, US Energy & Employment Report; IREC, National Solar Jobs Census; IRENA 2017, Renewable Energy Benefits: Leveraging Local Capacity for Onshore Wind; IRENA 2017, Renewable Energy Benefits: Leveraging Local Capacity for Solar PV; IEA 2022, Special Report on Solar PV Global Supply Chains; Georgia Public Service Commission, Vogtle construction monitoring reports; Dailian 2019, KHNP on the Shin Kori 5 and 6 workforce; Eash-Gates et al. 2020, Joule, sources of cost overrun in nuclear construction; Lok Sabha, December 2024, contract workers at NPCIL’s operating stations; India’s Ministry of Labour and Employment 2024, central minimum wages from October 2024; NTPC 2025, Business Responsibility and Sustainability Report FY2024-25; ReNew 2026, Business Responsibility and Sustainability Report FY2025-26; NYSERDA and New York Department of Public Service 2026, Master Plan for Responsible Advanced Nuclear Development in New York.

Solar and onshore wind prices by place: IRENA, Renewable Power Generation Costs in 2025, in 2024 dollars. Utility-scale solar, then blended with rooftop at the rooftop share exactly as the global price is: $519 per kW in China, $531 in India, $655 in Brazil, $802 in Europe, $816 in Africa, $845 in South America outside Brazil, $859 in Asia outside China and India (the Middle East and Central Asia included), $864 in Canada and Mexico, $897 in Oceania, $907 in Central America and the Caribbean, $972 in Eurasia (Russia, Türkiye and the Caucasus) and $1,150 in the US. Onshore wind: $769 in China, $1,142 in India, $1,249 in Brazil, $1,280 in South America outside Brazil, $1,361 in Eurasia, $1,510 in Africa, $1,625 in the US, $1,669 in Asia outside China and India, $1,681 in Europe, $1,741 in Oceania and $1,764 in Canada and Mexico. The global averages, $650 for solar and $951 for wind, are weighted by where projects were built, and China built most of them, so global wind is cheaper than every region’s but China’s. They price the world as a whole, and wind in Central America and the Caribbean, where IRENA gives no regional figure. Where IRENA reports a country on its own, which it does for the 30 largest onshore wind markets of 2025 and the 15 largest solar markets, that country’s figure replaces its region’s. Some sit far from their region: onshore wind cost $2,895 per kW in Japan and $2,215 in South Korea, against $1,669 for Asia outside China and India, and $1,781 in Germany and $1,883 in France against Europe’s $1,681, while Spain’s wind ($1,386) and solar ($672) came in below. Canadian provinces take Canada’s wind figure, $1,666. A single country rests on fewer projects than a region and moves more from year to year, so read these as one year’s prices. The full list is under Capital cost, below.

Hydro ceilings for US states: It is each state's current hydro generation plus two ORNL assessments for the Department of Energy that cover disjoint resources and so add: new stream-reach development on rivers carrying no hydro today (Kao et al. 2014, DOE/EE-1063, Table ES.2), plus turbines added to existing dams that have no powerhouse (Hadjerioua et al. 2012, DOE/EE-0711, Table 4, capacity converted at the 42.6% capacity factor implied by that study's own 12.1 GW and 45 TWh national totals). Summed with existing generation this puts the United States at about 770 TWh a year, and the national figure is computed as the sum of the fifty states so that picking “United States” and picking every state cannot disagree. Three caveats a reviewer should have. This is the unrestricted stream-reach total of 460 TWh; excluding reaches close to national parks, wild and scenic rivers and wilderness areas cuts it to 347 TWh, which would put the US near 640. That exclusion is policy rather than physics, and this ceiling is meant to be physical, so the larger figure is used. Alaska carries no ORNL generation estimate, only 4,723 MW of capacity derived by a different method from the lower 48, converted here at 50% because Alaskan flow is snowmelt-driven, below ORNL's 53–71% range for the other states. And a 2024 reassessment cuts national non-powered-dam potential from 12.1 GW to 4.1 GW on tighter screening but publishes no state table, so that term is an upper bound. Note the definitional asymmetry this creates: countries carry Gernaat's economic potential while US states carry ORNL's technical potential, which is the looser of the two. The US ceiling is therefore more permissive than France's, which cuts against this tool's own framing rather than for it.

Geothermal ceilings: For US states the ceiling is the US Geological Survey’s national assessment (Williams et al., Assessment of Moderate- and High-Temperature Geothermal Resources of the United States, Fact Sheet 2008-3082): each state’s mean identified resource plus its mean estimated undiscovered resource, converted at the survey’s own 8.77 GWh a year per MWe. The thirteen assessed states sum to 9,057 MWe identified and 30,033 MWe undiscovered, and USGS states that they hold every identified moderate- and high-temperature system, so every other state’s ceiling is zero. California’s comes to about 147 TWh a year, Nevada’s 51, and the United States’ 343, the sum of its states. Minnesota’s is zero: its heat flow is low (Gosnold et al., American Geophysical Union, 2011), and its geothermal potential lies in engineered systems at depth (Klenner, University of North Dakota). For countries, the ceiling comes from national assessments of conventional potential compiled in Gutiérrez-Negrín’s review, Evolution of worldwide geothermal power 2020–2023 (Geothermal Energy, 2024): Indonesia 23,357 MW, the Philippines 4,064, Mexico about 2,500, Kenya about 10,000, and Ethiopia 4,200 to 10,800, taken at its midpoint. A country with no geothermal power plants today gets no conventional resource beyond what it already runs. That follows the International Energy Agency’s The Future of Geothermal (2024), which finds conventional geothermal concentrated in a few countries with accessible, high-quality resources and calls India’s conventional potential highly limited; India’s often-quoted 10,600 MW is heat from hot springs, not electricity. Japan takes AIST’s national assessment, 23,470 MW (Muraoka and others, 2008). Ten more countries take the estimates in the same review: Türkiye 9,000 MW, Chile 16,000 (its ministry’s 2,086 is a build estimate for 2050, a different measure), Costa Rica about 1,000, Guatemala and Nicaragua about 1,000 and 1,500, El Salvador 550, Honduras 116, Colombia 1,170, Russia 1,650 (Kamchatka and the Kurils) and Taiwan 979, from hydrothermal resources only. Four take their own national assessments: New Zealand 3,600 MW, the resource capacity of all its high-temperature fields (Lawless, 2005, as summarized for New Zealand’s energy ministry); Italy 5,000, the Unione Geotermica Italiana’s conservative figure for medium- and high-temperature resources suitable for power (2016), the same temperature reach as the US survey; China 8,620, the China Geological Survey’s 7,120 MW from high-temperature resources in southern Tibet, western Yunnan and western Sichuan plus 1,500 MW from medium-temperature ones (Liu and others, 2021); and Iceland about 4,200, the middle of Iceland GeoSurvey’s 1,860 to 6,520 MW from high-temperature fields over 50 years (updated national assessment, 2025). The World takes the low end of Stefánsson’s world assessment (World Geothermal Congress, 2005): 240 GW most likely from fields already identified, and 5 to 10 times that hidden, so 1 to 2 TW in all; 1 TW counts identified and hidden fields together, the same reach as the USGS figure behind the US states, whose undiscovered resources are 3.3 times the identified. Countries that already generate geothermal power but have no assessment in this tool, among them Germany and France, are left uncapped rather than given a guessed ceiling, and so are Canada and its provinces.

Electricity data: Energy Institute Statistical Review of World Energy (2025 generation by fuel, by country). Country areas are land areas, inland water excluded, from the World Bank’s land area series (FAO); the world is Earth’s land surface, and Taiwan, which the series omits, keeps its official 36,197 km². Non-electric fossil energy is each region's gas, coal and oil consumption minus the share burned in power plants, from the same source.

US states: Generation by fuel from EIA-923 2024 final data for utility-scale plants, plus EIA’s estimate of small-scale rooftop solar, about 84 TWh nationally, from the Electric Power Annual, Table 3.21; Puerto Rico has no such estimate and stays utility-scale only. Areas are land areas from the Census Bureau’s state area measurements. Populations from Census Bureau 2024 estimates. Non-electric fossil energy is apportioned from the national total by population share, an approximation, since energy-heavy industrial states burn more than their share. Note that the 51 state rows sum to about 4,398 TWh while the "United States" country row reads 4,772 TWh. The two come from different sources and years (EIA net generation for 2024 versus the Energy Institute’s gross generation for 2025), which accounts for most of the 8% gap. Pick one level or the other for a given comparison rather than mixing them.

Canadian provinces and territories: Generation by source is Statistics Canada table 25-10-0015-01 for 2024. StatCan groups biomass, gas, coal and oil as combustible fuels, so the split between them comes from AESO for Alberta, SaskPower for Saskatchewan, the CER and Nova Scotia Power for Nova Scotia, and plant capacity for New Brunswick, Quebec, BC and the territories. The New Brunswick split is the least certain. Population is StatCan’s July 1, 2025 estimate, and Canada’s own figure is the sum of the thirteen, so Canada and all its parts agree. Areas are land areas from Statistics Canada’s Table 15.7. Capacity factors, fleet age and population growth fall back to Canada's national figures. Remaining hydro potential shares out Canada's national ceiling in proportion to each province's current hydro, a proxy that understates the territories and Labrador. Retail prices are Hydro-Québec's 2025 residential comparison for each province's largest city, which runs higher than the all-sector average used for US states. Provincial generation sums to 609 TWh against Canada's 660 TWh, because the national row is the Energy Institute's 2025 figure and StatCan's monthly table leaves out some industrial self-generation. Nunavut is not one grid: Qulliq Energy runs 25 separate diesel systems, one per community, so a single Nunavut grid understates how many reactors it would take. Reactors are sized to the grid they join, no single unit above 15% of peak on an islanded grid or 50% on a connected one, and a fleet takes at most one unit's construction time per unit, built one after another.

Quiz maps: Natural Earth 1:50m cultural vectors, Admin 0 countries and Admin 1 states and provinces, in the public domain.

Every constant lives in one labeled block at the top of this file's script, so numbers are easy to audit and update.

Revision history

Newest first, with earlier versions grouped.

v3.83 Each source’s build waits out its lead time before delivering; US states count rooftop solar; Canada uses StatCan’s population; shares of a place are of its land area.

v3.79 to v3.82 The calculator opens with Guess Watt It Takes, eight questions about your place answered at the standard settings, and says whether you are in the simple view or the full one; New York’s nuclear note follows its 2026 Master Plan.

v3.67 to v3.78 Build jobs come from measured construction records. India’s pay counts contract workers, geothermal is capped to published assessments worldwide, the one-pager shows all the CO₂ from energy, and modeled results are rounded to three significant figures.

v3.57 to v3.66 The simple view draws the build and the jobs. Pay in poorer countries comes from what power plants report paying.

v3.50 to v3.56 Costs fill gaps from NEA and EPRI’s 2025 study, and gas with capture checks for rock to store its CO₂.

v3.37 to v3.49 The results are reorganized into named sections, and pay follows the place.

v3.20 to v3.36 The methodology is rewritten with linked sources, and reactors fit the place: CANDUs in Canada, heavy-water reactors in India.

v3.00 to v3.19 Every source is priced at a mature cost beside NREL’s projections, with 2025 prices from IRENA.

v2.5 to v2.99 Capacity factors, land and build pace are measured place by place, and costs gain upkeep, interest, fuel and long-duration storage.

Before v2.5 The original model: 135 countries and every US state, with land, materials, build time, cost, lives and carbon for any mix.