Level

The model · US · to 2050

A US electricity transition model

This is a model. Unlike the rest of this site, its outputs depend on assumptions about the future that no citation can settle. Every assumption below is adjustable and disclosed. It takes decisions — build rates, retirement policy, demand growth — and evolves the fleet; the mix is an output, not an input.

It starts from the real US power fleet (EIA 2024) and answers “what would this fleet produce and cost,” not “could this happen” — transmission, markets, permitting and supply chains are out of scope. The base data is checked against EIA’s national totals; see every assumption, its source, and how the fleet reconciles.

How this model works

  1. 1. You make the decisions

    How fast to build solar, wind, gas, nuclear and batteries. Whether to retire ageing plants. How fast demand grows.

  2. 2. The model runs the fleet to 2050

    Year by year: old plants retire, new ones take years to build, and each source makes power on its own schedule — solar by day, nuclear around the clock.

  3. 3. You see what it produces

    The electricity mix over time, the deaths, CO₂, land and cost it implies — and whether it can actually keep the lights on, hour by hour.

Your scenario

How much to build each year (GW)

A big new power plant is roughly 1 GW. Drag to add more or less of each per year.

panels — cheap, but only makes power in daylight

turbines — variable, a bit stronger at night

stores a few hours of power for the evening

always-on, but takes ~7 years to build

the workhorse gas plant — flexible, burns fuel

gas plant for short demand spikes

1.0%
US demand has grown ~1%/yr; data centres and electric cars could push it higher.
When to retire old plants

In plain terms — your scenario

The short version of what the settings above produce. Dotted words have plain definitions — hover or tap them.

In 2050, this scenario can’t meet demand in about 27% of hours (roughly 5% of the electricity called for), with air-pollution and accident deaths down 63% vs today and climate pollution down 22% vs today. Every figure is a central estimate inside a range — the charts below show the full .

5% unmet
short in ~27% of hours, mostly evenings
Deaths per year
~7,800
↓ down 63% vs today
Climate pollution
~1,100 Mt
↓ down 22% vs today
Cost to run the fleet
~$370 bn/yr
2.5 $tn built 2025–2050

What the grid is made of, over time

Each coloured band is how much electricity a group of sources makes each year, stacked up. The dashed line is total demand. Where the colours stop below the dashed line, there isn’t enough — the fleet can’t meet demand. Select any year — pointer, tap or arrow keys — to read the numbers. (“Fossil fuels” combines coal, gas and oil; per-source detail is in the harm figures below.)

02.0k4.0k202520382050

Interactive chart. Use the left and right arrow keys to move year by year, Home and End for the first and last year, and Escape to clear. The same figures are available as a table below.

SolarWindFossil fuelsHydroOtherNuclear Demand
Year-by-year figures as a table
Generation mix in terawatt-hours per year. Demand shown as a dashed line.
YearNuclearOtherHydroFossil fuelsWindSolarDemand
2025782672432.5k4522194.3k
2026782662432.4k4483004.3k
2027782632432.4k4763804.4k
2028782602432.3k5014604.4k
2029777552432.3k5295404.5k
2030760492432.2k5516204.5k
2031742442432.1k5737004.6k
2032725392432.1k5867804.6k
2033721382432.1k5918604.7k
2034671352432.2k5929414.7k
2035641332432.2k6071.0k4.7k
2036635292432.1k6171.1k4.8k
2037603282432.0k6081.2k4.8k
2038602272431.8k6351.3k4.9k
2039611262431.8k6501.3k4.9k
2040611262431.7k6551.4k5.0k
2041592242431.7k6591.5k5.0k
2042599242431.8k6691.6k5.1k
2043598232431.8k6781.7k5.1k
2044559232431.8k6831.7k5.2k
2045508232431.9k6701.8k5.2k
2046448222431.9k6591.9k5.3k
2047402222431.9k6601.9k5.3k
2048369212431.9k6712.0k5.4k
2049365202432.0k6852.1k5.5k
2050353192392.0k7152.1k5.5k

The harms and the cost

The model applies the same per-source figures as the rest of the site to the electricity above. Each result is a shaded , and the range widens further into the future, where less is knowable. Switch to “Added up” to see the total over all years.

Deaths (deaths/yr)
050k100k150k202520382050
CO₂ (Mt CO₂/yr)
01.0k2.0k202520382050
Cost (USD bn/yr)
0200400600202520382050
Land (km² (annual footprint))
0100k200k202520382050

Land is a standing footprint, not a flow, so it is always shown annually — a cumulative integral would be area·years.

Can it keep the lights on?

The real test isn’t the yearly total — it’s every hour. Below is one typical summer day in 2050. Supply stacks up from the bottom; the dashed line is demand. Where the colours can’t reach the line, that hatched gap is — the lights flicker. Notice how solar vanishes through the evening and overnight just as demand stays high.

0500GW12am6am12pm6pmnightmorningafternoonevening

Interactive chart. Use the left and right arrow keys to move through the 24 hours, Home and End for the first and last hour, and Escape to clear. The same figures are available as a table below.

Always-on (nuclear, hydro…)Gas & coalWindSolarBatteryNot served Demand
Hour-by-hour figures as a table
Supply versus demand across a representative summer day in 2050, US local time
Local timeAlways-on (nuclear, hydro…)Gas & coalWindSolarBatteryDemandNot served
12am70 GW336 GW88 GW3 GW0 GW651 GW154 GW
1am70 GW336 GW86 GW2 GW0 GW621 GW128 GW
2am70 GW336 GW83 GW1 GW0 GW599 GW109 GW
3am70 GW336 GW80 GW1 GW0 GW586 GW100 GW
4am70 GW336 GW76 GW3 GW0 GW584 GW99 GW
5am70 GW336 GW73 GW16 GW0 GW591 GW96 GW
6am70 GW336 GW70 GW66 GW0 GW605 GW64 GW
7am70 GW308 GW65 GW185 GW0 GW627 GW0 GW
8am70 GW166 GW58 GW366 GW0 GW658 GW0 GW
9am70 GW54 GW55 GW516 GW0 GW695 GW0 GW
10am70 GW0 GW54 GW645 GW0 GW734 GW0 GW
11am70 GW0 GW54 GW718 GW0 GW771 GW0 GW
12pm70 GW0 GW54 GW743 GW0 GW802 GW0 GW
1pm70 GW0 GW54 GW746 GW0 GW826 GW0 GW
2pm70 GW0 GW56 GW731 GW0 GW845 GW0 GW
3pm70 GW33 GW58 GW698 GW0 GW859 GW0 GW
4pm70 GW93 GW61 GW643 GW0 GW867 GW0 GW
5pm70 GW174 GW64 GW557 GW0 GW865 GW0 GW
6pm70 GW270 GW67 GW444 GW0 GW851 GW0 GW
7pm70 GW336 GW71 GW302 GW52 GW831 GW0 GW
8pm70 GW336 GW75 GW179 GW140 GW805 GW6 GW
9pm70 GW336 GW82 GW85 GW0 GW770 GW198 GW
10pm70 GW336 GW87 GW20 GW0 GW728 GW216 GW
11pm70 GW336 GW89 GW5 GW0 GW687 GW188 GW

Reliability in 2050

hourly dispatch

261 TWh

4.7% of demand went unmet

Hours short

2,360 h

of 8,760 h in the year

80.4 TWh

available but nowhere to put it

-10%

780 GW dependable · 867 GW peak

When shortfalls fall — by season

Winter140 TWh
Spring0 TWh
Summer120 TWh
Autumn0 TWh

When shortfalls fall — by time of day

12amnoon12am

Unserved energy is the gap between demand and what this fleet can dispatch, hour by hour. It is reported, not judged: a scenario with little unserved energy may carry high deaths or CO₂, and vice-versa. Dispatch uses representative seasonal days, so it captures timing, not the single worst hour of a real year.

The coefficients behind the death figures

The model reuses these exact per-TWh death rates — the same chart, hatch pattern and log/linear scale as the rest of the site. Solid is counted deaths, hatched is modeled.

Deaths per terawatt-hour

Coal1 death / 670–6k
24.6–224
Oil1 death / 3k–8k
18.4–52
Gas1 death / 18k–54k
2.8–8.5
Biomass1 death / 13k–33k
4.6–12
Hydro1 death / 94k–3.8M
0.04–1.6
Nuclear1 death / 2.1M–5.0M
0.03–0.07
Wind1 death / 1.9M–7.5M
0.02–0.08
Solar1 death / 2.5M–15M
0.01–0.06

This scale is logarithmic: each step to the right is ten times the previous one, so the visual distance understates how much larger the fossil numbers really are.

counted deaths · modeled deaths

The small grey figure under each source is the same rate turned human: roughly how many people’s yearly electricity corresponds to one death. Coal, about 700–6,000 people; solar, a few million.

The figures as a table
Deaths per terawatt-hour by source, low to high, with the counted and modeled split.
SourceLowHighModeled1 death per
Coal24.622495%670–6k people
Oil18.45294%3k–8k people
Gas2.88.590%18k–54k people
Biomass4.61286%13k–33k people
Hydro0.041.62%94k–3.8M people
Nuclear0.030.0770%2.1M–5.0M people
Wind0.020.0825%1.9M–7.5M people
Solar0.010.0625%2.5M–15M people

Plain-language glossary

capacity

How much electricity a power plant CAN make at full tilt, measured in gigawatts (GW). Different from how much it actually makes over a year.

generation

The electricity a plant actually produces over time, measured in terawatt-hours (TWh). One TWh powers roughly 90,000 US homes for a year.

capacity factor

The share of the time a plant runs, averaged over a year. Solar is ~20% (only in daylight); nuclear is ~90% (almost always on).

unserved energy

Electricity that demand called for but the fleet could not supply in that hour. The model reports it plainly — it is the reliability cost of a scenario.

reserve margin

How much dependable, weather-independent capacity the grid has above its highest-demand hour. Positive is a cushion; negative means the firm fleet alone cannot meet the peak.

firm capacity

Capacity you can count on regardless of weather — nuclear, hydro, gas, coal, batteries. Wind and solar are not firm, because a calm night can zero them out.

ELCC

Effective Load Carrying Capability — how much a wind, solar or battery fleet counts toward keeping the lights on. It falls as you build more of the same thing (the 10th solar farm helps the evening peak far less than the 1st).

learning rate

How fast a technology gets cheaper as the world builds more of it. A 20% learning rate means every doubling of total installed capacity cuts the price about 20%.

uncertainty band

A low-to-high range instead of a single number, because the underlying science is a range. The band is wider further into the future, where less is knowable.

cumulative

Added up over all years so far, rather than a single year. Cumulative CO₂ is what matters for the climate; a single year is just one slice.

curtailment

Clean electricity that was available but thrown away because there was more than demand needed and nowhere (no battery) to store it.

net-summer capacity

A plant's dependable output on a hot summer afternoon — a bit below its nameplate rating. The standard basis US grid planners use.

Every number here is reproducible from the committed EIA 2024 snapshot and the pure model engines. Nothing is fetched at runtime. See every assumption →