Method and sources
How every number on this site is worked out, where it comes from, and when it changes. Every figure links here.
Humanity's power
Humanity's power is the world's total energy consumption in the latest full year that EIA has published (series INTL.44-2-WORL-QBTU.A, "Total energy consumption, World, Annual", in quadrillion Btu), turned into average power: the year's energy in joules divided by the seconds in that year.
power (W) = quadrillion Btu × 1015 × 1,055.05585262 J ÷ (days in the year × 86,400 s)
For 2024: 606.0 quadrillion Btu is 6.39
EIA counts fuels by their energy content and converts electricity from nuclear, wind, solar and hydro by its own convention, so other publishers (the Energy Institute, the IEA) give different world totals; we use EIA alone and never mix sources.
On every build we check this conversion against EIA's own figures in terajoules (series INTL.44-2-WORL-TJ.A) for every year, and we check the series name, units, range and the change from the year before. If a download fails any check, the site keeps its previous figures and their dates.
Source: U.S. Energy Information Administration, International Energy Statistics (updated March 2026), retrieved 12 October 2026 · data year 2024 · Refreshed when EIA publishes a new year
Kardashev rating
We use the decimal Kardashev scale (after Carl Sagan, 1973), the version Wikipedia uses, with power P in watts:
K = (log10 P − 6) ÷ 10
Sagan proposed the steps; the formula is the usual way of writing them (Gray 2020; Ćirković 2015, open access, which quotes Sagan's 1016 W).
So Type I is 1016 W (K = 1), Type II is 1026 W (K = 2) and Type III is 1036 W (K = 3). Ratings are shown to 3 decimals. For 2024: log10 of 20.2 TW is 13.306, so K = 0.731.
A second common definition of Type I is all sunlight on Earth: the IAU's nominal solar irradiance (1,361 W/m²) times Earth's cross-section, π × (6,378.1 km)², about 1.74
Sources: the decimal Kardashev scale (after Carl Sagan, 1973; Gray 2020; Ćirković 2015, Serbian Astronomical Journal 191: 1–15); IAU 2015 Resolution B3 · fixed values, so no refresh
Share of the Sun's output
The share of the Sun's output is humanity's average power divided by the Sun's total output, taken as the IAU's nominal solar luminosity of 3.828
For 2024: 20.2 TW ÷ 3.828
On this site, a "sentient sun" means the Sun's whole output turned to computing — Elon Musk's idea, not a forecast. On the decimal scale the Sun's whole output rates 2.058.
Source: IAU 2015 Resolution B3 (a fixed value) · data year 2024 for humanity's power · Refreshed with it
Solar on Earth
Solar on Earth is the world's solar electricity generation from EIA (series INTL.116-12-WORL-BKWH.A, "Solar electricity net generation, World, Annual", in billion kilowatthours), turned into average watts the same way. We use electricity actually generated, not installed capacity.
power (W) = billion kWh × 109 × 3,600,000 J ÷ (days in the year × 86,400 s)
For 2024: 2,090 billion kWh over 366 days (31,622,400 seconds) is an average of 238 GW.
This is electricity, while humanity's total is energy consumption counted at the source, so we don't divide one by the other. EIA counts solar at about its electricity content.
Source: U.S. Energy Information Administration, International Energy Statistics (updated February 2026), retrieved 12 October 2026 · data year 2024 · Refreshed when EIA publishes a new year
When would humanity reach Type I?
The Kardashev calculator projects forward from the data year's value — never from today's date — at a steady growth rate you can change:
years to go = log(PType I ÷ Pdata year) ÷ log(1 + growth rate)
The default is 1.8% a year: EIA's world total energy consumption grew by an average of 1.76% a year from 1980 to 2024, rounded.
At 1.8% a year from 2024's 20.2 TW, humanity would reach 1016 W in about 2372, and all sunlight on Earth (1.74
Source: U.S. Energy Information Administration, International Energy Statistics (updated March 2026), retrieved 12 October 2026 · data years 1980–2024 · Refreshed with humanity's power
Dyson swarm
The Dyson swarm calculator works out what a swarm of solar collectors around the Sun would capture, how large and heavy it would be, and the computing it could run. You choose the share of the Sun's light it captures. Every other input starts at a default, given below and under Computing, and you can change it; only the distance is fixed, at 1 au.
power captured = share × 3.828
electricity = efficiency × power captured
collector area = share × 4π × (1 au)²
collectors = collector area ÷ collector size
mass = collector area × mass per m²
- Distance: 1 au, Earth's distance from the Sun, 149,597,870,700 m exactly (IAU 2012 Resolution B2, restated in 2015 Resolution B3). A sphere with that radius has an area of 2.81
× 1023 m². The Sun's output spread over it is 3.828 × 1026 W ÷ 2.81 × 1023 m² = 1,361 W/m², matching the IAU's 1,361 W/m² at 1 au. The calculator keeps this distance fixed. - Conversion efficiency: 30.7%, a Spectrolab XTJ Prime space solar cell at the beginning of its life (Spectrolab's XTJ Prime datasheet, 2022; listed in Table 3-1 of NASA's small spacecraft survey).
- Collector mass: 1 kg per m², about a bare space-grade solar cell: Spectrolab gives 84–130 mg/cm² (0.84–1.3 kg per m²) for the XTJ Prime.
- Whole arrays weigh more than bare cells. NASA's survey finds flown arrays clustered around 30 W per kg (200 W per kg the highest). If they delivered the XTJ Prime's 415.9 W per m² (NASA Table 3-1), that would be about 2 to 14 kg per m², likely high, since whole arrays deliver less per m² than a new bare cell.
- Collector size: 1 km² each, your choice: no measured “typical collector” exists.
- Comparisons. Earth's surface, 5.10
× 1014 m²: 4π × the mean radius squared (6,371.0084 km; JPL). The Moon's mass, 7.346 × 1022 kg: its GM of 4,902.800 km³/s² (JPL) divided by the gravitational constant. Mercury's mass, 3.30103 × 1023 kg (JPL). Humanity's power is the data year's, as above.
Sources: IAU 2015 Resolution B3; Spectrolab, XTJ Prime space solar cell datasheet (2022); Small Spacecraft Technology State of the Art, 3.0 Power (Table 3-1; specific power of flown arrays); JPL planetary and satellite physical parameters · fixed values, reviewed when a source updates
Computing
Computing is shown two ways, and the two are never added or compared directly: one counts arithmetic, in FLOP/s (calculations a second), the other the erasing of bits.
computing with today's chips = electricity × FLOP/s per watt
physical ceiling on erasing bits = electricity ÷ (k × T × the natural log of 2)
- Today's chips: 2,300 GFLOP/s per watt. AMD lists the Instinct MI350X at a peak of 2.3 PFLOP/s of dense BF16 matrix arithmetic (without structured sparsity, which AMD lists separately at 4.6 PFLOP/s) and 1,000 W typical board power: 2.3 × 1015 ÷ 1,000 = 2.3 × 1012 FLOP/s per watt. That counts the chip's board only, not cooling or the rest of a data center. Chosen as the highest dense BF16 FLOP/s per watt of published board power among the current data-center GPUs we checked (AMD Instinct MI355X: 2.5 PFLOP/s at 1,400 W). NVIDIA has not published board power for its Rubin GPU.
- The physical ceiling on erasing bits: the Landauer limit. Erasing one bit of information releases at least k × T × the natural log of 2 in heat (Landauer, 1961), where k is the Boltzmann constant, 1.380649
× 10−23 J/K exactly (NIST), and T the temperature. At the default 300 K that is 2.87 × 10−21 J per bit, so each watt allows at most 3.48 × 1020 bit erasures a second. The result is shown as “bit erasures a second”, not FLOP/s. - Temperature: 300 K, room temperature, a reference between a black ball at 1 au (about 278 K) and a flat black collector radiating from both faces (about 331 K), each found by balancing 1,361 W/m² of sunlight against the heat it radiates (σT4, σ = 5.670374
× 10−8 W/m²K⁴, NIST CODATA). The lowest allowed is 2.7 K, just under the temperature of deep space: the cosmic microwave background is 2.72548 K (Fixsen 2009).
Sources: AMD Instinct MI350X GPUs (specifications); AMD Instinct MI355X GPUs (specifications); R. Landauer, “Irreversibility and Heat Generation in the Computing Process”, IBM Journal of Research and Development 5(3): 183–191, July 1961, doi.org/10.1147/rd.53.0183; NIST, CODATA value: Boltzmann constant; NIST, CODATA value: Stefan–Boltzmann constant; D. J. Fixsen, “The Temperature of the Cosmic Microwave Background”, arXiv:0911.1955 (2009) · reviewed at each update for a better published figure per watt
History
The history uses the same EIA series for every year from 1980 to 2024, each converted the same way. Humanity's average power rose from 9.37 TW in 1980 to 20.2 TW in 2024. If EIA revises a past year, the next build uses the revised figure.
Source: U.S. Energy Information Administration, International Energy Statistics (updated March 2026), retrieved 12 October 2026 · data years 1980–2024