Power is not energy. Nameplate is not output.
What does 90 GW of AI power really require?
Turn a viral shorthand into an auditable capacity plan. Adjust data-center demand, efficiency, generation mix, capacity factors, build rates, emissions, and land assumptions.
The useful correction
90 GW is a rate. Plants are capacity.
If 90 GW means a continuous facility load, the system uses 788.4 TWh each year. Ninety 1-GW reactors have 90 GW of nameplate capacity, but at a 92% capacity factor they average about 82.8 GW. Supplying a flat 90-GW load with nuclear energy alone would require about 97.8 GW of nuclear nameplate capacity before reserve margin, transmission losses, or outages beyond the capacity-factor assumption.
The “90 plants” line is a useful order-of-magnitude image. It is not a complete grid plan.
Power90 GWinstantaneous demand
Annual energy90 × 8,760 = 788.4 TWhif sustained all year
Nuclear nameplate90 ÷ 0.92 = 97.8 GWat 92% capacity factor
Build a capacity plan
Start with IT demand. PUE adds cooling and power-distribution overhead; utilization converts peak facility capacity into average load. Generation shares are normalized automatically, so the energy mix always totals 100%.
AI electricity scenario
Ready. Adjust any assumption.
Server, accelerator, memory, storage, and network equipment.
Average IT draw as a share of IT nameplate load.
Facility power ÷ IT power. Lower is more efficient.
Extra firm capacity for contingencies; shown separately from annual energy.
Used to compare required capacity with assumed annual build rates.
Facility peak90.0 GWIT load × PUE
Average facility load90.0 GWafter utilization
Annual electricity788.4 TWhaverage GW × 8,760
Firm planning target103.5 GWaverage load + reserve
Claim check: Loading calculation…
Annual energy by source788.4 TWh total
Nameplate capacity required0 GW total
Nuclear
Gas
Wind
Solar
Natural gas
Wind
Solar
Compare the consequences
These are scenario outputs, not forecasts. Emissions use life-cycle median-style planning factors; land uses editable model defaults and project-area conventions that vary significantly by site.
Life-cycle emissions
Annual emissions from the normalized generation mix.
0 Mt
Equivalent context loading…Project land area
Approximate project area, including wind spacing rather than only disturbed land.
0 km²
Land assumptions shown below.Slowest build lane
Years required if each technology builds at its selected annual rate in parallel.
0 years
Compared with the selected horizon.| Technology | Normalized share | Capacity | 1-GW plant equivalents | Life-cycle g CO₂e/kWh | Project km²/GW |
|---|
Edit emissions and land assumptions
Nuclear
Natural gas
Wind
Solar
Three worked examples
Use these to sanity-check any headline. The formulas are intentionally visible so a GW claim can be translated without trusting the simulator.
Continuous 90-GW load
A grid load that never changes.
90 GW × 8,760 h ÷ 1,000 = 788.4 TWh/yearAt 92% capacity factor: 97.8 GW of nuclear nameplate, or about 98 one-GW units.
90-GW IT fleet
IT hardware at 85% utilization and PUE 1.25.
90 × 0.85 × 1.25 = 95.625 GW average facility loadAnnual energy becomes 837.7 TWh, larger than a flat 90-GW grid load because facility overhead matters.
Efficiency as supply
The same 90-GW IT fleet at 85% utilization, improving PUE from 1.25 to 1.10.
90 × 0.85 × (1.25 − 1.10) = 11.475 GW avoidedThat saves about 100.5 TWh/year, before any change in computing output.
Check your grid intuition
Each answer explains the unit or planning concept involved. A score only appears after all questions are answered.
What does 90 GW describe?
Why can 90 one-GW reactors average less than 90 GW?
What does PUE add to IT electricity?
Which value converts average GW into annual TWh?
What the shorthand leaves out
A grid plan must match demand hour by hour, not just over a year. These are the first follow-up questions to ask.
Does annual renewable energy equal firm 24/7 power?
No. Wind and solar capacity factors help estimate annual energy, but hourly production may not align with load. Storage, transmission, demand flexibility, overbuilding, and firm generation can close that timing gap. This simulator is an energy-and-capacity explainer, not a dispatch model.
Is reserve margin already included in the nameplate chart?
No. The chart calculates nameplate capacity needed to generate the selected annual energy at each capacity factor. The separate firm planning target shows the selected reserve margin. Converting that target into a reliable portfolio requires technology-specific capacity credit and outage analysis.
Are the land comparisons universal?
No. Site boundaries, disturbed land, turbine spacing, dual-use solar, transmission corridors, fuel infrastructure, and local geography produce very different answers. The defaults are transparent planning approximations and can be edited.
Does reducing PUE reduce computing demand?
It reduces facility overhead for the same IT load. It does not by itself make chips more efficient. Better hardware, software, scheduling, and model efficiency can reduce IT energy per unit of useful computation.
Sources and model boundaries
The simulator uses authoritative definitions and transparent defaults. It does not forecast AI adoption, grid prices, construction lead times, or local interconnection constraints.
U.S. Energy Information Administration
Capacity factor is actual output divided by maximum possible output. EIA reports the U.S. nuclear fleet at about 92% in 2024.
Capacity and generation definitionsU.S. Department of Energy
PUE is total data-center facility energy divided by IT equipment energy. DOE's current design guide cites an average around 1.6 and much lower values for highly efficient facilities.
Data center design guideNational Renewable Energy Laboratory
Default life-cycle factors use NREL harmonization work: roughly 12 g CO₂e/kWh for nuclear and wind, 45 for solar PV, and 450 for combined-cycle gas.
Life-cycle emissions datasetDOE data-center outlook
DOE reported 176 TWh of U.S. data-center electricity use in 2023 and a projected 325–580 TWh range for 2028. This tool lets you compare a 90-GW scenario with those annual totals.
2024 U.S. data-center energy report releaseReplace the headline with assumptions you can inspect.
Copy your scenario, share the exact parameterized link, or reset the model and test a different claim.