# U.S. AI electricity and household equivalents, 2024

All electricity and household data refer to the United States in 2024.

## Inputs and provenance

- AI-specialized servers: approximately 50 TWh, visually read from the dark teal segment of the 2024 bar in the user-supplied Figure 9, printed page 25, Berkeley Lab's United States Data Center Energy Usage Report: 2025 Update (published June 2026). This is segment thickness, not the height of its top above zero. The segment spans roughly 145–195 TWh on the axis. These endpoints are approximate; the precise sector total is 192 TWh in the report text. This visual reading is not an exact tabulated AI value.
- Whole-sector electricity: 192 TWh, reported in the report's overview.
- AI-facility PUE: 1.145 for 2024, report section 3.5, printed page 24. PUE = total facility electricity / IT electricity. Applying it to AI servers allocates facility overhead in proportion to their electricity use.
- Household denominator: 863 kWh/month per U.S. residential utility customer, EIA 2024 Table 5A; 863 × 12 = 10,356 kWh/year. The monthly figure is rounded. This is mean purchased electricity per utility customer, not median consumption or all household energy.

Report: https://www.energy.gov/documents/united-states-data-center-energy-usage-report-2025-update

EIA: https://www.eia.gov/electricity/sales_revenue_price/pdf/table_5A.pdf

## Calculations

Equivalent residential customers = TWh × 1,000,000,000 / 10,356.

| Definition | Calculation, TWh | Million residential-customer equivalents (unrounded calculation) | Display result |
|---|---:|---:|---|
| AI servers only | 50 | 4.8281189649 | ~50 TWh; ~4.8 million |
| AI servers + allocated facility overhead | 50 × 1.145 = 57.25 | 5.5281962148 | ~57 TWh; ~5.5 million |
| All data centers, AI and non-AI | 192 | 18.5399768250 | 192 TWh; ~18.5 million |

Allocated overhead alone = 7.25 TWh, or about 0.70 million residential-customer equivalents. It increases the AI-server-only estimate by 14.5%. Unrounded arithmetic above is for reproducibility and does not imply measurement precision.

## Reading sensitivity and scope

An illustrative 45–55 TWh range around the visual 50 TWh reading gives 51.525–62.975 TWh after PUE, equivalent to approximately 5.0–6.1 million residential customers. This is an analyst-selected ±10% reading sensitivity, not a confidence interval and not the report's modeled uncertainty.

The AI estimate includes the report's AI-specialized server category and allocated cooling/power-distribution/other facility overhead. It does not separately allocate external storage or networking electricity to AI. Those are IT equipment, not overhead supplied by the PUE multiplier. Thus this is a server-based AI operational estimate with facility overhead, not a complete AI lifecycle footprint or a precisely measured total for all AI workloads. Do not add all infrastructure in Figure 9 to AI: it also supports non-AI equipment. The all-sector PUE of 1.45 is not used for the AI category.

The source values are model estimates. Equivalent homes are a scale comparison, not a claim that electricity was diverted from those homes. The chart contains one derived aggregate estimate per definition, with no statistical replication or inferential error bars.
