US Energy

Published

July 7, 2026

LAST UPDATED: 2026/07/07 at 13:46 EDT
AUTHOR: Brennen Slaney 
GITHUB: @realslimslaney

This page tracks how much energy the United States uses, where it comes from, and what it costs, from seventy years of annual history down to what the grid did the day before yesterday. Consumption and prices come live from the EIA; the AI data-center section at the bottom asks how much of all this computing actually uses, with numbers from the DOE-commissioned LBNL report and the IEA. Units: a quad is a quadrillion British thermal units; 1 quad ≈ 293 TWh of thermal-equivalent electricity. Sources are listed at the bottom.

  • EIA: US Energy Information Administration, the Department of Energy’s independent statistics agency
  • DOE: US Department of Energy
  • LBNL: Lawrence Berkeley National Laboratory, a DOE national lab
  • IEA: International Energy Agency, the Paris-based intergovernmental energy body
  • MER: EIA’s Monthly Energy Review, the audited consumption statistics
  • SEDS: EIA’s State Energy Data System, the expenditure statistics
  • EIA-930: EIA’s daily grid monitor for the contiguous Lower 48
  • Btu / quad / GWh / TWh: British thermal unit; a quadrillion Btu; gigawatt-hours; terawatt-hours

How much energy does the US use?

96.3 quads
total primary energy, 2025

77 TWh-eq / day
that’s 0.264 quads every day

13,942 GWh
Lower-48 electricity demand on Jul 06, 2026

$1.51T
spent on energy in 2024 ≈ $4.1B per day

The long view first: total primary energy consumption, every Btu of fuel and renewable generation the country uses in a year, including what power plants burn to make electricity.

Source: EIA Monthly Energy Review Table 1.3 via the EIA API v2 (annual, 1949+). The same year simply divided by 365 in the per-day tab; actual daily variation only exists for electricity (see the grid, day by day).

The monthly view shows the rhythm underneath the annual totals: winter heating and summer cooling put a double hump in every year.

Source: EIA Monthly Energy Review via the EIA API v2 (monthly, shown from 2000; ~2–3 month publication lag).

Where the energy comes from

Five sources cover essentially all US energy. Fossil fuels (petroleum, natural gas, coal) still carry most of the load; the fuel switch of the last two decades has been coal out, natural gas and renewables in.

Source: EIA Monthly Energy Review Table 1.3. Petroleum excludes biofuels (they count under renewables), so the five sources sum to the total.

Renewables deserve their own breakdown. “Renewables” was mostly hydropower and wood for half a century; the growth since ~2008 is wind and solar.

Source: EIA Monthly Energy Review Table 10.1 (annual, 1949+).

The 2025 energy mix
Annual consumption per source, with its per-day equivalent
Source Quads / year Share % TWh-eq / day
Petroleum 35.9 37.3 28.8
Natural gas 34.6 36.0 27.8
Renewables 8.8 9.2 7.1
Coal 8.7 9.0 7.0
Nuclear 8.2 8.5 6.6
Total 96.2 100.0 77.3
Source: EIA Monthly Energy Review Table 1.3. TWh-equivalent uses 1 quad ≈ 293.07 TWh (thermal).

The grid, day by day

Total energy only exists as monthly and annual statistics, but electricity is measured hour by hour. The EIA-930 grid monitor reports what the Lower-48 grid actually did, about a day behind real time.

Source: EIA-930 Hourly Electric Grid Monitor via the EIA API v2. Contiguous Lower 48 only (Alaska and Hawaii excluded); operational data with a 1–2 day lag, subject to revision.

Where each day’s electricity came from. The daily fuel mix makes the sun visible: solar rises and sets inside every week of the band below.

Source: EIA-930 via the EIA API v2, trailing 90 days. Battery and pumped storage are net values and can be negative (charging). Lower 48 only; 1–2 day lag.

The grid on Monday, July 06, 2026
Generation by fuel vs. the same weekday a year earlier (Jul 07, 2025)
Fuel GWh Share % vs. year ago %
Natural gas 5,862 43.5 −7.6
Coal 2,576 19.1 −9.6
Nuclear 2,306 17.1 0.4
Solar 1,170 8.7 23.6
Hydro 658 4.9 −6.3
Wind 629 4.7 9.4
Other 229 1.7 39.1
Petroleum 28 0.2 −26.7
Battery storage 16 0.1 −26.5
Pumped storage 8 0.1 −29.0
Source: EIA-930 via the EIA API v2. Lower 48 only; storage rows are net of charging.

What the US spends on energy

The State Energy Data System tracks what end users actually pay: fuel bought by homes, businesses, industry, and transportation, plus retail electricity bills. Power-plant fuel purchases are excluded (that cost reaches consumers through the electricity line), so the sources below sum to the total without double counting.

Source: EIA State Energy Data System (SEDS) via the EIA API v2, 1970+. Real dollars use the OMB GDP chained price deflator shared with the US Fiscal tab (the deflator is fiscal-year based; SEDS years are calendar years; the small mismatch is accepted for a 50-year chart).

Source: SEDS (spending by source) and EIA retail electricity sales (average price, all sectors, 2001+), via the EIA API v2.

Note

SEDS expenditure estimates lag about 18 months; the latest full year here is 2024. “Petroleum / Natural gas / Coal” are end-use purchases only: fuel bought by power plants is excluded and reaches consumers through the “Electricity (retail)” line instead, so the stack sums to the total without double counting.

What impact do AI data centers have?

Note

Every number in this section is a model-based estimate. No agency meters data centers directly. The US numbers below are LBNL’s bottom-up estimates built from server-shipment data (commissioned by the Department of Energy, published December 2024), and the global context is the IEA’s. The wide 2028 range is by construction, not sloppiness: it spans scenarios of how many GPUs get built and how hard they run. Cryptocurrency mining is tracked separately and is not in these totals (EIA’s 2023 estimate for Bitcoin mining alone is another ~70 TWh).

US data centers consumed an estimated 176 TWh in 2023, about 4.4% of US electricity, roughly tripling over the decade, with GPU-accelerated AI servers accounting for roughly 40 TWh of it. LBNL’s scenarios put 2028 between 325 and 580 TWh (6.7%–12% of US electricity). The US runs about 45% of the world’s data-center electricity (IEA). Whether that is “a lot” depends entirely on the denominator; the ladder chart below walks from all US energy down to AI.

Source: LBNL, 2024 United States Data Center Energy Usage Report. Only values stated in the report text are plotted (hence the sparse markers); lines connect estimates, they are not annual data. The AI-server line excludes the cooling overhead attributable to those servers.

LBNL’s published shares (its denominator is total US electricity consumption) with an independent check dividing the same TWh by EIA retail electricity sales; the two bases differ by a couple percent of themselves, which is exactly the kind of slack these estimates carry.

Each rung as a share of everything above it: electricity retail sales are 14% of all US energy on a thermal-equivalent basis, data centers are 4.5% of electricity but only 0.6% of total US energy, and AI servers alone are 0.15% of total US energy. The thermal-equivalent basis understates electricity’s economic role (about ⅔ of fuel burned for power becomes waste heat), so these shares are the conservative, apples-to-Btu version.

The AI data-center numbers, with their uncertainty
All values are model-based estimates; nobody meters data centers
Estimate TWh Share
US data centers, 2023 (LBNL) 176 4.4% of US electricity
US data centers, 2028 low scenario (LBNL) 325 6.7% of US electricity
US data centers, 2028 high scenario (LBNL) 580 12% of US electricity
World data centers, 2024 (IEA) 415 ≈1.5% of world electricity; US ≈45% of it
World data centers, 2025 (IEA) 485 AI-focused use grew ≈50% in 2025
World data centers, 2030 projection (IEA) 945 ≈ Japan's total consumption today
Sources: LBNL 2024 United States Data Center Energy Usage Report (Dec 2024); IEA Energy and AI (Apr 2025) and IEA data-centre update (Dec 2025). Vendored with page-level citations in data/energy/.

So, do AI data centers move the needle? On electricity, yes: 4.4% of US electricity in 2023 and plausibly 7–12% by 2028 is real load, concentrated in a handful of regions and growing 13–27% a year in LBNL’s 2023–2028 scenarios. On total US energy, not much: data centers are on the order of 1–2% of all US energy use, and AI servers specifically are a fraction of a percent. Both things are true at once, and the honest summary is the range itself: the difference between the 2028 low and high scenarios (325 vs. 580 TWh) is larger than everything data centers consumed in 2023.

Sources & methodology

Data Source Access Coverage Refresh
Primary energy by source EIA Monthly Energy Review (Tables 1.3, 10.1) EIA API v2 total-energy, live at render 1949+ annual, 1973+ monthly every render (24h cache)
Daily grid demand & fuel mix EIA-930 Hourly Electric Grid Monitor EIA API v2 electricity/rto, live Jul 2015+ / Jul 2018+, Lower 48 every render (12h cache)
Energy expenditures EIA State Energy Data System (SEDS) EIA API v2 seds, live 1970+, ~18-month lag every render (24h cache)
Electricity sales & retail price EIA retail sales EIA API v2 electricity/retail-sales, live 2001+ every render (24h cache)
US data-center electricity LBNL 2024 US Data Center Energy Usage Report (DOE) vendored data/energy/ 2014–2023 + 2028 scenarios manual, ~annual
Global data-center context IEA Energy and AI (Apr 2025; Dec 2025 update) vendored data/energy/ 2024–2030 manual, ~annual
  • Units: a quad is 10¹⁵ Btu; 1 quad ≈ 293.07 TWh thermal-equivalent. Where TWh and quads meet on this page, the conversion is thermal, with no adjustment for the ~⅔ conversion loss in thermal generation.
  • Live EIA data needs a free key: set EIA_API_KEY (register here). Without it the page still renders: EIA charts show an unavailability note and the vendored AI section is intact.
  • EIA-930 scope: contiguous Lower 48 only; operational sensor data with a 1–2 day lag, revised after the fact. The annual/monthly MER series are the audited statistics.
  • SEDS spending convention: end-use purchases plus retail electricity; power-plant fuel excluded (no double counting). Expenditures lag ~18 months.
  • Real dollars: OMB GDP chained price deflator, reused from the US Fiscal tab (fiscal-year deflator against calendar-year spending).
  • Data-center numbers are estimates: vendored with page-level citations in data/energy/README.md; only values stated in the source text are used, nothing is read off a chart. Crypto mining is excluded from those totals.

Commentary lives on the blog.