Environment and energy: what the evidence shows

Evidence on environment and energy in the United States shows a power system that in 2025 still drew about 58% of its utility-scale electricity from fossil fuels, record domestic energy production, net greenhouse gas emissions in 2022 that EPA put 17% below 2005 levels, and retail electricity prices that rose in nominal terms from 2023 through 2025. The figures most contested in public debate are not those measurements but estimates built on assumptions: what each power source really costs, how reliable the grid will be as demand from data centers grows, what a ton of emissions is worth in damages, and how long permitting takes. Several of these estimates moved sharply between 2024 and 2026, and some core federal data series are published on a delay.

In plain words

Environment and energy numbers come in two kinds.

Some energy numbers are counts. How much electricity came from gas last year is a count, and it is pretty reliable.

Other numbers are forecasts or estimates. How much a new solar farm will cost over 30 years, or whether the grid will have enough power in 2030, depends on guesses about the future. Different experts make different guesses, and they get different answers.

When you see an energy number in an argument, ask three things. What year is it from? Is it counting something that happened, or predicting something? What does it leave out?

Defined terms

Environment and energy data on this page use these terms.

• Utility-scale generation. Electricity from power plants of at least 1 megawatt of capacity; small rooftop systems are counted separately by EIA.
• Quad. One quadrillion British thermal units, the unit EIA uses for total U.S. energy production and consumption.
• Gross and net emissions. Gross emissions are total releases; net emissions subtract carbon absorbed by forests and land (the land sector’s “sink”).
• Levelized cost of electricity (LCOE). An estimate of the cost to build and operate a generator over its cost recovery period, expressed per megawatthour. Levelized avoided cost of electricity (LACE) estimates the revenue, and so roughly the value to the grid, that the generator would earn.
• Loss of load. Hours or events in which supply cannot meet demand, a standard way to measure reliability risk.
• Interconnection queue. The list of proposed power plants and storage projects waiting for approval to connect to the transmission grid.
• Social cost of greenhouse gases. “A monetary estimate of the economic impacts associated with emitting an additional (marginal) ton” of a greenhouse gas, in the Congressional Research Service’s definition.⁠¹⁸
• Discount rate. The rate used to convert future costs and benefits into present values; a lower rate gives more weight to the future.

Evidence

Evidence on environment and energy comes from agencies and organizations with different jobs, and the differences matter. The Energy Information Administration (EIA) is the federal statistical agency for energy and publishes measured data and model projections. EPA compiles the national greenhouse gas inventory. NOAA measures temperature. The North American Electric Reliability Corporation (NERC) assesses grid reliability using utility forecasts. National laboratories, such as Lawrence Berkeley National Laboratory, publish research. Agencies also publish analyses that support a policy, and advocacy groups publish analyses that oppose one. This page labels which is which.

Two cautions apply throughout. First, vintages differ: EIA’s 2025 generation data are already out, while EPA’s inventory page still links 2022 as its latest complete year.⁠⁵ Second, measurement and projection are different kinds of claims; a forecast can be reasonable and still wrong.

What the data show

Environment and energy data from federal agencies and independent researchers show the following, each with its year and source.

Electricity generation. In 2025, total U.S. utility-scale generation was about 4.43 trillion kilowatthours. Natural gas supplied about 41%, coal about 17%, nuclear about 18%, wind about 11%, solar about 7% and conventional hydropower about 6%, for about 58% from fossil fuels and about 24% from renewables.⁠¹ EIA separately estimated about 0.09 trillion kilowatthours from small-scale solar under 1 megawatt.⁠¹

Energy production and consumption. U.S. energy production reached a record 107 quads in 2025, up 3.4% from the previous record in 2024 and the fourth consecutive annual record.⁠³ Crude oil production averaged a record 13.6 million barrels per day, and dry natural gas production rose more than 4% to a record 39 trillion cubic feet.⁠³ EIA reports that the United States remained the world’s largest crude oil producer in 2025, about 40% above the next two largest producers, Russia and Saudi Arabia.⁠⁴ EIA also states that “since 2019 total energy production has been greater than consumption in the United States”.⁠² On the consumption side in 2025, petroleum was about 38% of U.S. primary energy consumption, natural gas about 36%, renewables about 9% and coal about 9%.⁠²

Greenhouse gas emissions. EPA reports that U.S. greenhouse gas emissions in 2022 totaled 6,343 million metric tons of carbon dioxide equivalent, and that 2022 emissions after accounting for land sector sequestration were 17% below 2005 levels; the land sector offset 13% of emissions, leaving net emissions of 5,489 MMT CO2e.⁠⁵

Temperature. NOAA reported that 2024 was the warmest year in its record beginning in 1850, at 1.29°C (2.32°F) above the 20th-century average and 1.46°C above the 1850 to 1900 pre-industrial average, and that the 10 warmest years since 1850 all occurred in the past decade.⁠⁶

Prices. EIA’s Electric Power Monthly reports average retail prices, in nominal cents per kilowatthour, of 12.68 (all sectors) and 16.00 (residential) in 2023; 12.94 and 16.48 in 2024; and 13.56 and 17.23 in 2025. Preliminary data for August 2026 show 14.87 and 18.16, against 14.15 and 17.57 in August 2025.⁠⁷ By simple arithmetic, the residential annual average rose about 7.7% in nominal terms from 2023 to 2025; these figures are not adjusted for inflation. Prices vary widely by state: in EIA’s preliminary 2025 data, Hawaii averaged 35.72 cents and North Dakota 8.20 cents per kilowatthour.⁠⁸

Demand. A Lawrence Berkeley National Laboratory report released by the Department of Energy in December 2024 estimated that data centers consumed about 4.4% of total U.S. electricity in 2023 and could consume about 6.7% to 12% by 2028.⁠¹¹ NERC’s January 2026 assessment forecast summer peak demand rising by over 224 gigawatts over 10 years, 69% more than its prior projection of 132 gigawatts, with new data centers accounting for most of the increase.⁠¹²

Where the data are disputed

Environment and energy disputes are often presented as disputes about values. Many are also disputes about which number to trust and what it includes.

What power sources cost. EIA’s April 2025 estimates of levelized cost for plants entering service in 2030, in 2024 dollars per megawatthour and including tax credits as then modeled, put the simple averages at $29.58 for onshore wind, $31.86 for solar, $64.55 for natural gas combined-cycle, and $81.45 for advanced nuclear.⁠⁹ Advocates of renewables cite numbers like these as proof that wind and solar are cheapest. EIA itself warns that “direct comparisons of LCOE or LCOS across technologies are misleading as a method to assess the economic competitiveness,” because LCOE “does not capture all factors that contribute to investment decisions,” including value to the system and reliability.⁠⁹ Critics of relying on wind and solar argue that the costs of backup, storage and transmission needed to make variable output reliable belong in the comparison. The tax inputs have also changed: under legislation enacted July 4, 2025, the main clean electricity credits end for wind and solar facilities placed in service after December 31, 2027, unless construction began by July 4, 2026.⁠¹⁰ Estimates prepared before that change cannot simply be reused.

Whether the grid will be reliable. NERC found that 13 of 23 assessment areas face resource adequacy challenges over the next decade.⁠¹² A July 2025 Department of Energy report went further, stating that “retirements increase the risk of power outages by 100 times in 2030” and that outage hours “could increase from single digits today to more than 800 hours per year”; it counted 104 gigawatts of firm generation slated to retire by 2030 and 209 gigawatts of planned additions, of which 22 gigawatts were firm baseload.⁠¹³ DOE said the methodology informs possible use of its emergency authority to keep plants running.⁠¹³ In a September 8, 2025 rehearing request on one such order, a group of environmental and public interest organizations including the Sierra Club, the Natural Resources Defense Council and the Environmental Defense Fund argued that the report “identifies no present or imminent emergency” and, using what they called “deeply flawed methodology,” at most “a theoretical shortfall of generation in 2030,” and that DOE relied on the most conservative assumptions of a regional survey while ignoring the rest of its range.⁠¹⁴ Both sides agree demand is rising; they disagree about how fast replacements can be built and how to count them.

How fast new supply can connect. Lawrence Berkeley National Laboratory’s May 2026 edition of Queued Up reported about 1,312 gigawatts of generation and about 749 gigawatts of storage in active interconnection queues at the end of 2025, a median of over five years from interconnection request to commercial operation for projects built in 2025 in regions with data, and that only 13% of capacity that requested interconnection from 2000 to 2020 had reached commercial operation, while 75% had withdrawn.⁠¹⁵ A long queue can be read as evidence of strong demand to build, of a broken approval process, or of speculative applications; the data are consistent with all three.

How long permitting takes. The Council on Environmental Quality’s June 2020 report, covering 1,276 environmental impact statements with final statements published from 2010 to 2018, found a mean of 4.5 years and a median of 3.5 years from notice of intent to record of decision, with one quarter taking more than 6.0 years.⁠¹⁶ A January 2025 CEQ release, using a different endpoint (notice of intent to final statement), reported a median of 2.2 years for statements issued in 2024 and 3.1 years for 2017 to 2020.⁠¹⁷ The same release noted that environmental impact statements are “roughly one percent of environmental analyses” under NEPA.⁠¹⁷ Because the two reports measure different intervals, the decline cannot be read directly from them, and both sides of the permitting debate can cite a CEQ number that suits them.

The value of avoided emissions. The Congressional Research Service reports that EPA’s 2023 approach, using discount rates of 1.5% to 2.5%, produced social cost of carbon estimates for 2030 emissions of $140 to $380 per metric ton in 2020 dollars, compared with an interagency working group’s 2021 estimate of $89 for 2030 at a 2.5% rate.⁠¹⁸ CRS notes that “the choice of discount rate can significantly increase or decrease values”.⁠¹⁸ Executive Order 14154 of January 20, 2025 withdrew the working group’s documents “as no longer representative of governmental policy,” directed EPA to address the estimate’s “harmful and detrimental inadequacies,” including “consideration of eliminating” it, and directed agencies to follow the 2003 version of OMB Circular A-4 on discount rates and domestic versus international effects.⁠¹⁸ Supporters of a high value argue that future and global damages are real and that low discount rates are ethically appropriate for long-lived harms; critics argue that the models’ damage functions are speculative and that U.S. rules should count primarily U.S. costs and benefits.

Unequal impacts. EPA’s 2021 report modeled, at 2°C of global warming, that Black and African American individuals are 40% more likely than others to live in areas with the highest projected increases in extreme-temperature mortality and 34% more likely to live where childhood asthma diagnoses are projected to rise most from climate-driven particulate pollution, and that Hispanic and Latino individuals are 43% more likely to live where labor hour losses in weather-exposed industries are projected to be highest.⁠¹⁹ These are modeled relative likelihoods for a warming scenario, not measured outcomes; the report analyzed each group independently and did not analyze interactions among its measures of vulnerability.⁠¹⁹ The report’s former EPA landing page no longer resolves, though the PDF remains at its EPA address.

Local opposition and canceled projects. The Sabin Center found that about 40% of 567 renewable projects facing significant local opposition had been canceled, but cautioned that this “may not be solely due to local opposition”.⁠²⁰ Opposition is real; its causal weight in cancellations is not established by this count.

Science and law are different disputes. Gallup found in March 2026 that 64% of U.S. adults attribute global warming mainly to pollution from human activities, including 90% of Democrats and 28% of Republicans.⁠²¹ EPA’s 2026 rescission of the vehicle endangerment finding rested on what the agency called the “best reading” of the Clean Air Act’s text.⁠²² A reader should not assume that a legal argument about statutory authority is a scientific claim about warming, or the reverse.

What is unknown

Several facts that matter to environment and energy decisions in the United States are not knowable yet, or not available in published form.

• Recent national emissions. EPA’s inventory page, last updated June 10, 2026, still links the 1990 to 2022 report as its latest inventory; a draft covering 2023 was released for comment with comments due March 17, 2025, and the page does not say whether it was finalized.⁠⁵
• Future demand. LBNL’s 2028 data center range runs from 6.7% to 12% of U.S. electricity,⁠¹¹ and NERC’s 10-year growth projection rose 69% in one year.⁠¹² The spread is itself evidence of uncertainty.
• Effects of policy changes since 2025. The 2025 tax law changes,⁠¹⁰ the 2026 vehicle rule rescission,⁠²² and pending litigation (see who decides) have not yet produced multi-year data on prices, emissions or investment.
• Price causation. Retail prices rose from 2023 through 2025.⁠⁷ EIA lists generation costs, fuel prices, transmission and distribution costs, weather and demand among the factors that affect prices.⁠⁸ The sources reviewed here do not apportion the recent rise among those factors or among specific policies.
• The size of future climate damages. The range of social cost estimates, and its dependence on discount rates,⁠¹⁸ reflects uncertainty about damages and about values that additional data will narrow only partly.

Competing interpretations

The same environment and energy evidence supports different stories, each held by serious people.

A transition underway. One reading emphasizes that renewables reached about 24% of utility-scale generation in 2025,⁠¹ that net emissions in 2022 were 17% below 2005,⁠⁵ and that wind and solar have low levelized costs.⁠⁹ On this reading, policy should keep removing barriers such as interconnection delays.⁠¹⁵

An abundance story. Another reading emphasizes record production,⁠³ the U.S. position as the largest crude oil producer,⁠⁴ and the 58% fossil share of generation.⁠¹ On this reading, abundant domestic fuel is an economic and strategic asset, and policy should not force the pace of change.

A reliability warning. A third reading starts from demand growth.⁠^11,12 and argues that whichever sources are preferred, the grid needs more firm capacity soon, and that retirements should slow⁠¹³

A process failure. A fourth reading, shared by some on each side, argues the central problem is that building anything (transmission, pipelines, plants) takes too long,⁠^15,16 and that disputes about fuels distract from it.

The subject overview sets out the value arguments behind these readings.

Application to this framework

My Vote Forward’s framework treats evidence and priorities as separate inputs. Environment and energy shows why.

1. Evidence does not dictate priorities. Two people can accept every number on this page and rank the subject differently, because they weigh cost, risk and the future differently. See tradeoffs.
2. Proposals should carry their evidence and its vintage. A proposal citing an LCOE figure should say which year, whether tax credits were included, and what EIA says the figure omits.⁠⁹ See proposals.
3. Published results need the same discipline. A participation result is a measurement too; it needs a denominator, method and limits. See measurement and question design.
4. Participants’ opinions are not evidence about the world. A majority opinion that prices rose because of a particular policy is a fact about the participants, not about prices. See opinion and choice.

Unresolved questions

Environment and energy evidence leaves these questions open.

• Which reliability forecast should a voter or a representative trust when NERC, DOE and outside critics read the same system differently?
• Should levelized cost comparisons in public debate always be paired with value or system cost measures, and who would enforce that?
• What discount rate, if any, should public decisions use for harms that fall decades away, and is that a technical question or a moral one?
• How long will the gap between measured emissions and published federal inventories last, and what should fill it?
• How can modeled estimates of unequal impacts be checked against outcomes as they occur?