⚡ Energy
California's Main Grid Ran 36% on Batteries at Its Evening Peak. The Whole US Fleet Holds 20 Minutes.
The US installed a record 20.2 gigawatt-hours of batteries in the second quarter, into a fleet that holds 20 minutes of national demand and whose California slice, in July, covered a third of that state's main grid at its peak hour.
Twenty minutes. That is how much of the country's electricity the entire US utility-scale battery fleet stores: 165 gigawatt-hours against generation that averaged 506 gigawatts last year. Nobody builds a battery to run a country; they build it to run an evening, and the fleet's CAISO slice, on its record evening, covered a third of peak-hour demand on that grid. On July 9 CAISO's batteries discharged 12.99 gigawatts at the evening peak, 36 percent of everything the grid operator was serving then.
A record is the occasion for the arithmetic. On September 1 the Solar Energy Industries Association and Benchmark Mineral Intelligence reported that the country installed 20.2 gigawatt-hours of storage in the second quarter, a record, with the year's forecast at 71 gigawatt-hours. About 18 of those gigawatt-hours were utility-scale; the 165-gigawatt-hour fleet counts utility-scale only. What none of the coverage we could reach supplied was a denominator.
Twenty Minutes Is the Wrong Denominator
Start with the honest small number: in 2025 the country generated 4,430 terawatt-hours, an average of 506 gigawatts around the clock. Divide 165 gigawatt-hours by the average and the fleet holds 19.6 minutes of the country; divide by the Lower 48's record peak of 759 gigawatts and it holds 13. SEIA's 683 gigawatt-hours for 2030 would be 81 minutes at last year's average. A rounding error, by that measure.
The job has a different denominator: a grid's hard hour comes after sunset, when solar drops off a cliff and air conditioners do not. CAISO's fleet held 62 gigawatt-hours in June, and at July 9's 12.99-gigawatt record rate that June fleet would run for nearly five hours on paper. Those same 62 gigawatt-hours are seven minutes of the country, 62 divided by 506. Same battery, different question.
| Quantity | Energy or power | Measured against | What it covers | Type |
|---|---|---|---|---|
| US utility-scale fleet, mid-2026 | 165 GWh | 506 GW average US generation, 2025 | 19.6 minutes, if deliverable | Estimate (SEIA) over measured (EIA) |
| Same fleet | 165 GWh | 759 GW record Lower-48 peak, July 2025 | 13 minutes, if deliverable | Estimate (SEIA) over measured (EIA) |
| SEIA 2030 forecast, all segments | 683 GWh | 506 GW average US generation, 2025 | 81 minutes, if deliverable | Forecast (SEIA) over measured (EIA) |
| CAISO fleet, June 2026 | 62 GWh | 12.99 GW record discharge, July 9, 2026 | 4.8 hours at the record rate, a ceiling | Measured (CAISO) over measured (CAISO) |
| CAISO record hour, July 9, 2026 | 12.99 GW discharged | ~36 GW CAISO demand that hour | 36% of the peak hour | Measured (CAISO) over implied |
| Arizona, Q2 2026 additions | 6.2 GWh, ~1.55 GW at 4 hours | 17.6 GW combined APS and SRP record peaks | 9% of two utilities' peak, a ceiling | Measured (SEIA) over measured (APS, SRP) |
| Daily average build, Q2 2026 | 222 MWh, 55.5 MW at 4 hours | $219 per kWh, BNEF US turnkey | $48.6 million a day | Measured (SEIA) at survey price (BNEF) |
The CAISO-fleet and Arizona rows are ceilings; the 36 GW is implied by the share; assumptions in Limitations.
The Evening Is the Job
Batteries compete with the peaker plant, not the grid's total, and that is the whole case. A gas peaker runs only when demand spikes; a four-hour battery charged on midday solar does the same job with a different fuel. CAISO's batteries set discharge records in June and July, the June one at 8:10 p.m.
Duration is where this gets tested. Dividing SEIA's energy by the Energy Information Administration's 52 gigawatts of power gives the national fleet about 3.2 hours, a little short of a summer evening, and the four-hour system is the benchmark in BNEF's survey, though the survey does not say what gets built.
What a Quarter Buys
Over 91 days, 20.2 gigawatt-hours is 222 megawatt-hours a day, a 55.5-megawatt four-hour battery daily, and Arizona, which installed 6.2 gigawatt-hours in three months, shows the shape of it: its two big utilities set their all-time peaks between 5 and 6 p.m., and one quarter's batteries can cover at best 9 percent of those two peaks combined for four hours. The EIA counts about 52 gigawatts by mid-2026, and developers plan 24 for the year, 28 percent of the 86 gigawatts of utility-scale capacity planned for 2026.
The Price of Building Here
This record was set in the most expensive market BNEF reported. BloombergNEF's 2025 cost survey puts a turnkey storage system at $219 per kilowatt-hour in the United States and $73 in China, against a global average of $117. Three times China's price. This year's 71 gigawatt-hours cost about $15.5 billion at the US price and $5.2 billion at China's, a $10 billion gap that is a benchmark against China's domestic price, not a saving US buyers could book.
Trade policy explains part of that premium, with labor, permitting, interconnection and margin carrying the rest, and BNEF's $219 predates January's tariff increase, when the Section 301 rate on non-EV lithium-ion batteries rose from 7.5 to 25 percent and foreign-entity rules began withholding the 45X manufacturing and 48E investment credits above the foreign-content thresholds. Both measures aim at a domestic cell supply chain, and this piece does not judge whether they work.
China is also the scale comparison. It added 189 gigawatt-hours of new-type storage in 2025, 3.2 times America's 59, and reaching SEIA's 683 gigawatt-hours by 2030 means adding roughly 480 more on top of the 205 the fleet should reach by year-end, which is $105 billion at US prices or $56 billion at the global average.
What You Can Do
Reading a storage headline: convert gigawatt-hours to hours by dividing by the load that matters, a state's evening peak. Anyone modelling 2030 should carry both prices, $219 and $117; the $49 billion between them on 480 gigawatt-hours is a US-cost line item of which trade policy is only part.
Limitations
All figures come from SEIA and Benchmark's report, EIA releases, CAISO and utility records via secondary citations, and BloombergNEF's survey; primary pages were unreachable. SEIA's 165 gigawatt-hour figure is a trade group's estimate, its release crediting "the first 18 months of the Trump administration"; we use the number, not the framing.
The 20.2, 71 and 683 gigawatt-hour figures cover all segments while the 165 gigawatt-hour fleet is utility-scale only; EIA's 52 gigawatts is a separate count, so the 3.2-hour duration is approximate; the minutes-of-the-country figures use 2025 generation, the latest full year, and assume a 52-gigawatt fleet delivering at 506. The 36 percent is CAISO demand, not all California load, with its peak-hour load implied from the share; the 4.8-hour figure assumes a fully charged fleet discharging to empty; and we computed record hours only, not an average evening. The Arizona 9 percent puts statewide four-hour additions over two utilities' peaks, different years; dollar figures hold BNEF's 2025 turnkey averages flat over all-segment volumes; the 205 base adds 2026's remaining 40 to a utility-scale 165; and China's total is CNESA's "new-type" storage, some of it non-lithium.
The Strongest Case Against This Analysis
Nobody serious ever measured batteries in minutes of the country, so the piece knocks down a framing of its own making, and the real questions are whether four-hour batteries cover multi-day heat waves and winter dark, which they do not, and whether the record can continue under tariffs and foreign-entity rules added to a price already three times China's. CAISO's 36 percent was one ISO on one evening, not a fleet average, Arizona's 9 percent assumes every battery is full at 5 p.m., and the 2030 forecast comes from the industry that sells the batteries.
That first objection is fair: the national number is what a reader reaches for when a gigawatt-hour total lands with no denominator attached, and the article's job is to replace it with the evening-peak number. Multi-day weather is a real limit, hence the duration figure; one evening is the right unit for a record, the wrong one for an average. Arizona's 9 percent is a ceiling; a fleet 80 percent charged covers about 7. Wood Mackenzie, which sells research and not batteries, projected in June 655 gigawatt-hours by 2031, below SEIA's 683 and a year later. If prices hold near $219, that forecast is the first thing that breaks.
The Bottom Line
America's grid batteries hold 20 minutes of the country. California's 62 gigawatt-hours hold nearly five hours of a CAISO evening at the record rate on paper, and that is the number that describes the job. That much is settled. The 2030 forecast turns on whether the country keeps paying $219 per kilowatt-hour against a $117 global average.