Energy Storage Revolutions

Energy storage as a system service: how many hours of energy a grid can hold, when it dispatches them, what it earns for doing so, and which technologies serve which durations. The subject of this brief is duration and dispatch. It is not chemistry, and the distinction is the whole argument — a cell can win decisively on cost per kilowatt-hour and still leave a grid unable to survive a windless fortnight, because the number of hours a system is built to hold is a separate decision from what is inside it.

A unit convention, stated first because almost every disagreement in this subject is a unit error. Where this brief writes a figure in dollars per kilowatt-hour it means system energy-capacity cost — the installed cost of a complete plant divided by its rated energy — not cell cost, and never quoted without the technology, the duration, the market and the year of the survey attached. Those qualifiers are not decoration. Lithium-ion four-hour systems were surveyed at about $304/kWh globally in BloombergNEF's 2024 cost work and at about $107/kWh in Wood Mackenzie's March 2026 survey of the Chinese market. That is a threefold spread between two defensible numbers, and most of it is date and geography rather than disagreement. An unqualified dollar-per-kilowatt-hour figure in this field carries no information at all.

Where this brief stops. What is inside a cell — lithium iron phosphate against nickel-manganese-cobalt, sodium-ion, solid-state, cell-level cost, cycle life, degradation chemistry — belongs to advanced battery technologies and is not re-argued here. The arbitration rule where the two briefs meet is the unit of account. If the question is what a kilowatt-hour of cell costs and why, it belongs there. If the question is how many hours anybody bought, what the plant cost per kilowatt-hour of system energy, how it was dispatched and what it earned, it belongs here. One consequence of that cut is worth flagging: lithium-ion appears in this brief mostly as a system competitor to compressed air, thermal and flow storage at eight hours and above, which is a role its own advocates did not expect it to be playing.

Established The one-sentence version of what follows. A very fast, very successful and increasingly unprofitable industrialisation of the two-to-four-hour battery is under way, alongside a quiet Chinese pumped-hydro boom that dwarfs it in stored energy; the long-duration problem that determines whether a high-renewables grid survives a three-week windless January remains almost entirely unaddressed, and the technologies nominated to address it are collectively smaller, in 2026, than a single mid-sized gas turbine.

The reason that reading is not the usual one is that the usual one counts gigawatts. Storage does its work in gigawatt-hours, and the two series have diverged. This brief counts energy wherever a source permits it, reports duration with every deployment figure, and states conditions with every price.

Established The revolution is real, it is large, and it is a two-and-three-quarter-hour revolution. BloombergNEF counts 112 GW and 307 GWh of battery storage added worldwide in 2025, up 48%, and forecasts 158 GW and 459 GWh for 2026. Divide the one by the other and the fleet-average duration of 2025 additions is 2.74 hours; the 2026 forecast implies 2.9 hours. The arithmetic is this brief's; the inputs are sourced. What scaled is a diurnal solar-shifting device, and it is a superb one. It is not an energy reserve.

Established Every independent national series corroborates the two-to-three-hour centre of mass, and the four-hour figure that dominates Western commentary is a regional artefact. China, which is 54–60% of world additions, sits at 2.58 hours.

Established Storage above six hours is a rounding error in that build. BloombergNEF describes long-duration storage of six hours or more as minimal in 2025 and expects it to quadruple to 2 GW in 2026 — which implies roughly 0.5 GW in a 112 GW year, about 0.4% of additions. The arithmetic is this brief's; the quadrupling and the 2 GW are the analyst's.

Established Two authoritative counts of the same year disagree and this brief reports both rather than averaging them. The International Energy Agency counts 108 GW added in 2025, up 40%, roughly four-fifths of it utility-scale, with lithium iron phosphate about 90% of deployments; BloombergNEF counts 112 GW. The gap is about 4% and is almost certainly definitional — whether behind-the-meter systems are included, whether the date is commissioning or energisation. Neither is a primary statistic; both are secondary reporting of proprietary datasets, and the intergovernmental figures are themselves sourced from a commercial minerals-intelligence firm, so the two are not fully independent estimates and should not be treated as a triangulation.

Established The incumbent nobody mentions holds roughly nine-tenths of the world's stored electrical energy, and it is having its best decade. Pumped storage hydropower reached 201 GW installed at end-2025 with 11.6–11.7 GW added during the year, its largest single year on record, inside a total hydropower fleet of 1,469 GW. In energy terms the IEA puts global pumped-storage capability at ≈8,500 GWh and states that pumped hydro is over 90% of total global electricity storage capacity; the industry association puts it at up to 9,000 GWh and over 94% of the world's long-duration capacity. Two sources of very different standing converge on 8.5 to 9 TWh. Set that beside the battery build: the entire 2025 global battery installation, 307 GWh, is about 3.5% of the pumped-hydro energy stock, which is to say the incumbent's stored energy is roughly twenty-eight times everything the battery industry installed in its record year.

Frontier And in energy terms pumped hydro is still outgrowing much of the battery build. At the eight-hour-plus durations that independent tracking attributes to pumped storage, 11.6 GW added in 2025 implies roughly 93 GWh of new energy capacity — about 30% as much energy capacity as every battery installed worldwide that year. That is this brief's arithmetic on a sourced power figure and a sourced typical duration, not a sourced energy figure, and it is flagged frontier for that reason. The direction of the finding is robust to a wide range of assumed durations; the precise 30% is not.

Established The long-duration category, after two decades of promise, delivered fifteen gigawatt-hours. Wood Mackenzie's March 2026 assessment puts 2025 global long-duration deployment above 15 GWh, up 49% — and at 6% of all energy storage installations, with 93% of cumulative long-duration capacity in China. The 2025 technology split was compressed air 45%, thermal 33%, vanadium flow 21%, with gravity absent from the split altogether. Read the accompanying duration figures carefully: lithium-ion averages about 2 hours, vanadium flow and compressed air about 4 hours, thermal about 8. Most of what the industry counts as long-duration storage is a four-hour asset.

Established The Western flagships, verified one by one, are demonstrators. Form Energy's entire fleet delivering to a grid is the 1.5 MW / 150 MWh hundred-hour iron-air installation at Cambridge, Minnesota — roughly one twenty-thousandth of one per cent of 2025's global additions — with its 300 MW / 30 GWh Google and Xcel project shipping first modules at the end of 2028. Hydrostor's 1.75 MW Goderich facility in Ontario has been operating since 2019 as a commercial reference plant, seven years, while its 500 MW / 4,000 MWh Willow Rock project in California had its offtake amended three times, the April 2025 amendment slipping commercial operation thirteen months and raising the contract price 15.5%, from US$775 million to US$895 million, with financing still unsecured despite a conditional United States loan guarantee of up to US$1.76 billion. Energy Vault has shifted from gravity to lithium-ion and from engineering contractor to independent power producer, redirecting research to software and artificial-intelligence infrastructure. ESS Inc., the iron-flow manufacturer, reported second-quarter 2025 revenue of US$2.4 million against an operating loss of $11.6 million and US$0.8 million of cash, with its annual filing stating that substantial doubt exists as to its ability to continue as a going concern. Long-duration funding fell 30% in 2025 and its venture capital fell 72%.

Established Where a genuine technology-neutral long-duration auction actually ran, pumped hydro and lithium-ion won it. Ofgem's cap-and-floor scheme in Great Britain set an eight-hour minimum, assessed 77 projects totalling 27 GW in October 2025, and on 26 June 2026 announced sixteen projects totalling 7,645 MW at durations of eight to twenty-two hours. Three pumped hydro schemes — Coire Glas, Earba and Loch Kemp — take the lion's share of capacity, and Field won five lithium-ion battery projects totalling 1,600 MW / 26.8 GWh at an average 16.75 hours. Compressed air, vanadium flow and zinc hybrid cathode won only small allocations. The scheme was technology-neutral only after initial resistance to admitting lithium-ion at all. This is the empirical answer to the question of which technology gets long duration when a regulator pays for it, and it is not the answer the long-duration sector expected.

Frontier Meanwhile the revolution that did happen is eating its own returns. ERCOT battery revenues tracked about $26.0/kW through November 2025 for a full-year projection near $29.4/kW — down 50% from $56/kW in 2024 and 83% from $193/kW in 2023. Great Britain fell from above £150,000/MW/yr in 2022 to about £50,000 in 2023 once frequency services saturated, with October, November and December 2025 benchmarking at £77,000, £59,000 and £47,000 per megawatt-year respectively. CAISO's fleet earned $53/kW-yr in 2024 against $78 in 2023, with 82% of market revenue now coming from energy rather than ancillary services. Flagged frontier not because the revenue series are uncertain — they are measured — but because the inference that the compression is structural rather than cyclical rests on a mechanism argued in the bottlenecks section rather than on a controlled comparison.

Established The honest positive finding, which this brief holds simultaneously with everything above. Short-duration storage demonstrably works. On 29 March 2025 CAISO curtailed over 58,000 MWh of solar while batteries absorbed 38,897 MWh; the analysis concludes curtailment would have been 67% higher without storage. California's battery share of the evening 7–8 pm window rose from 3% to 22% between 2021 and 2025 while fossil generation's share of that window fell from 44% to 34%, and Italian large-scale batteries discharged an average 1.1 GW in the same hours in September 2025, about 3% of demand. The thing works. It is already suppressing the price signal that pays for it.

Four things in this subject are genuinely open. Several more sound open and are not, and separating them is most of the analytical work, because the storage literature routinely presents a solved diurnal problem and an unsolved seasonal one as a single continuum with a cost curve running through it.

Frontier The first genuine unknown is the size of the requirement, and the one rigorous published estimate is sobering. A May 2026 Nature Communications study of European Dunkelflaute — extended wind-and-solar droughts — analysed thirty-five historical weather years and found that under policy-relevant interconnection assumptions the least-cost European system requires 351 TWh of long-duration storage capacity, about 7% of annual European electricity demand. With unconstrained geographic balancing across the continent that falls to 159 TWh, 3.2% of demand. The extreme events it models “may last up to several months and span across the turn of years”, the most severe in the record being winter 1996/97. Set that against the stock: 351 TWh is roughly forty times the entire world's pumped-hydro energy capability of about 9 TWh, and on the order of five hundred times the world's installed battery energy. Even the low-end 159 TWh is about eighteen times world pumped hydro. This is a modelling result, not a measurement, and it is flagged frontier for that reason — but it is peer-reviewed, it spans thirty-five weather years, and no consulted source offers a smaller rigorous estimate.

Frontier The second is the cost threshold, and the peer-reviewed answer is brutal. Sepulveda and Jenkins, in Nature Energy in 2021, mapped the design space and found that long-duration storage energy-capacity cost must fall to ≤US$20/kWh to reduce electricity costs by 10% or more, and to ≤US$1/kWh to fully displace all modelled firm low-carbon generation, with the highest-impact systems having durations exceeding one hundred hours. Energy-capacity cost and discharge efficiency dominate the outcome; power-capacity cost and charge efficiency are secondary. The same paper concludes that in northern-latitude electrified systems, full displacement of firm generation requires performance combinations unlikely to be feasible with known long-duration technologies. A separate 2024 Nature Communications study puts the threshold at which seasonal operation becomes cost-effective below US$5/kWh, at which point modelled optimal duration jumps from about nine hours to about eight hundred hours. Actual delivered system energy-capacity costs across the technologies in this brief run $107 to $643/kWh. That is a gap of roughly five to thirty times against the $20/kWh threshold and one hundred to six hundred times against the $1/kWh threshold, and it has not narrowed for the challenger technologies.

Frontier The third is where the crossover actually sits, and here two commercial surveys are in direct conflict. BloombergNEF's 2024 cost work — 278 data points across seven long-duration technology groups and twenty technology types, explicitly excluding conventional pumped hydro and hydrogen — concluded that the least expensive novel technologies were already cheaper than lithium-ion for durations over eight hours. Wood Mackenzie's March 2026 survey of the Chinese market has thermal storage 78% more expensive and compressed air 88% more expensive than lithium-ion at four hours. The likely resolution is that lithium-ion's system cost fell from about $304/kWh to about $107/kWh in China between the two surveys, moving the crossover outward — plausibly past twelve hours. This brief does not resolve the conflict. The defensible statement is that the crossover is somewhere between eight and twelve-plus hours and has been moving away from the challengers, not toward them, which is precisely why Wood Mackenzie forecasts lithium-ion holding 85% market share through 2034 and describes a strategic squeeze on long-duration technologies in the economically critical four-to-eight-hour band.

Frontier The fourth open question is whether anything anybody built actually delivers, and it is unanswerable from the public record. Every capacity figure for a long-duration demonstrator in this brief is a nameplate. No source consulted publishes megawatt-hours delivered, cycles achieved, or availability for Energy Vault's Rudong gravity plant, Hydrostor's Goderich facility, Form Energy's Cambridge installation, or Antora's Big Stone City project. This is the single largest evidentiary hole in the subject and the brief states it as such rather than filling it with vendor claims. Round-trip efficiency is the specific casualty: Form Energy does not publish one for iron-air, and the only figure available is a third-party estimate of ≈40–50%, possibly nearer 40%; Energy Vault has published no measured efficiency for Rudong anywhere reachable; Antora has not disclosed one for Big Stone.

Frontier What sounds open but is not: whether the intermediate durations have an optimum. They do, and it is modelled with reasonable consistency. Cost-optimal storage duration in a zero-emissions system is 6 to 10 hours in solar-dominant zones and 10 to 20 hours in wind-dominant zones, and a 50% reduction in available hydropower pushes average optimal duration in affected load zones from 6.3 hours to 23 hours. The open question is not what duration is wanted. It is why nobody is buying it.

Handwave What sounds open and is a marketing artefact: gravity storage's round-trip efficiency. Energy Vault's project page claims efficiency exceeding 80% and a thirty-five-year life for Rudong. There is no independent verification, no published measured output, and no efficiency figure in the commissioning announcement itself; the testing was conducted by the investor and builder. The Chinese national pilot-demonstration designation the project carries is a policy label, not a performance audit. Treat the 80% as unevidenced.

Handwave And a specific dataset this brief refuses to use. Systemiq's December 2024 levelised-cost table gives flow batteries at €25/MWh, an order of magnitude below every other source consulted and below lithium-ion in the same table. The figure is implausible and inconsistent with the $423–701/kWh capital costs reported for the same technology elsewhere. The rest of that paper's deployment findings are used; its cost table is not.

Speculative The genuinely speculative frontier, stated as speculation. If any technology reached the $5/kWh energy-capacity threshold, the modelled response is not incremental — optimal duration jumps roughly ninety-fold and a twenty-terawatt-hour storage mandate is modelled to cut high-demand-hour prices by over 70%. That is a discontinuity, not a trend, and no deployed technology is within an order of magnitude of triggering it. The only routes plausibly approaching $5–20/kWh at the margin are pumped hydro with favourable geology and hydrogen in salt caverns — precisely the two categories the commercial cost surveys exclude, which means the cheapest candidates are the least measured.

Established First and binding: the revenue that pays for storage is collapsing, and it is collapsing because storage works. ERCOT battery revenues fell from $193/kW in 2023 to $56/kW in 2024 to a projected $29.4/kW in 2025 — 83% in two years. Two drivers are separable. One is weather: 2025 had only eleven days at a statewide average of 85 °F or above, against 28 to 71 in historical years, 61–85% fewer extreme heat days, and that is cyclical. The other is structural: about 9 GW added in 2024–25 on a fleet up seventyfold since 2020, with the ancillary services market saturated and arbitrage spreads compressed as batteries bid lower. The structural signature is visible in the concentration statistics — the top thirty days produced 30% of annual revenue in 2025 against 44–68% previously, and no single day exceeded 3% of annual revenue against 8–9% historically. A fleet large enough to flatten its own scarcity events has removed the tail it was earning from.

Established Second: ancillary services are a volume-capped market, and every market with a material fleet has saturated or is saturating. A transmission operator needs a fixed few gigawatts of frequency response regardless of how much storage exists. Great Britain reached the wall at the end of 2022 and revenues fell from above £150,000/MW/yr to about £50,000 in a single year. The mechanism has been identified precisely: when the overbuild ratio — available capacity divided by service demand — exceeds about 1.5, frequency prices converge toward battery opportunity cost. Germany is on the same curve and the arithmetic is forecastable: about 580 MW prequalified for automatic frequency restoration reserve, roughly 30% of operational batteries, against 2 GW procured by German transmission operators, with the battery fleet heading to 5.7 GW by end-2026. If even 35% of that fleet prequalifies, supply exceeds procurement.

Established Third: arbitrage is volume-uncapped but self-cannibalising, which is a slower version of the same problem. CAISO's fleet now earns 82% of market revenue from energy rather than ancillary services, having provided 84% of regulation up and regulation down while charging with 14.7% of system load in the late-morning hours. The compression is measurable: the four-hour day-ahead top-to-bottom spread was $4,200/MW in October 2025, down 17.7% year on year, and monthly revenue was $2.99/kW-month, down 22.7% year on year, on a 14.7 GW fleet. Each additional two-to-four-hour asset flattens the very price curve it monetises.

Frontier Fourth: the energy-capacity cost of every long-duration candidate is one to three orders of magnitude above the level at which duration pays. This is the bottleneck that binds the category rather than the incumbent, and it is set out in the frontier section. What makes it a bottleneck rather than merely a gap is the direction of travel — lithium's system cost has been falling faster than the challengers', so the crossover duration has been moving outward, squeezing the four-to-eight-hour band where a challenger would first find a market.

Established Fifth: capital has already drawn the conclusion. Long-duration funding fell 30% in 2025 and venture capital into the category fell 72%, in the same year that batteries added 112 GW. The balance-sheet consequences are visible at company level: ESS Inc. with US$0.8 million of cash and a going-concern warning, rescued by a $31 million injection in July 2025 and pivoted to a twelve-to-fourteen-hour product; Energy Vault redirecting research away from its own flagship technology while retaining about US$75 million of gravity licences it no longer builds against. Capital moved toward the thing that works now and away from the thing a decarbonised grid requires, which is a rational response to the cost gap and a bad outcome for the system.

Established Sixth: permitting, interconnection and financing bind before technology does. Willow Rock's stated reasons for a thirteen-month slip and a 15.5% price increase were continued permitting and interconnection challenges, state permitting delays, the incumbent utility's inability to complete transmission upgrades on time, and continuing failure to secure project financing — all this after a conditional federal loan guarantee of up to US$1.76 billion in January 2025 and a final state licence on 20 December 2025. Nothing in that list is about compressed air.

Frontier Seventh: the entire operating evidence base for long duration sits in one country. 93% of cumulative long-duration capacity is in China. Whatever is learned about degradation, availability, dispatch behaviour and real round-trip efficiency at scale will be learned there first, under a reporting regime that publishes commissioning announcements and not operating data. A Western developer cannot currently point to an independent operating record for its own technology class, and that is a financing bottleneck as much as an epistemic one.

Established What is not a bottleneck, stated plainly. Chemistry is not the bottleneck for flow batteries — they have scaled in China to gigawatt-hour class and they work. Geology is not the binding limit on pumped hydro at current build rates, with 621 GW in development and 243 GW under construction globally. Curtailed energy is not scarce. The bottleneck is that nothing outside pumped hydro delivers energy capacity at a price that justifies holding it for more than a few hours, and no market yet pays anyone for holding it.

Established This brief depends on no unresolved physics, and that is the correct sentence. Every technology described here has been built and connected to a grid somewhere. Pumped hydro is a century old, compressed air runs at 300 MW in Hubei, flow batteries run at gigawatt-hour scale in Xinjiang, thermal storage runs at 5 GWh in South Dakota. What the field waits on is cost, market design and measured operating evidence.

Established It depends on energy-capacity cost falling, which is largely the sibling brief's subject and is not a result the sibling produces. Advanced battery technologies owns cell cost and chemistry; what this brief needs is a price reaching a level, not a discovery being made. Typing a dependency edge from a system brief to a price would be an over-reach in the opposite direction from failing to name the dependency at all, so the constraint is stated in prose and recorded in the tree.

Established It depends on market designs that pay for capacity held rather than energy moved, and those belong to no technology. The revenue stack has migrated twice already — from ancillary services to arbitrage, and now toward administratively set capacity payments: resource adequacy in California, capacity remuneration under China's Document 114, the British cap-and-floor, European capacity markets. A storage fleet whose returns increasingly depend on regulated capacity payments rather than market spreads is not a technology story any more. It is a rate-design story, and rate design is the actual dependency.

Frontier It depends weakly on geology and siting. Salt caverns for compressed air and hydrogen, and topography plus water rights for pumped hydro, are real constraints that no amount of capital removes at a given site. No consulted source identifies them as binding at current build rates, with 621 GW of pumped hydro in development and 243 GW under construction, but they are what makes the cheapest long-duration options non-portable.

Established What depends on it is the more consequential direction. Every high-renewables scenario on this map — and every argument that firm generation can be retired rather than merely displaced — assumes storage at durations the market is not building. That assumption is generic rather than specific to any one route here, so it is named rather than typed. Hydrogen economies and energy corridors are the two substitutes: a molecule stored in a salt dome and a wire to somewhere the wind is blowing both do the job storage would otherwise do, and both compete for the same decarbonisation budget.

Established The most informative experiment in this field was an auction, and it produced a result its designers had resisted allowing. Ofgem's cap-and-floor Window 1 was a technology-neutral competitive procurement with an eight-hour minimum — the only one of its kind with results. It assessed 27 GW across 77 projects and awarded 7,645 MW across sixteen projects at eight to twenty-two hours on 26 June 2026, with final determinations due later in 2026 after a consultation closing 7 August 2026. Three pumped hydro schemes took the bulk of capacity; Field's five lithium-ion projects took 1,600 MW / 26.8 GWh at an average 16.75 hours, described by its chief executive as among the longest duration in the world. Compressed air, vanadium flow and zinc won small allocations. As a test of whether the challenger technologies are competitive at the durations they were designed for, run by a regulator with no stake in the answer, this is as clean as the subject offers.

Established The second is a natural experiment in policy design, and China ran it. Document 136 of February 2025 abolished the mandatory requirement that renewable projects co-locate storage. The result was immediate and directional: by early 2026 standalone systems were 84.7% of capacity additions against 8.4% co-located. Document 114 of January 2026 then extended national capacity remuneration to standalone batteries, and by mid-2026 twelve provinces had published capacity remuneration standards, with a national capacity price floor for standalone grid-scale storage tied to coal benchmarks paying for available capacity rather than discharged energy. Removing a mandate and replacing it with a price produced a larger and better-utilised standalone fleet — cycling rose to 299 from 199 — which is a result any jurisdiction contemplating a storage mandate should weigh.

Established The third is a negative result reported by the company that produced it. Energy Vault built a gravity storage plant, connected it to the grid in December 2023, and then shifted to lithium-ion and an independent-power-producer model, redirecting research to software and artificial-intelligence infrastructure while targeting data-centre power purchase agreements. Its first-quarter 2026 revenue was US$21.9 million, up 156% year on year, with full-year guidance of US$225–300 million, 340 MW operational or in development, a pipeline above 3 GW and a 150 MW Texas lithium project broken ground — growth entirely from the incumbent chemistry. When the flagship technology is set aside by its own developer in favour of the thing it was meant to displace, that is a stronger signal than any modelled comparison. The caveat belongs in the same sentence: the company retains about US$75 million of gravity licences and publicly maintains the technology has been extensively tested and validated.

Established The fourth is a single day of grid operation that settles the diurnal question. On 29 March 2025 CAISO curtailed more than 58,000 MWh of solar while its batteries absorbed 38,897 MWh. The published analysis concludes curtailment would have been 67% higher without storage, and 38% higher without the storage added since 2024. Over January to May 2025 the curtailment rate fell from 13% to 11.5% of solar generation while solar output rose 18%, though absolute curtailment still rose 4.1%. Short-duration storage does the job it is bought for, measurably, at fleet scale.

Frontier The experiment worth watching next is a regulatory instrument, not a device. The enabling innovation behind Antora's 5 GWh Big Stone project is a thermal market energy pricing rider approved by regulators in Minnesota, North Dakota and South Dakota, under which the developer and the utility exchange day-ahead data so that charging tracks surplus renewable output. If long-duration storage becomes viable anywhere in North America this decade, a tariff of that shape is the more likely proximate cause than a cost breakthrough — and the rider is replicable in a way a salt cavern is not.

Frontier And the decisive missing experiment is the one nobody is running: publishing operating data. A single year of measured delivered energy, achieved cycles, availability and round-trip efficiency from Rudong, Goderich, Cambridge or Big Stone would settle more of this subject than any new demonstrator. No such dataset exists for any of them. Until one does, every comparison in the category is nameplate against nameplate.

Established The mechanism that organises this entire subject is that storage capital cost has two terms and they scale differently. Cost decomposes as dollars per kilowatt of power, plus dollars per kilowatt-hour of energy multiplied by hours. At two hours the power term dominates, and lithium-ion's cheap inverters, mature engineering-procurement-construction supply chain and manufacturing scale win comfortably. As hours rise the energy term dominates, and the question becomes what one more hour costs. For lithium-ion, one more hour is one more rack of cells — the cell is the energy store, so cost scales close to linearly with duration. For pumped hydro, compressed air, thermal and hydrogen, one more hour is a slightly larger reservoir, cavern, brick stack or salt dome, and the marginal energy cost is low and sublinear. The peer-reviewed statement is explicit: batteries and pumped hydro have high costs for energy capacity, while the long-duration technologies have high power-capacity costs and low long-duration energy-capacity costs. That is why the challengers should win on paper above some duration, and the only question is where.

Established Here are the two surveys that answer it differently, with their conditions attached. They are not directly comparable and the table says why: different years, different geographic scope, different durations.

Established Round-trip efficiency is the second axis and it is where the compressed-air and iron-air cases are weakest. The measured Chinese compressed-air plants report 70.4% at Zhangjiakou and 64% at Hubei Yingcheng, with up to 70% claimed as achievable. Independent commentary assessing a proposed Ontario project puts advanced adiabatic compressed air at 60–65% against lithium-ion's 85–95% at direct-current level, figures consistent with the measured Chinese plants. Iron-air's third-party estimate is 40–50%, and the consequence is not a modest efficiency penalty but a cycling limit: a hundred-hour system at 40% round-trip efficiency takes 250 hours to charge fully against 143 hours at 70%, capping it at roughly 25 cycles a year instead of 36. That arithmetic is a battery consultant's, reported in March 2026, and it locates the correct place to be sceptical about iron-air. The question is not whether iron rusts reversibly. It is whether an asset that can cycle twenty-five times a year can earn back capital in any market that pays for energy delivered rather than for insurance held.

Handwave One efficiency figure in circulation should never be placed beside the others. Rondo Energy claims round-trip efficiency above 97% for its thermal storage. That is a company claim, and it is meaningful only as electricity-to-heat: the device converts electricity to heat at near-unity and delivers heat. It is not an electricity-to-electricity round trip and comparing it with lithium-ion's 85–95% is a category error, however often it is done.

Established The compressed-air deployment record, since it is the largest long-duration category by 2025 share. Hubei Yingcheng, at 300 MW / 1,500 MWh over five hours, is the largest operating compressed-air plant in the world; it is built in salt caverns, cost CNY 1.95 billion (about US$270 million, roughly $180/kWh), and reached full-capacity operation on 12 January 2025 as confirmed by Chinese government publication. Zhangjiakou is 100 MW / 400 MWh at four hours, a third the size. Jiangsu is 60 MW / 300 MWh at five hours. Sanmenxia in Henan, at 700 MW / 4,200 MWh over six hours, would be far larger, but it was only approved on 24 November 2025 with construction expected to begin in 2026 and must not be described as operating. Against that, the Western record is Hydrostor's Goderich plant in Ontario at 1.75 MW discharge and more than 10 MWh (7 MWh contracted to the provincial system operator), operating since 2019.

Established Thermal storage is the one long-duration class delivering at eight hours and above, and it mostly sells heat rather than electricity. Rondo Energy's operating fleet totals about 132 MWh — a 100 MWh unit at Holmes Western Oil in California in commercial operation since 16 October 2025, storing heat above 1,000 °C and delivering steam alongside existing gas boilers, plus a 32 MWh cement-plant installation in Thailand from November 2025 — with a 100 MWh Covestro project at Brunsbüttel in Germany, €75 million funded, commissioning targeted for end-2026 and supplying about 10% of site steam. Antora Energy's Big Stone City project in South Dakota is 50 MW / 5 GWh of solid-carbon storage running to about 2,400 °C, built in under twelve months and expected to be fully operating later in 2026. At five gigawatt-hours Big Stone is five times the energy capacity of the world's largest flow battery and larger than most national battery fleets by energy — and it delivers process heat for bioethanol production, not electricity to a grid. The most consequential long-duration asset in the West is not a grid asset.

Established And the one place eight-to-sixteen-hour electricity-out storage is being built at scale is barely discussed in Western literature. China connected nine new concentrating solar power plants in 2025 totalling 900 MW, reaching twenty-seven plants and 1,738.2 MW cumulative. The thermal storage durations on the 2025 cohort were 6, 8, 8, 9, 12, 12, 12 and 16 hours, and plants are now being equipped with electric heating so they can function as long-duration storage stations independent of the solar field. The figures come from an industry-adjacent research network; the intended peer-reviewed cross-check could not be obtained, so they stand unchecked against academic literature.

The nearest neighbour is advanced battery technologies, and the relationship is a clean division of the same object. That brief owns the cell: chemistry, cell-level cost, cycle life, degradation. This one owns the system: duration, dispatch, revenue and installed cost per kilowatt-hour of system energy. They share the same technology and answer opposite questions about it, and the arbitration rule is the unit of account. The most useful thing to read across the boundary is that lithium iron phosphate — about 90% of 2025 deployments, and a chemistry story — is what made lithium-ion competitive at sixteen hours in a British auction, which is a system outcome nobody predicted from the chemistry.

Hydrogen economies is adjacent in the strong sense: hydrogen in salt caverns is one of only two routes this brief's sources identify as plausibly approaching the $5–20/kWh energy-capacity cost at which seasonal storage becomes rational, and it was explicitly excluded from the commercial cost survey used here. Any serious treatment of the 351 TWh European requirement runs through that brief, not this one.

Energy corridors is adjacent as a substitute rather than a complement, and the peer-reviewed Dunkelflaute work quantifies the trade: unconstrained geographic balancing cuts the modelled European long-duration requirement from 351 TWh to 159 TWh. Transmission is worth roughly 55% of the storage problem in that model. That is the sharpest available statement of why the two topics belong on the same page.

Small modular reactors and geothermal megaprojects are adjacent as the alternative to the whole enterprise: firm low-carbon generation is what long-duration storage would displace, and the peer-reviewed finding that displacing it entirely needs $1/kWh storage is the strongest argument in the corpus for keeping firm capacity on the table. Planetary-scale energy systems is where the terawatt-hour arithmetic here becomes a design constraint rather than a footnote.

Established The defining institutional fact about storage is that the buyer of duration does not exist in most jurisdictions. Short-duration storage has buyers everywhere — merchant developers selling into energy and frequency markets, utilities meeting resource-adequacy obligations. Duration above about six hours has almost no buyer at all, because no market pays for energy that is held rather than moved. Storage of six hours or more was 0.4% of 2025 additions, and that is a procurement fact rather than a technology fact.

Established One regulator has built the missing institution and it worked. Ofgem's cap-and-floor scheme guarantees a revenue floor and caps the upside for assets meeting an eight-hour minimum, which is exactly the instrument required for capacity that earns rarely. Window 1 drew 27 GW of applications from 77 projects and awarded 7,645 MW across sixteen projects at eight to twenty-two hours, with final determinations due later in 2026. The lesson is not that novel technologies win when you pay for duration — they mostly did not. It is that duration itself gets built when an institution pays for it, and does not otherwise.

Established California authorised the same thing in a stricter form and has not yet run it. The Public Utilities Commission's centralised procurement decision of August 2024 authorises up to 10.6 GW including 1 GW at twelve hours or more and 1 GW multi-day, with the Department of Water Resources as central procurer and lithium-ion explicitly excluded from the long-duration tranches. That exclusion is an industrial-policy judgement, not a technical one, made on the reasoning that alternatives need procurement volume to reach scale economies — and the British result, where excluded lithium-ion won the largest non-hydro share, is the natural test of it. Solicitations were to begin in 2026 for resources online 2031–2037. Nothing has been awarded.

Established China replaced its mandate with a price, which is the most consequential institutional change of the period. Document 136 of February 2025 ended mandatory storage co-location; standalone rose to 84.7% of additions. Document 114 of January 2026 extended national capacity remuneration to standalone batteries, and the national development and energy administrations set a capacity price floor tied to coal benchmarks paying for available capacity rather than discharged energy, with twelve provinces publishing standards by mid-2026. In principle a capacity payment favours longer duration; in practice no minimum duration is specified in the published policy, and China's fleet-average duration remains 2.58 hours. Whether capacity remuneration without a duration floor produces duration is an open institutional question with a large natural experiment already running.

Established One thing in this subject has overshot its official target, and it is the incumbent. China's historic pumped-storage policy targets — over 62 GW by 2025 and about 120 GW by 2030 — now read as conservative against 66.8 GW already in the cumulative fleet and a construction pipeline that an independent tracker puts at 168 GW and the industry association at 218 GW. That 50 GW discrepancy is larger than the entire United States battery fleet, and this brief does not resolve it: the higher figure comes from an interested party, the lower from an independent tracker, and no reconciliation was available in any consulted source. What is not in doubt is the direction. The only storage technology comfortably exceeding its own government's plan is the hundred-year-old one.

Frontier The most interesting institutional innovation in North America is a tariff, not a technology. The thermal market energy pricing rider approved by regulators in Minnesota, North Dakota and South Dakota lets a storage developer and a utility exchange day-ahead information so that charging tracks surplus renewable output. That instrument, not a cost breakthrough, is what made a 5 GWh thermal project financeable. It is also replicable in a way a salt cavern is not, and it points at where the binding institutional work actually is.

Established Who funds, and what changed. Public capital: a $147 million grant and a conditional loan guarantee of up to $1.76 billion in the United States; €75 million from a philanthropic climate fund and the European public investment bank for a German thermal project. Private capital: a $550 million Series C for solid-carbon thermal in August 2026 with a single external investor, against a 72% collapse in long-duration venture capital across the category as a whole. The pattern is concentration — a few large bets alongside a general withdrawal — which is what a sector looks like when investors believe the category is real and the median company is not.

Frontier The institution that does not exist anywhere is a performance registry. No body collects or publishes delivered energy, achieved cycles, availability or measured round-trip efficiency for long-duration assets. Grid operators publish it for their battery fleets — the Californian operator's annual battery report is the model — and nobody does for the demonstrators that public money built. Creating one would be cheap, would settle several of this brief's open questions within a year, and is nobody's mandate.

The evidence base for this subject is dominated by interested parties, and the brief is explicit about which claims rest on whom. Established The pumped-hydro figures come from a hydropower industry association; the United States deployment figures from a solar and storage trade association; the European figures from a solar industry body; the concentrating-solar durations from an industry-adjacent research network; and every long-duration performance claim from the developer that made it. The independent counterweights available are the intergovernmental agency, the United States and Chinese governments, the Californian grid operator, the German regulator, the British regulator, an independent non-governmental infrastructure tracker, an independent think tank, and four peer-reviewed papers. Where those disagree with an interested party, this brief follows the counterweight and says so.

Established Two of the most-quoted datasets in the field are paywalled and cannot be audited. The BloombergNEF and Wood Mackenzie cost surveys underpin most of the comparative economics here, and everything reproduced comes from their own press releases and trade-press relay. Their methodology and definitional boundaries — notably whether behind-the-meter storage is included consistently — are not checkable from outside. That is a structural weakness in the public evidence on a question of considerable public importance, and it is not the analysts' fault; it is what happens when the statistics a policy debate depends on are commercial products.

Established The research-integrity issue in this field is not fraud; it is nameplate reporting. Every long-duration capacity figure in circulation is a rating, and no source publishes megawatt-hours delivered, cycles achieved or availability for any of the Western demonstrators. A sector that has taken public grants, public loan guarantees and ratepayer-funded tariffs on the strength of hundred-hour and eight-hour claims has published no operating evidence that those durations are achieved in service. Publishing measured performance is an obligation that follows public money, and it is not being met.

Frontier Public money in this sector carries risk that is real and under-discussed. A federal loan guarantee of up to US$1.76 billion stands behind a project whose offtake has been amended three times, up 15.5% in price and thirteen months in schedule, and which had still not secured financing. A $147 million grant supports an 85 MW iron-air project with no verified construction progress. Neither is a scandal — early-stage support is exactly what public capital is for, and the alternative is that nothing at this scale is ever attempted. But it should be described as early-stage support rather than as deployment, and the risk should be stated where the announcement is made.

Established A political intrusion into the evidence base worth recording plainly. On 2 October 2025 the United States Department of Energy terminated 315 awards across 223 projects worth $7.56 billion. A federal district court subsequently ruled the terminations unlawful under the Fifth Amendment's equal protection guarantee, the government having conceded that grantees' state political identity played a preponderant role. Whether the Lincoln, Maine iron-air award was among the terminated awards could not be verified and is left open here rather than assumed either way.

Frontier The opportunity-cost question, posed as a question because this brief cannot settle it. Long-duration funding fell 30% and its venture capital fell 72% in 2025, while capital poured into an asset class whose revenues fell 83% in two years. Both movements are individually rational and jointly perverse: money left the thing the system will need and entered the thing the system is already saturating. Whether the correct response is more public support for long duration, or a rate design that pays for held capacity, or an honest decision to retain firm generation instead, is a genuine policy choice, and it is currently being made by default through the absence of a buyer rather than through deliberation.

Established The civilisational question this topic answers is whether a grid run mostly on weather can survive its worst weather, and the honest answer from the evidence is that nobody has built the thing that would let it. A hundred and fifty-eight gigawatts a year of two-and-three-quarter-hour batteries makes solar-heavy grids work beautifully on ordinary days — California's evening window went from 3% battery-supplied to 22% in four years while fossil fell from 44% to 34%, which is a real and rapid decarbonisation of the hardest hours. It does nothing for a three-week windless January, and the modelled requirement for those events in Europe alone is 159–351 TWh against a world pumped-hydro stock of about 9 TWh.

Established The general principle this case illustrates is that a technology can succeed completely and still leave its motivating problem untouched. Storage was argued for as the answer to intermittency. What got built answers intermittency-within-a-day, which turned out to be the tractable half, and the marketing did not distinguish the halves. The corpus's recurring question — what stands between a demonstrated capability and a built world — is answered here not by physics and not even mainly by cost, but by the absence of a buyer for energy held in reserve.

Frontier A second principle, visible in the revenue series, is that a successful grid asset destroys its own revenue. Batteries earn from price spreads and negative prices; batteries at scale remove price spreads and negative prices. California and Texas both saw negative-price frequency decline in 2025, attributed partly to a marked increase in average hourly battery charging around midday, even as seven European countries recorded negative prices in at least 5% of hours. This is not a failure. It is the asset doing its job, and it means the merchant model cannot be the long-run financing mechanism for the thing a decarbonised grid needs most. Any energy system built on weather will end up with a large fraction of its capital earning administratively set payments, which is a political fact before it is an economic one.

Speculative And the largest stake is what happens if the gap never closes. If nothing reaches $20/kWh of energy capacity, the system either keeps firm low-carbon generation — nuclear, geothermal, retained gas with capture — or accepts a reliability standard it has not yet debated in public. The peer-reviewed finding that full displacement of firm generation in northern-latitude systems requires performance combinations unlikely to be feasible with known technologies is the most consequential sentence in this brief's sources, and it is stated as a modelling result rather than a verdict.

Established What already happened, because the chronology usually starts too late. Pumped storage has been the world's dominant electricity store for decades and passed 201 GW in 2025 with its largest annual addition on record. Compressed air reached grid scale in China in 2022 at Zhangjiakou and full-capacity operation at 300 MW / 1,500 MWh at Hubei Yingcheng on 12 January 2025. The world's first gigawatt-hour-scale flow battery — Jimusaer, 200 MW / 1,000 MWh over five hours, about $530/kWh at RMB 3.8 billion — was commissioned on 31 December 2025. Rondo's 100 MWh Californian thermal unit entered commercial operation on 16 October 2025. Ofgem announced its long-duration awards on 26 June 2026.

Frontier 2026: the year the short-duration build accelerates and its revenue keeps falling. BloombergNEF forecasts 158 GW / 459 GWh, implying 2.9 hours; the United States has 24 GW planned after 15 GW added in 2025, with Texas over half of it; Germany's battery fleet heads to 5.7 GW against 2 GW of frequency-reserve procurement. The number to watch is fleet-average duration. If it stays near three hours, nothing structural has changed.

Handwave 2026–2027: the long-duration schedules, all of which are intentions. Antora's Big Stone is expected to be fully operating later in 2026. Rondo's Covestro plant targets commissioning at end-2026 and its Heineken Portugal project April 2027. Hydrostor's Willow Rock groundbreaking was planned for mid-2026 with commercial operation now around July 2029 after a thirteen-month slip. Form Energy's Cambridge project was originally due in service at end-2025 and its own site still described the full project as expected online in 2026. Sanmenxia's 700 MW compressed-air plant expects construction to start in 2026 from a November 2025 approval. Every one of these dates has either slipped once already or belongs to a category whose dates slip; they are recorded as intentions, not forecasts.

Frontier 2028: the first date at which a Western long-duration asset could plausibly matter. Form Energy's chief executive expects to ship first modules for the 300 MW / 30 GWh Google and Xcel project at Pine Island, Minnesota by the end of 2028. That is the earliest credible transition from demonstrator to system-relevant scale in the iron-air programme, and it is two and a half years away from a base of 1.5 MW operating.

Frontier 2031–2037: California's long-duration procurement, if it happens. The state authorised up to 1 GW at twelve hours or more and up to 1 GW of multi-day storage in August 2024, with solicitations beginning in 2026 for resources online between 2031 and 2037. Nothing has been awarded.

Frontier 2034: the forecast that would falsify this brief's reading if it fails. Wood Mackenzie projects lithium-ion holding 85% of the storage market through 2034, with vanadium flow at 5% and compressed air at 3%, and vanadium flow remaining 240% above equivalent lithium-ion cost. If a challenger takes a material share of the four-to-twelve-hour band before then, the sticky-crossover argument here is wrong.

Speculative Beyond: seasonal storage at terawatt-hour scale. The modelled European requirement is 159–351 TWh and no deployed technology is within an order of magnitude of the cost at which it becomes rational to build. A date is not available and this brief declines to supply one.

• Depends on Nothing on this map, and the reason is worth being precise about. Energy-capacity cost per kilowatt-hour is the sibling battery slot's subject, but what this brief depends on is that cost reaching a number — a price, not a result the other brief produces. Typing a brief-to-brief edge to a price point would be an over-reach in the opposite direction from ignoring the dependency. All the technologies described here have been built; none waits on a discovery. Both real constraints are recorded below.
• Requires (not on this map) System energy-capacity cost falling from the $107–643/kWh actually delivered toward the $20/kWh at which peer-reviewed modelling finds a ten per cent system cost reduction — a gap of five to thirty times. Market designs that pay for capacity held for rare events rather than energy moved daily, since arbitrage self-cannibalises and ancillary services are volume-capped. A procurement institution with authority and money at eight hours and above, which currently exists in Great Britain and almost nowhere else. Salt caverns, reservoirs and the water and land rights that go with them, which are what make the cheapest options non-portable. And measured operating data — delivered energy, achieved cycles, availability, real round-trip efficiency — from the demonstrators already built, which no institution collects and no developer publishes. Four of the five are institutional, financial or evidentiary rather than technical, and that is the finding.
• Enables Multi-day and seasonal balancing, which every high-renewables scenario on this map assumes and none of which specifies a route. The assumption is generic rather than specific to one technology, so no typed enabling edge is claimed here — the dependency is real and it runs from a scenario rather than from a brief.
• Adjacent Advanced battery technologies, the sibling slot that owns everything inside the cell; hydrogen economies, the seasonal candidate whose low energy-capacity cost in salt caverns is the one route plausibly near the modelled threshold and whose round-trip penalty is the reason it is not obviously the answer; small modular reactors and geothermal megaprojects, the firm-generation alternatives to storing anything at all; energy corridors, since transmission and storage substitute for one another; and planetary-scale energy systems, where the terawatt-hour numbers in this brief become the constraint.

“Batteries have solved grid storage.” Established They have solved the diurnal shift, and solved it well. The 2025 global fleet averaged 2.74 hours and storage of six hours or more was about 0.4% of additions. A two-and-three-quarter-hour asset moves a midday solar peak into an evening peak. It cannot bridge a multi-week wind drought at any plausible cost, because the terawatt-hours required are three to four orders of magnitude above what diurnal cycling justifies building.

“Two-to-four hours dominates, so call it four.” Established Right conclusion, wrong centre of mass, and the error is systematic. The global median is 2.5–2.7 hours. Four hours is a Californian and American artefact — CAISO's fleet at 3.64 h, new United States build at 3.4 h — driven by resource-adequacy accounting rules that credit a four-hour asset fully. China, which is 54–60% of world additions, sits at 2.58 hours. Writing that four hours dominates overstates the world's installed storage energy by roughly half.

“Pumped hydro is legacy technology.” Established It holds 8,500–9,000 GWh, more than 90% of global storage capacity on the intergovernmental accounting and more than 94% of long-duration capacity on the industry association's, against a global battery build of 307 GWh in 2025. And 2025 was its largest year on record at 11.6 GW added, which at its typical eight-hour-plus duration is roughly 30% as much new energy capacity as every battery installed worldwide. The revolution narrative counts gigawatts and ignores gigawatt-hours, which is precisely the axis on which a storage system does its work.

“California has an eight-hour storage procurement.” Established The threshold and the jurisdiction are both wrong, and the error is common enough to be worth correcting in detail. California's August 2024 decision under AB 1373 authorises up to 10.6 GW in total, including up to 1 GW at twelve hours or more and up to 1 GW of multi-day storage, procured centrally by the Department of Water Resources, with lithium-ion deliberately excluded from the long-duration tranches on the reasoning that alternatives lack the procurement volume to achieve scale cost reductions. Solicitations were to begin in 2026 for resources online 2031–2037, and nothing has been awarded. The eight-hour minimum belongs to Ofgem in Great Britain, and that one has real awards.

“China mandates storage.” Established China went the other way. Document 136 of February 2025 abolished the mandatory co-location requirement for renewable projects, after which standalone systems rose to 84.7% of capacity additions against 8.4% co-located by early 2026. Current policy is capacity remuneration — Document 114 of January 2026, a national capacity price floor tied to coal benchmarks, twelve provinces with published standards by mid-2026 — not duration mandates, and no minimum duration is specified in the published policy. If there is a Chinese storage-mandate story, it is a story about a mandate being removed and replaced with a price.

“China's Dalian flow battery is 175 MW / 700 MWh.” Established That conflates two projects and one of them is in a different province. Dalian, Liaoning is 100 MW / 400 MWh — phase one, grid-connected in late May 2022, of a planned 200 MW / 800 MWh whose phase-two completion this brief could not verify. The 175 MW / 700 MWh four-hour vanadium system is Ushi in Xinjiang, completed December 2024. The current world's largest flow project is Jimusaer, Xinjiang: 200 MW / 1,000 MWh over five hours, commissioned 31 December 2025 at about $530/kWh.

“Zhangjiakou, or a Shandong plant, is the world's largest compressed-air storage.” Established Zhangjiakou is 100 MW / 400 MWh at 70.4% round-trip efficiency, respectable and a third the size of the leader. The largest operating compressed-air plant in the world is Hubei Yingcheng at 300 MW / 1,500 MWh over five hours, 64% round-trip efficiency, about US$270 million or $180/kWh, in salt caverns, at full-capacity operation since 12 January 2025. No Shandong compressed-air plant at comparable scale could be found; the large Shandong facility usually cited in this context appears to be a pumped storage station, though this brief has not proven that negative. Sanmenxia in Henan at 700 MW / 4,200 MWh will be larger and was approved on 24 November 2025 with construction expected from 2026 — it is not operating and should not be described as though it were.

“Flow batteries have failed to scale after forty years.” Established Reframe this one. Flow has scaled, in China, to gigawatt-hour class, and it works. What it has not done is scale outside China — the largest United States project is 2 MW / 8 MWh, and Canadian, British and Australian projects sit around 20 MWh, a factor of fifty to a hundred adrift — or reach durations above about five hours. Its constraint is cost, not chemistry: $423/kWh in China and $701/kWh outside it at eight hours, against lithium-ion at $107–304/kWh depending on market and year. And note where the competition actually happens: flow's commercial durations are four to five hours, so flow batteries are losing in lithium's home territory rather than failing above it.

“Energy Vault abandoned gravity storage.” Established Substantially yes, with one caveat and one sharper point. The caveat: it retains about US$75 million of gravity licences across China, Africa and the Middle East and publicly maintains the technology has been extensively tested and validated. The sharper point: Rudong is a 25 MW / 100 MWh four-hour system. The flagship gravity asset was never a long-duration asset, its claimed round-trip efficiency above 80% has no independent verification, and its own project page still described it as commissioning three years after grid interconnection. Wood Mackenzie's 2025 long-duration technology split does not list gravity at all, and BloombergNEF's 2024 survey put it at $643/kWh, the most expensive of seven categories.

“Iron-air is delivering hundred-hour storage to the grid.” Frontier It is delivering, and the fleet is 1.5 MW. Form Energy stated in March 2026 that it had shipped, installed and operated its first commercial iron-air system in the fourth quarter of 2025 and validated more than one hundred hours of continuous discharge under sub-zero conditions. That is a company claim with no independent measurement, no published efficiency and no energy-delivered data. The Lincoln, Maine project — 85 MW / 8,500 MWh with a $147 million federal grant awarded in August 2024 — has no verified 2025–26 construction progress. The flagship is now the 300 MW / 30 GWh Google and Xcel project with first modules shipping at the end of 2028. And the substantive objection is efficiency: a third-party estimate of 40–50% round-trip implies roughly twenty-five cycles a year rather than thirty-six.

“Curtailment proves we need long-duration storage.” Frontier This is the single most common error in storage advocacy and the inference simply does not hold. California's entire 2024 curtailment was 3.4 TWh, 93% of it solar and the bulk of it in spring midday hours. That is a high-frequency, low-energy-per-event, diurnal problem, and the correct response is more two-to-four-hour assets cycling near-daily — which is exactly what is being built and demonstrably works, as the 38,897 MWh absorbed on a single March day shows. The long-duration case rests on multi-week renewable droughts, events in which there is no surplus to curtail at all. Curtailment statistics and the long-duration argument describe opposite failure modes, and using one to justify the other is a mistake of kind, not of degree.

“Germany curtailed X terawatt-hours in 2025.” Established Specify which X you mean, because two credible figures differ by a factor of about 5.5 and both are right about different things. An independent think tank puts German wind and solar curtailment at about 9.6 TWh, roughly 4% of renewable output — total curtailment including grid-congestion redispatch and feed-in management. A market analytics firm puts it at 1,749.7 GWh, a record and about 25% above 2024 — but that series is explicitly price-sensitive curtailment, volumes cut because prices went negative. The distinction matters for storage specifically: only the price-sensitive portion is directly addressable by arbitrage-driven storage, and the congestion portion is a transmission problem that storage helps only if sited correctly.

“China's rising curtailment shows the grid is overwhelmed.” Established Partly, and partly it is administrative. Solar curtailment rose to 6.6% in the first half of 2025 from 3.9% a year earlier and wind to 5.7% from 3.0%, against 268 GW of new wind and solar added in that half-year alone, with provincial extremes in Tibet of 30.2% for wind and 33.9% for solar. But the national curtailment limit was relaxed from 5% to 10% in the same period, so much of the reported increase is policy headroom being used rather than a purely physical constraint. Read the Chinese curtailment series as partly a regulatory series.

“Storage is a good business.” Frontier Revenues in the largest merchant market fell 83% in two years, British revenues fell roughly two-thirds in one year once frequency services saturated, and the mechanism is not weather but arithmetic: ancillary markets are volume-capped and arbitrage is self-cannibalising. The returns are migrating to administratively set capacity payments in every market with a material fleet. That may be a perfectly good business. It is a regulated one, and it should be described that way.