The Long-Duration Storage Playbook: Which Battery Wins Above Eight Hours

In 2025 the world installed more grid batteries than in any year before it: 112 GW of new capacity, according to BloombergNEF, crossing the 100-gigawatt annual threshold for the first time. Over the same stretch, venture funding for the storage technologies built to outlast those batteries fell 72 percent.

Both numbers describe the same market. Neither is wrong.

The 100-gigawatt era arrived faster than any storage forecast assumed. It took four years for annual additions to climb from 10 GW to more than 100 GW; solar needed roughly eight years to make the same jump and wind fifteen. BloombergNEF expects 158 GW of additions in 2026, and cumulative capacity outside pumped hydro has reached 2.9 TW. Batteries now cover more than 40 percent of California's evening load on good solar days, and supplied over 60 percent of the state's hour-to-hour ramping in the first quarter of 2026, according to the International Energy Agency (IEA).

Yet almost all of that machine discharges in under four hours. The average commissioned utility project ran three hours in 2025, up from two in 2023. Everything beyond that mark belongs to a different industry with different physics, different buyers, and until very recently, no dependable revenue line. This guide maps which technology owns which slice of the duration ladder, who is actually paying for it now, and where the financing story still fails.

The Duration Ladder: Chemistry Follows Hours

Storage selection is duration selection. Below eight hours, lithium iron phosphate (LFP) wins on bankability, and Wood Mackenzie expects lithium-ion to hold roughly 85 percent of the storage market through 2034 regardless of what emerges above it. Past the eight-hour mark, lithium's cost advantage stops compounding because the cells scale linearly with energy stored while utilization per dollar falls. That is the opening the alternatives are fighting over.

Two structural facts sit behind the table. First, flow batteries and compressed air carried nearly all early commercial LDES deployment precisely because they are bankable today: proven components, financeable offtake, no exotic materials science. Second, the newest entrants compete on a different axis. Iron-air does not try to beat lithium on efficiency. It charges from power that would otherwise be curtailed, often at near-zero marginal cost, so a 50 percent round-trip loss lands on electricity nobody could sell anyway. The relevant question stops being efficiency and becomes whether discharged electricity during hour ninety of a grid stress event is worth more than the input power cost during hour one. During a multi-day scarcity event, locational prices can clear above $1,000/MWh. That asymmetry is the entire business case.

What AI Changed: A Buyer Who Pays for Hours

For a decade, LDES developers pitched utilities and waited for capacity markets to price duration. The pitch mostly died in committee. Then data center demand rewrote the customer profile. An AI campus is an always-on load that measures outages in failed training runs, signs ten-year contracts, and increasingly cannot get a grid connection inside its construction window. Such a buyer will pay for duration directly, contract by contract, without waiting for regulators to design a market for it.

The deal, announced at CERAWeek in Houston in March 2026, matters less for its size than for its structure. Crusoe secured reserved volume, pricing and delivery terms under a "bring your own capacity" model: power assets procured alongside compute rather than requested from a utility. All systems ship from Form Factory 1 in Weirton, West Virginia, a former steel mill site now scaling toward 500 MW of annual module production. Form Energy's pipeline under agreement exceeds 75 GWh, anchored by a 30 GWh project with Xcel Energy and Google in Minnesota that ranks among the largest battery commitments ever announced by energy capacity.

Eight months later the pattern repeated at larger scale and off-grid entirely. Energy Vault signed its biggest commercial agreement: 1.25 GW of integrated power infrastructure serving a hyperscaler AI campus in Texas. The package combines battery energy storage systems (BESS), grid-forming power conversion and generator sets. The configuration deliberately bypasses the Electric Reliability Council of Texas (ERCOT) interconnection process, where queues in some US markets now run five years or longer. Projected revenue sits between $500 million and $600 million. Weeks earlier the same company broke ground on a modular data center campus in Snyder, Texas with Crusoe itself.

"It establishes a repeatable commercial platform that we believe can support substantial future expansion as hyperscaler AI infrastructure investments continue to accelerate globally."— Robert Piconi, chairman and CEO, Energy Vault

Repeatable is the operative word. One deal is an anecdote; two deals across two architectures, plus a 30 GWh utility anchor and a quadrupling of LDES additions to 2 GW in 2026 per BloombergNEF, start to look like a procurement channel. Data centers are becoming the first customer class to buy multi-day storage at scale without being asked to by a regulator.

The Money Problem Nobody Has Solved

Here is the paradox underneath the boom. Wood Mackenzie counted 49 percent growth in LDES installations in 2025 and, in the same outlook, a 72 percent collapse in venture capital funding for the sector. China absorbed 93 percent of the world's long-duration capacity additions. Private American and European investors looked at that scoreboard and walked.

The walk-away is rational on the numbers those investors were shown. LDES startups spent a decade promising they would repeat lithium's cost curve, and lithium refused to cooperate: cell prices kept falling through exactly the duration bands where alternatives needed headroom. Priya Shrivastava, an energy storage supply chain researcher at Wood Mackenzie, put the verdict plainly: "The dramatic cost reductions lithium-ion achieved over the past decade will be difficult for emerging LDES technologies to replicate."

The unit economics explain both lines at once. Storage projects live or die on levelized cost of storage (LCOS): total lifetime cost divided by energy actually delivered. Lithium amortizes cheap cells across daily cycling, thousands of cycles a year. A multi-day asset cycles perhaps fifty times annually, spreading fixed cost over thin volume, so its LCOS only clears the market when either the input fuel is nearly free, the output commands scarcity pricing, or both. The US Department of Energy's ARPA-E (Advanced Research Projects Agency-Energy) DAYS program set its benchmark accordingly: long-duration systems must reach roughly five cents per kilowatt-hour-cycle across 10 to 100-hour durations, a target its own documentation concedes likely requires round-trip efficiency above 50 percent plus rock-bottom storage media cost. Iron, water and air qualify on the second condition. Curtailed renewable power qualifies on the first.

What collapsed with venture funding was never the technology roadmap. It was the bridge financing between demonstration and bankability, the phase where first-of-a-kind projects burn cash proving insurers, lenders and grid operators wrong. The buyers who replaced venture investors did so for their own reasons. A hyperscaler facing a five-year queue treats an expensive storage contract as the cheaper option. A utility signing a 100-hour project buys resource adequacy insurance, not an arbitrage machine. Both buyers internalize reliability cost that merchant markets still refuse to price. Until capacity markets or state mandates pay for duration directly, that substitution is the sector's financing model.

Where Each Technology Breaks

Honest limits, technology by technology. Iron-air's weakness is geometry: the modules store modest energy per square meter, so a gigawatt-hour site spreads over land several times larger than an equivalent lithium array, and the rust cycle wastes roughly half the input electricity. Vanadium flow batteries sidestep degradation, the electrolyte effectively never wears out, but the metal itself is a commodity position as much as a component, and miners' pricing power shows up in every project pro forma. Compressed air delivers excellent cost where salt caverns exist and nowhere else; geography picks the winners. Liquid air and gravity schemes remain niche entries hunting for their first at-scale references. Hydrogen remains the only seasonal answer and the most capital-intensive one, a bet on the 2030s rather than this decade's procurement cycles.

None of these constraints is secret, which is why the financing gap persists despite genuinely working hardware. Investors are not doubting that a 100-hour battery runs. They are doubting anyone will pay its full cost before 2030.

Underwriting Duration: Four Checks Before the Pro Forma

For anyone pricing one of these projects, four checks separate fundable from hopeful. First, read the input power contract before the technology sheet. A multi-day battery is a conversion machine for otherwise worthless electricity; if the offtake behind it pays retail-scale prices for charging power, the efficiency penalty stops being academic and starts eating the margin the pitch assumed away.

Second, interrogate the cycle count. Levelized cost math flatters any asset by assuming it runs. Fifty deep cycles a year is the honest planning number for multi-day duty in most markets today, not the daily-cycling assumption imported from lithium models, and every fixed dollar divides by that smaller denominator. The ARPA-E five-cent-per-kilowatt-hour-cycle target remains the cleanest public yardstick: a project claiming better economics without explaining its path to that order of cost is selling narrative, not arithmetic.

Third, identify who actually buys the output and on what tenor. The 2026 deals share one property: a named counterparty with an investment-grade balance sheet buying reliability directly, whether a Crusoe reservation or a utility resource-adequacy contract. Merchant LDES, selling into spot scarcity pricing with no capacity payment underneath, still has no completed financing at scale anywhere. The buyer line item matters more than the chemistry column.

Fourth, price the geography honestly. Compressed air lives or dies on salt geology, iron-air on land and interconnection position, hydrogen on cavern availability, and the American buildout itself concentrates in three states per US Energy Information Administration (EIA) data, with Texas alone planning 12.9 GW of battery additions next year. Queue position has become part of the asset. Two identical plants can differ by years of revenue purely on where they stand in line, which is precisely why the Energy Vault structure sold: it bought its way out of the line entirely.

One more number belongs in every memo: 93 percent. That is China's share of long-duration capacity added last year per Wood Mackenzie, which makes Western LDES projects simultaneously infrastructure plays and industrial-policy bets. Neither framing invalidates the other, but underwriters should know which one they are being paid for.

The Bear Case: Ten Days a Year

The strongest argument against the whole category comes from Wood Mackenzie's own demand analysis: two-to-eight-hour systems already cover about 90 percent of storage needs in most regions. Genuine multi-day discharge events occur fewer than ten days per year. Building hundred-hour assets for the tenth-busiest day of the year looks, from a spreadsheet, like insuring a beach house against meteorite strike.

The rebuttal is that spreadsheets do not sign interconnection agreements. As we wrote in August, the distributed version of the same idea, millions of parked EVs acting as a virtual fleet battery, remains stuck exactly because the grid has no mechanism to pay fleet owners for availability. Purpose-built multi-day storage faces the identical valuation gap from the opposite direction: the physics works, the accounting does not. What changed in 2026 is that a customer appeared with both the incentive and the balance sheet to ignore the accounting problem. If AI campuses keep buying duration directly, capacity market reforms pending in the UK and US gain proof points. If hyperscaler spending cools, the sector returns to waiting on regulators with weaker finances than it had in 2024.

Watch the order book, not the demos.