The milestone is real. Sometime in early 2026, the cumulative capacity of grid-scale battery energy storage systems globally crossed one terawatt-hour — a number that would have been considered implausible as recently as 2018. The IEA’s Global Energy Review 2026 documented 108 GW of new additions in 2025 alone, a 40% year-on-year increase, and confirmed that global battery storage capacity has grown more than 12-fold since 2020. In the United States, the EIA projects 24 GW of utility-scale battery storage additions in 2026, up from a then-record 15 GW in 2025.
What the headline numbers do not capture is that the fastest-growing grid storage market in the world — Texas under the ERCOT grid operator — just watched its revenue model collapse by nearly 90% in two years. The industry has proved it can deploy at scale. It has not yet proved it can generate predictable returns at scale. That gap is the constraint that will determine the pace of investment in non-Texas markets, and it is the constraint that data center power planners need to understand before building battery storage into long-term power procurement strategies.
The Deployment Picture
IEA data shows global grid-scale battery capacity grew more than 12 times between 2020 and 2024, reaching 124 GW by end-2024 before 2025 additions pushed the cumulative figure across 1 TWh. Installed cost fell alongside: from $511/kWh in 2019 to approximately $213/kWh in 2024, a 58% reduction driven by LFP (Lithium Iron Phosphate) chemistry displacing NMC (Nickel Manganese Cobalt) as the dominant grid storage technology. LFP’s lower cell cost, longer cycle life, and better thermal stability have made the chemistry shift essentially complete for stationary storage applications.
In the US, the geographic distribution of deployment is highly concentrated. Three states account for approximately 80% of planned 2026 battery capacity additions: Texas at 53% (12.9 GW), California at 14% (3.4 GW), and Arizona at 13% (3.2 GW). The largest single project is the 621 MW Lunis Creek battery energy storage system (BESS) in Jackson County, Texas. Three of the four largest US battery projects scheduled for 2026 completion are in Texas.
The concentration in Texas is not accidental. It reflects the ERCOT market structure: an energy-only, real-time-priced grid with no capacity market, where price spikes during demand peaks can reach the $9,000/MWh market cap. Battery systems that respond in seconds to frequency deviations earned outsized ancillary service fees. For several years, ERCOT was the most profitable jurisdiction in the world for standalone grid storage. That era has ended.
The ERCOT Revenue Collapse
ERCOT battery storage revenues peaked in 2023, when operators earned an average of $149 per kilowatt in annual revenue, with 84% of that revenue derived from ancillary services — primarily frequency regulation and spinning reserve. By 2025, projected average annual revenue had fallen to $17/kW-year, an 89% decline. The share from ancillary services dropped from 84% to 48%. Most ERCOT battery operators are now posting profitability below 2.2% year-to-date.
The cause is structural, not cyclical. As ERCOT battery capacity grew from 8.6 GW at end-2024 to 11 GW by mid-2025 — having expanded from a low single-digit GW base in 2021–2022 — the scarcity premium that made ancillary services lucrative was competed away. This is the defining economic property of grid storage: adding more batteries in the same market compresses the price spreads that justified the investment. Energy arbitrage — buying cheap overnight power and selling into afternoon peaks — has become the dominant revenue source, but the spreads are narrower and more volatile than the ancillary service fees they replaced.
ERCOT launched its Real-Time Co-optimization plus Batteries (RTC+B) programme in December 2025, designed to help operators simultaneously optimise ancillary service participation and energy dispatch. Whether RTC+B stabilises operator economics across a full market cycle is not yet measurable. The market is also pushing toward longer-duration systems — 4-hour and 8-hour assets rather than 2-hour — partly in response to new Effective Load Carrying Capability (ELCC) accreditation signals and partly because longer duration preserves arbitrage value more effectively as the 2-hour spread compresses.
The ISO-by-ISO Problem
The ERCOT experience is being watched carefully by developers and investors in every other North American grid market, because it illustrates a pattern that plays out wherever storage deployment scales rapidly without concurrent market structure reform.
CAISO (California) currently offers the most favourable revenue environment outside ERCOT’s pre-saturation peak. Capacity payments around $200/kW-year exist alongside arbitrage and ancillary service revenues. But California’s interconnection queue is lengthy, and the state’s duck curve dynamics — oversupply midday, steep ramp at evening peak — mean that batteries are increasingly competing with each other for the same arbitrage window.
PJM (Mid-Atlantic/Midwest) has a capacity market that should, in theory, provide a bankable revenue floor for storage. In practice, PJM’s first interconnection transition cycle awarded agreements to only 21 BESS projects, adding 1.9 GW — most of which will not reach commercial operation until 2028–2030. Ongoing capacity accreditation rule changes for storage under ELCC methodology create uncertainty about what a battery project built today will actually earn in the capacity market five years from now. Non-recourse project financing is difficult when lenders cannot model revenue with confidence across the life of a 20-year asset.
MISO (Midwest) is further behind, with capacity accreditation modeling reforms still pending. The result is a two-speed deployment map: markets with energy-only pricing (Texas) or strong capacity payment structures (California) attract most of the capital, while markets where the revenue stack depends on multiple uncertain regulatory variables attract far less. The Morgan Lewis 2026 energy storage procurements analysis notes that standalone merchant storage is increasingly difficult to finance on a non-recourse basis in any market, and that hedging contracts and tolling arrangements are becoming structurally necessary.
- Global grid-scale battery storage crossed 1 TWh cumulative capacity in early 2026, with IEA confirming a 12-fold increase since 2020.
- US adds 24 GW of utility-scale battery storage in 2026, with 53% concentrated in Texas — a market whose ancillary service revenues have since collapsed 89%.
- Revenue models vary sharply by ISO: CAISO offers capacity payments, PJM has structural delays, MISO has no settled framework. Standalone merchant storage is difficult to finance outside contracted structures.
- AI data center operators are emerging as a new contracting counterparty — willing to sign long-term power agreements that provide the revenue certainty grid storage cannot obtain from spot markets alone.
The AI Data Center Demand Signal
The most significant near-term development in grid storage demand is one that sits outside the traditional utility procurement model entirely: the hyperscale AI data center build-out. Data center operators building for AI workloads need power that is reliable, available 24/7, and increasingly required to be low-carbon. The problem is that grid infrastructure expansions take five to ten years; data centers need to come online in 18 to 24 months. Battery storage — co-located with on-site generation and tied to renewable offtake — fills the gap that the grid cannot.
In March 2026, Form Energy and Crusoe announced a strategic capacity agreement for 12 GWh of multi-day iron-air battery storage to support AI data center power needs, with deliveries beginning in 2027. Meta has separately reserved up to 100 GWh of multi-day storage from startup Noon Energy. BloombergNEF has tracked 4.9 GW of energy storage announcements co-located with on-site fossil fuel generation at data centers — about 32% of all announced global on-site data center battery capacity.
BloombergNEF projects global AI data center battery shipments will reach 272 GWh annually by 2030. That trajectory makes AI data center procurement a more important demand driver for grid storage manufacturers than utility-scale power arbitrage within five years, by volume. As we covered in our analysis of data center power procurement, the hyperscale sector’s appetite for firm, low-carbon power is structurally driving new contracting models that do not exist in utility procurement.
A data center operator willing to sign a 15-year power services agreement with a storage developer provides exactly the contract certainty that lenders require to underwrite non-recourse project debt — bypassing the ISO revenue stack problem entirely. This is the same logic that enabled renewable power development to scale through power purchase agreements (PPAs) before capacity markets matured. The counterpart risk is that multi-day storage technologies — iron-air, flow batteries, long-duration lithium — are at varying stages of commercial readiness. Form Energy’s iron-air chemistry has operating pilots but has not yet delivered at the scale implied by the Crusoe agreement.
What to Watch
The key 12-month signal is ERCOT operator financials under the new RTC+B regime. If average annual battery revenue stabilises above $40–50/kW — closer to the CAISO capacity payment floor — the market has found a new equilibrium. If it stays near $17/kW, the deployment pipeline in Texas will slow despite the headline capacity numbers, and the concentration of US storage investment will begin to shift toward California and contracted AI data center deals.
In PJM, watch the ELCC capacity accreditation rule revision. A favourable update could materially improve the revenue model for the 1.9 GW of projects already holding interconnection agreements and would accelerate the next wave of financing. An unfavourable or delayed update extends the bankability gap for PJM and MISO projects by at least another 12 to 18 months — and pushes the geographic diversification of US storage investment further into the future.
On AI data centers, the Form Energy / Crusoe 12 GWh agreement will serve as the benchmark for whether multi-day iron-air storage can deliver at contracted volumes and timelines. If 2027 deliveries proceed as scheduled, it validates a new long-term contracted demand category that the grid storage industry has not previously accessed at scale. The industry has crossed 1 TWh. The next milestone worth watching is not a capacity number — it is a revenue model that holds up as capacity keeps growing.
This article was produced with AI assistance and reviewed by the editorial team.Related posts:
The Grid Cannot Keep Up: How AI’s Power Demand Is Forcing Hyperscalers to Buy the Infrastructure Themselves
The Solid-State Battery Gap: Strong Lab Numbers, Unproven Factories
Hyperscaler AI Capex Has Crossed Into Debt Financing. Here Is What That Structural Shift Means.
Section 232 Semiconductor Tariff Report: What AI Infrastructure Buyers Must Model Now
