Blog
AI Demand Fuels Long-Duration Energy Storage Boom
Estimated reading time: 5 minutes · Last updated:
Long‑duration energy storage is attracting fresh commercial momentum because AI data centers need power that can be delivered for tens of hours rather than a few. Since early 2025 the United States has seen at least 150 GWh of long‑duration storage projects announced, approved or financed, and tech firms are placing capacity orders that span roughly from 10 hours to as long as 100 hours. That U.S. surge was disclosed on 18 September by Bloomberg. Major customers such as Meta and Google are now backing deployment and pilots that could turn long‑duration concepts into operating systems.
Four hours of energy storage is far from sufficient.
Amanda Yang, head of clean and renewable energy at Meta
Key takeaways
- The disclosure put U.S. long‑duration projects since early 2025 at least 150 GWh, more than five times the planned capacity at the end of 2024.
- Meta has an agreement with Noon Energy for systems that Meta says can store power for as long as 100 hours.
- Google is partnering with Energy Dome to deploy a carbon‑dioxide‑based system in Arizona; Energy Dome says the installation will offer roughly 10 hours of storage.
- AI data centers now account for roughly half of Noon Energy's planned deployments, Aric Saunders said.
Table of contents
- Key takeaways
- Why AI data centres change the storage brief
- Which technologies and deals are moving to pilots or contracts
- How orders change the finance and scale equation
- What remains to be demonstrated before broader industry adoption
- Case for and against near‑term growth
- What to be careful about
- Frequently asked questions
Why AI data centres change the storage brief
AI data centers run large clusters of servers with sustained, high electricity use and an increasing preference for lower‑carbon supply. Conventional lithium‑ion battery installations that provide three to four hours of backup or peaking capacity match many renewables profiles but not the multi‑day or multi‑tens‑of‑hours energy windows some AI workloads can require when paired with variable renewables.
Long‑duration energy storage is defined by its ability to discharge electricity for tens of hours or even several days; companies and project proposals cite durations ranging from about 10 hours to as much as 100 hours. Amanda Yang, head of clean and renewable energy at Meta, summarised the gap plainly: "Four hours of energy storage is far from sufficient." That shortfall is the commercial logic behind tech buyers seeking systems that shift generation and consumption over much longer intervals.
Which technologies and deals are moving to pilots or contracts
Tech firms are taking different technical routes and placing visible orders that give startups and manufacturers commercial validation. Google is working with Energy Dome on a carbon‑dioxide‑based system in Arizona; Energy Dome says that installation will offer roughly 10 hours of storage, while Meta has signed with Noon Energy for installations that Noon Energy describes as capable of delivering up to 100 hours.
Other pairings named include Crusoe with Energy Vault on gravity‑based storage and a purchase order to XL Batteries from Prometheus Hyperscale that XL Batteries' co‑founder Tom Sisto says provided market validation. These partnerships show buyers testing several storage chemistries and architectures rather than converging on a single technical standard in the near term.
How orders change the finance and scale equation
Large procurement by hyperscale customers does two things for nascent storage firms: it provides revenue visibility and it reduces financing frictions. Aric Saunders, Executive Vice President of Noon Energy, said AI data centers now account for roughly half of the firm's planned deployments; two years ago that share was nearly zero. Tom Sisto of XL Batteries described a Prometheus Hyperscale purchase order as the kind of commercial signal that helps unlock capital for production scale‑up.
Yiyi Zhou of BNEF expects that operational experience and scaled production will drive down costs. She said deployment in AI data centers will generate real‑world data likely to lower costs and broaden applications, moving some long‑duration technologies from pilot projects to repeatable commercial projects.
What remains to be demonstrated before broader industry adoption
Long‑duration storage has historically faced steep costs and limited large‑scale validation, which helps explain why the sector was slow to commercialize. The principal technical and commercial hurdle remains proving round‑trip efficiency, lifetime durability and safety at utility scale while competing with ever‑cheaper short‑duration lithium batteries for many grid services.
There is also customer concentration in early orders: Noon Energy's recent planned deployments are roughly half tied to AI data centers, which accelerates scale but raises exposure if that procurement pace slows. Finally, different technology pathways mean outcomes will vary by vendor and chemistry; some approaches avoid critical minerals, Lucia Tian of Google noted, but others may face different supply or manufacturing bottlenecks.
| Company | Technology | Claimed duration | Notable customer/partner |
|---|---|---|---|
| Energy Dome | Carbon‑dioxide based thermal storage | Up to 10 hours | |
| Noon Energy | Long‑duration storage (company technology not specified in text) | Up to 100 hours | Meta |
| Energy Vault / Crusoe (collaboration) | Gravitational potential energy | Not stated in the material | Crusoe |
| XL Batteries | Company technology not specified in text | Not stated in the material | Prometheus Hyperscale (purchase order) |
Case for and against near‑term growth
The case for
- AI data centers are placing large orders and partnerships that provide revenue visibility and financing support for startups.
- Deployment at scale in hyperscale data centers should produce operational data and production learning that, as BloombergNEF's Yiyi Zhou noted, can lower costs and expand applications.
- Some technologies under test do not rely on critical minerals such as graphite, a point Lucia Tian at Google highlighted, which could ease material‑supply constraints for particular pathways.
The case against
- Long‑duration storage was historically limited by high costs and lack of large‑scale validation, a constraint that still needs to be overcome before widespread grid adoption.
- Early demand is concentrated: Aric Saunders says AI data centers now account for roughly half of Noon Energy's planned deployments, which creates commercial exposure if those customers slow procurement.
- Multiple technical approaches are in play, and differing efficiency, lifetime and cost profiles mean winners are not guaranteed across the board.
What to be careful about
- Commercial performance risk: systems must demonstrate reliable round‑trip efficiency and durability at scale before buyers beyond hyperscalers commit.
- Demand concentration: Noon Energy reports roughly half of its planned deployments are for AI data centers, creating exposure if that procurement falters.
- Cost and validation: the sector’s historical barrier has been high cost and limited long‑duration operational data; failure to lower costs would limit adoption outside specialised buyers.
The bottom line
AI data centers have shifted a nascent clean‑energy category from concept to contracting by offering large, creditworthy buyers and repeatable project scale. The 150 GWh pipeline in the United States since early 2025 and named deals from Meta and Google show manufacturers and startups are securing the orders they need to test economics at scale. Success will depend on published operational performance, cost reductions from mass production, and whether demand diversifies beyond hyperscalers; until then the field will be defined by pilot outcomes and the first multi‑week operating runs.
What to watch
- Watch for public commissioning or operational reports from Google and Energy Dome’s Arizona deployment; no date has been set.
- Watch for first commercial operation or performance data from Meta’s Noon Energy installations that Meta has described as capable of up to 100 hours; no date has been set.
- Watch for published operational metrics or cost‑of‑storage updates from startups whose projects were financed since early 2025; no date has been set.
Frequently asked questions
What is long‑duration energy storage?
Long‑duration energy storage stores and releases electricity for tens of hours or several days rather than the single‑digit hours typical of many lithium‑ion systems; companies and project briefs cite durations ranging from about 10 hours to as long as 100 hours.
How big is the recent U.S. pipeline for these projects?
The disclosure put U.S. long‑duration projects since early 2025 at least 150 GWh, a figure described as more than five times the planned capacity at the end of 2024.
Are these systems already in commercial use?
Some systems are moving to commercial pilots and early contracts: Google is deploying an Energy Dome CO2 system in Arizona — Energy Dome says it will provide around 10 hours of storage — and Meta has agreements with Noon Energy for systems Noon describes as capable of delivering as much as 100 hours, but the sector still needs broader long‑term operational data.
Related reading