Energy storage spans a deployed core (lithium-ion cells for vehicles and the grid) and a research frontier (zinc-ion, hydrogen materials, supercapacitors, quantum batteries); the near-term action in the sources is battery energy storage as the physical buffer that lets gigawatt-scale AI data centres connect to stressed grids.
Summary
The mechanism that matters commercially in this source set is straightforward. Grid operators are moving to connect-and-manage interconnection, which lets very large new loads such as AI data centres connect without waiting for network upgrades, in exchange for accepting real-time curtailment and time-varying limits on power exchange at the point of common coupling. On-site batteries absorb the mismatch between what the compute wants and what the grid will allow, and can additionally be dispatched into ancillary-service markets. Where the workload has fast structure, a hybrid arrangement splits the disturbance, sending the energy-dominant component to the battery and the fast-varying component to a supercapacitor.
The parameters that decide value are interconnection strictness, degradation cost and state estimation. Value from co-located BESS more than doubles once peak-load and ramping limits become binding. Degradation-aware dispatch is treated as a first-class constraint in these models, which puts pressure on the weakest link: today’s monitoring relies on electrical and sometimes temperature measurements, limiting insight into internal electrochemical state. Two lines of work attack this: ultrasonic non-invasive interrogation, sensitive to mechanical changes correlated with electrochemical ones, and deep learning models forecasting remaining capacity and lifetime for predictive maintenance.
One caution on terminology. The phrase also appears in unrelated physics, for example the unidirectional energy storage of a deformed optical microcavity used to narrow a semiconductor laser linewidth. That work has nothing to do with grid or vehicle storage and should not be counted as evidence for it.
Evidence base
- Lithium-ion is described as the leading technology in consumer electronics, EVs and grid-level electrochemical storage, with monitoring limited to electrical and sometimes temperature measurements, June 2026.
- Daily value of co-located battery storage at a data centre increases by a factor of two or more when peak-load and ramping interconnection limits become binding, June 2026.
- Connect-and-manage interconnection lets gigawatt-scale AI data centres connect without prior network upgrades in exchange for real-time curtailment, with the operator’s acceptance rule opaque to the site, June 2026.
- Aqueous zinc-ion batteries are attractive for grid-scale storage on safety and cost grounds but current cathodes fade at rates below C/2; over 2000 synthesised compounds were computationally screened for replacements, June 2026.
- In situ multimodal STEM of an all-solid-state cell mapped three ordered LixLaTe3 phases from x = 1/3 to 3 with up to 5 per cent in-plane strain and found a reversible three-layer superdense tetragonal lithium phase, June 2026.
- Quantum battery work remains at minimal physical scale, for example a three-level battery in a single trapped 40Ca+ ion charged from an engineered thermal reservoir, May 2026.
Open questions
- Do the simulated benefits of co-located BESS under connect-and-manage, including the doubling of daily value when limits bind, hold in a metered deployment rather than an IEEE 39-bus case study?
- Can ultrasonic signals be rigorously linked to electrochemical state, the gap the 2026 review explicitly names, enough to enter commercial battery management systems?
- Does any cathode from the screen of over 2000 compounds actually eliminate zinc-ion capacity fade below C/2 in a cycled cell?
- How much of the data centre flexibility assumed by these frameworks, checkpoint-constrained training and DVFS-controlled inference, will operators actually concede in practice?
Assessment drafted 2026-08-31 from up to 18 KB sources using the technology-scorecard framework; scores are a draft read pending review.