Traditional UPS battery systems were designed for one job: bridge the gap between a grid failure and generator startup. That job takes 10 to 30 seconds and requires a controlled, predictable discharge. AI GPU clusters do not produce controlled, predictable power demand. They produce millisecond-scale transient spikes as compute cores switch between idle and full utilization — a load profile that standard lead-acid and lithium-ion UPS batteries were never engineered to absorb.
The conventional response has been overbuilding: size the UPS for the spike, not the average, and accept the cost and footprint penalty. 84% of operators now prioritize total cost of ownership over upfront price (per the Data Center Energy Storage Industry Insights Report), and 57% say they need higher power density in less space. The math on oversized legacy UPS systems does not close against those requirements.
Nickel-zinc electrolyte: water-based. Thermal runaway: not possible (no flammable electrolyte). Power density: higher than lead-acid, competitive with lithium-ion in smaller footprints. Lifecycle greenhouse gas emissions: 25–50% lower than competing chemistries (third-party verified). Recyclability of primary materials: over 90%. Fire suppression requirements: reduced, enabling deployment closer to critical equipment.
A 1,000-GPU training cluster does not draw 1,000 times the power of a single GPU continuously and evenly. It draws varying amounts as different layers of a model compute simultaneously, as data loads between memory and compute, and as batch processing cycles start and stop. The peak-to-average ratio for a large GPU cluster can be 3:1 or higher over sub-second timescales.
Power distribution systems designed for steady-state loads see these spikes as fault events. Oversizing the UPS system adds capacity margin to absorb them. An alternative approach — using battery chemistry optimized for rapid charge-discharge cycles at the UPS level — addresses the problem by making the battery itself the transient absorber, rather than requiring the entire power distribution system to be sized for peaks that represent milliseconds of actual operation.
Lithium-ion is the default data center UPS battery chemistry in 2026. It offers good energy density and a known maintenance profile. Its liabilities are thermal runaway risk and a supply chain that has shown significant price volatility. Lithium-ion cells require active thermal management — which adds another cooling load to a facility already managing hundreds of megawatts of server heat.
Nickel-zinc uses a water-based electrolyte. Thermal runaway is chemically impossible — there is no flammable solvent to ignite. That safety profile allows deployment in closer proximity to critical equipment without the fire suppression infrastructure requirements that lithium-ion imposes. The footprint reduction is real and measurable. For a facility where every square meter of floor space has a power and cooling cost attached, that reduction compounds.
Power systems and cooling systems in a data center are not independent. UPS batteries that require active thermal management add to the facility's cooling load — and add a distinct thermal management requirement (battery chemistry has specific temperature ranges for optimal performance and safe operation) separate from the server cooling requirement. A battery chemistry that does not require active cooling, and can be deployed in locations within the facility that are not in the primary cooling zone, simplifies the overall thermal management architecture.
This is not the central argument for nickel-zinc in AI data centers, but it is a real one. Every component of the facility that can be removed from the active cooling requirement is a component that reduces the thermal load the CDUs, dry coolers, and heat rejection systems have to manage.