AI Power Demands Push Battery Makers Toward Layered Storage Designs
Powering artificial intelligence infrastructure is pushing energy storage beyond what a single battery architecture can deliver, according to EVE Energy. The cell manufacturer used RE+ 2026 to present a tiered product lineup built around the argument that AI computing imposes three distinct response timescales simultaneously — ranging from minutes down to milliseconds.
Traditional grid-tied storage handles demand management and frequency regulation on a minute-scale basis. AI data centers add a second layer: rapid load swings that must be absorbed close to server racks in under two milliseconds. Bridging both extremes within one system design, EVE contends, produces unavoidable compromises in efficiency and cycle life.
The company’s grid-facing product is a 20-foot containerized system rated at 5 MWh with a specified round-trip efficiency of 93.6 percent. It relies on 315-ampere-hour cells optimized for frequency regulation and, per EVE’s own specification, shows no lithium-plating degradation during high-intensity testing. A second container addresses longer-duration needs at 6.9 MWh, rated for 10,000 cycles to 70 percent state of health. Both units are specified to hold internal temperature variation within five degrees Celsius.
At the distribution tier, high-power cabinets configured at 10C can sustain 660 kilowatts for five minutes. At rack level, a dedicated backup module using cylindrical cells is specified to switch in two milliseconds, targeting the ultra-short transients characteristic of GPU compute bursts.
Coordinating cell behavior, battery management, and operating strategies across all three tiers directly affects reliability and total cost of ownership — a design consideration growing more central as energy storage and digital infrastructure converge.
Source: PV Magazine
Become a member of 100 RE UA
Switching to 100% renewable energy in Ukraine is possible!