Battery Storage for Data Centres and Server Rooms in Kyiv: an Engineer’s Approach

Why a Kyiv Server Room Has a Different Battery Problem

The city’s data centre stock sits inside converted office towers, logistics terminals and pharma cold stores — rarely in purpose-built shells with unlimited floor rating. A 40‑string VRLA bank for a 400 kW IT load weighs roughly 12–14 tonnes. In a leased building on Verkhnii Val Street, that weight eats into usable square metres or demands structural reinforcement that the landlord will not pay for. LiFePO4 racks, at equivalent usable kWh, land at about 40 % of the mass, so the same floor cutout can carry twice the autonomy or free up space for revenue racks. That alone shifts the conversation from “battery replacement” to “remodel the power room”.

The second Kyiv-specific driver is the DTEK Kyiv Electric Grids connection protocol. Most commercial buildings in the central business districts connect at 10 kV through dual-transformer substations with a normally‑open tie. On loss of the primary feeder, the tie closes in 1.2–1.8 seconds, not milliseconds. During that interval the site’s own generation, if it exists, is still synchronising. The UPS bridge is the only thing that keeps servers running. That bridge must be designed for a deeper discharge window than the 15‑second VRLA sizing rule allows when the secondary feeder is under maintenance — a scheduled event DTEK notifies 72 hours ahead but still removes the automatic transfer option for the whole working day.

A Normal Working Day with the BESS Online

08:30. The DC plant is running at 78 % of rated IT load — 312 kW on the rack PDUs, plus 94 kW of cooling and auxiliary. DTEK supplies steady 10.2 kV ± 2 % through the main feeder; the on‑site 10/0.4 kV transformer stays at 60 % capacity. The BESS inverter‑chargers sit in grid‑following mode, floating the DC bus at 832 V. The EMS continually polls thirty‑six LiFePO4 racks, each at 99 % SOC, temperature spread across cells < 1.5 °C. Charge current is zero because the SOC dead‑band is set to 98–100 %. The only BESS‑driven cooling load is the glycol pump for the liquid‑cooled cabinets — roughly 1.1 kW continuous, less than the CRAC fan power of the old VRLA room it replaced.

14:00. A cloud‑computing customer spins up a render batch; IT load jumps from 312 to 447 kW in eight minutes. The transformer tap‑changer handles the voltage drop, but the BESS inverter’s droop control catches the 70‑millisecond dip before the generator controller even registers a frequency deviation. No batteries discharge — the inverter sources reactive current only, a feature impossible with a diode‑rectifier VRLA UPS. Facility staff see a 0.4‑cycle event on the power quality monitor and nothing on the server logs.

Emergency Outage: Zero‑Transfer Behaviour at 0.4 kV

15:22. A cable fault on the Naberezhne Highway 10 kV ring trips the primary feeder breaker. Voltage at the DC bus drops to 0 V in 3.8 ms. The BESS inverter bridges the gap in under 250 µs — the power modules switch from grid‑following to voltage‑source mode before the server power supplies’ hold‑up capacitors discharge below the 12‑ms typical hold time. IT load sees no disturbance. The static transfer switch stays closed; no electromechanical switching is involved.

The EMS immediately signals the generator set to start and opens the grid‑side contactor. The BESS alone carries the full 541 kVA (IT + cooling) for the next 42 seconds until the generator reaches rated speed and the synchronisation relay closes. During those 42 seconds the battery SOC drops from 99 % to 96.3 % — a discharge C‑rate of 0.6C, well within the 1C continuous rating of the CATL 280 Ah cells. Once the generator takes the load, the BESS returns to grid‑forming mode, running in parallel with the generator to supply any subsequent load steps the engine cannot follow. DTEK crews restore the feeder at 18:10; the BESS re‑synchronises with the grid, recharges at 0.3C and is back at 99 % SOC by 19:35.

Seasonal Peak: July Heat and the PUE Penalty

Kyiv sees two weeks every July with a wet‑bulb temperature above 22 °C. Air‑cooled chillers on the roof of a glass‑façade business centre lose 15–18 % of their rated capacity. The CRAC units work harder; PUE drifts from 1.35 to 1.52. That extra 70–100 kW of cooling load is electrical, so it lands on the UPS bus too.

A VRLA bank sized for 10 minutes at nominal load now delivers 7 minutes before the end‑of‑discharge voltage is reached — and if the battery room shares the same chilled‑water loop, its ambient drifts above 25 °C, accelerating plate corrosion. LiFePO4 racks with liquid cooling reject their heat directly to the building’s condenser water loop via a plate heat exchanger. Cell temperature stays below 30 °C even when the battery room hits 38 °C. The EMS derates nothing; the 10‑minute autonomy figure holds. Sizing for Kyiv therefore means using the July design‑day PUE, not the nameplate PUE, and adding a 15 % cooling margin on top of the IT load forecast. Without that margin the battery will be undersized exactly when you need it most.

VRLA‑to‑LiFePO4 Swap: Numbers That Pay for the Project

A 400 kW / 600 kWh VRLA bank replacement cost at end‑of‑life year 5 runs about 3.8–4.2 million UAH (cells, racks, cabling, installation). A LiFePO4 bank of the same usable kWh — 600 kWh delivered at 90 % DoD — costs roughly 5.6–6.0 million UAH installed. The capex delta is 1.6–2.0 million UAH. That delta recovers in 3.2–4.1 years from three line items: avoided VRLA replacement every 4–6 years, 22 % lower cooling energy (the battery room no longer needs dedicated precision cooling), and space recovered for revenue racks. If the freed 12 m² can host three 8 kW racks at a colocation rate of $700 per rack per month, the payback drops below 2.5 years.

At 0.4 kV, 200–500 kW is the sweet spot. Below 200 kW the fixed cost of the EMS and liquid‑cooling loop dilutes the return. Above 2 MW at low voltage, cable cross‑sections become unmanageable and we push the design to a 10 kV containerised BESS that sits on the MV side of the transformer — a different project with different economics, but one that works with the same DTEK connection study we always file before offering a proposal.