Battery Energy Storage for Data Centres and Server Rooms

What Actually Breaks in a Data Centre Power Chain

Start with the physics. A data centre load is nonlinear and peaky: switch-mode PSUs draw current in short pulses, so RMS current is 40–60% higher than the sine-wave average. Your UPS output transformer and filters see that as heat, and the battery sees it as high-rate discharge. Lead-acid strings sag under pulse load, which is why you need 20–30% oversizing to keep the DC bus above the inverter cut-off. That oversizing costs floor space and cooling.

The second failure point is autonomy. At 0.8 power factor, a 1000 kW rack pulls 1250 kVA. A typical VRLA string rated for 15 minutes at C-rate is actually good for 11 minutes at 25°C, and 6 minutes at 35°C — which is the ambient in most server rooms. So you design for 15, get 6, and the generator may not be ready. That is not a hypothetical; it is the standard failure sequence in post-incident reports.

How LiFePO4 Changes the Equation

LiFePO4 cells have a flat discharge curve — 3.2 V per cell drops to 2.8 V only at 80% depth of discharge. That means the inverter sees a stable DC bus, so you can run the system down to 10% DoD without worrying about cell reversal. The energy density is 3–4 times lead-acid, so the same footprint holds 3–4 times the capacity. Thermal runaway is rare with LFP, and the BMS monitors cell temperature, voltage, and current continuously.

The transfer time is the real killer feature. A double-conversion online UPS switches to battery in under 4 ms, but the transfer from UPS to generator is a mechanical breaker — 100–200 ms. During that window, the load is on the UPS inverter, which is battery-fed. If the battery is already at 80% DoD, you are on borrowed time. A BESS with a static transfer switch and a grid-forming inverter can hold the load indefinitely, because the BESS is always online. No break, no mechanical movement, no momentary dip.

So the swap is not just chemistry. You are replacing a single-purpose device (UPS) with a multi-function asset: it does backup, it does peak shaving, it does frequency regulation, and it does it with a 20-year life instead of 5.

Sizing: Power and Capacity, the Rough Way

For power, take the critical load kW and divide by the inverter efficiency (0.96–0.98). Add 20% headroom for inrush and future growth. For 200–2000 kW, that is your inverter rating. But do not size the battery for 15 minutes. Size it for the time it takes your generator to start and synchronise, plus a safety margin. In Ukraine, grid outages can last 4–6 hours, so if you run on generator for more than an hour, the BESS is a bridge, not a replacement. For a 1 MW load, 30 minutes of autonomy needs 500 kWh at 1C — that is a 20 ft container with liquid cooling.

Capacity is load (kW) × autonomy (h) ÷ DoD (0.9) ÷ efficiency (0.95). Example: 500 kW for 2 hours = 1000 kWh ÷ 0.9 ÷ 0.95 = 1170 kWh. That is a 1.2 MWh system. At today's prices (LiFePO4 at $130–150/kWh wholesale), the battery pack is around $150k–180k. Add inverters, switchgear, EMS, and installation — total CAPEX is $250–300k. Payback comes from three streams: UPS replacement savings (VRLA strings every 5 years, disposal costs), PUE improvement (less cooling load), and energy arbitrage if you have time-of-use tariffs. In Ukraine, industrial tariffs have a night/day spread of 30–50%, so a 1 MWh system can shift 1 MWh per cycle, saving 3000–5000 UAH per day.

What Happens If You Size It Wrong

Oversize the battery and you waste CAPEX — the cells will sit at high state of charge most of the time, which accelerates ageing. Undersize the power and the inverter clips on a cold start, dropping frequency and tripping your servers. Undersize the capacity and you get the classic failure: generator starts, but the battery is already empty, so the UPS goes to bypass and the load dips.

There is also a subtle thermal issue. If you put a 1 MWh LFP system in a room designed for lead-acid, the heat rejection is lower (LFP is 95% round-trip efficient vs VRLA 80%), but the room still needs ventilation. With liquid cooling, the cabinets reject heat to a chiller loop — if that loop fails, the BESS derates. Our designs always include a dry cooler and a dual pump set, because a data centre cannot afford a single point of failure.

What We Need From You

To produce a feasibility study, we need three things. First, a load profile: 1-minute or 15-minute power data for at least one full week, including the critical load and the IT load. Second, your electricity bill for 12 months, so we can calculate the arbitrage and demand charge savings. Third, a single-line diagram of the existing power distribution, including the UPS, bypass, and generator connections.

Without the load profile, any sizing is a guess. We are not going to sell you a 2 MWh system on a napkin calculation. We are an EPC contractor — we do audit, design, supply, installation, commissioning, and service. We have built 50 kW to 5 MW systems, and we know the pitfalls of retrofitting into an existing data centre.