BESS for data centers:
Why Big Tech is Investing Massively in Drives
The generative AI revolution has created an unprecedented demand for computing power. A single query to GPT-4 consumes 10 times more energy than a Google search query. Data centers have become the fastest growing segment of electricity consumption in the world — and BESS is turning from "nice to have" into a critical infrastructure component.
Explosive growth of data center capacity
According to the IEA, the global energy consumption of data centers in 2022 was ~460 TWh* (~2% of global consumption). By 2026 years, 1000+ TWh* is predicted, and by 2030 — potentially 2000-3000 TWh*. The main driver is the training and inference of AI models. One NVIDIA H100 cluster on 10,000 GPUs consumes 30-40 MW continuously.
Global energy consumption of data centers (TWh*h)
Why data centers need BESS
Tier III/IV data centers require 99.982-99.995% uptime (26-53 minutes of downtime per year). For AI clusters, the requirements are even stricter — every second of downtime means the loss of millions of dollars in computing resources and SLA violations. The traditional scheme "mains + DSU + UPS" no longer copes with the scale and requirements.
Problems of the traditional scheme
- Diesel generators (DGU) need 10-30 seconds to reach the nominal power. This "gap" is closed by lead-acid UPS, which degrade in 3-5 years and require a significant area.
- Lead-acid UPS — bulky (5-10x more than LFP per unit of energy), toxic (lead, acid), limited resource (500-800 cycles), high conditioning costs (optimal temperature 20-25C).
- Diesel fuel — ESG risk, dependence on logistics, noise, emissions, bans in many jurisdictions (Amsterdam, Singapore).
- Grid connection — in many regions, the queue to connect to the network is 3-5 years. BESS allows you to obtain permission for a smaller amount of network power and compensate for battery peaks.
BESS as a replacement for traditional UPS
| Parameter | Lead-acid UPS | BESS (LFP) |
|---|---|---|
| Resource (cycles) | 500-800 | 6000-10000 |
| Term of service | 3-5 years | 15-20 years |
| Energy density | 30-50 W*h/kg | 150-200 W*h/kg |
| Working temperature | 20-25°C (narrow range) | 15-35°C (wide range) |
| Area (1 MW*h) | ~100 m² | ~20-30 m² |
| Additional revenue | No | Peak shaving, arbitration, FCR |
| TCO for 15 years | $500-700/kW*h | $200-350/kW*h |
A key advantage of BESS
Unlike UPS, which simply "sits and waits" for an accident, BESS works every day: peak shaving, energy arbitrage, frequency regulation. This turns the "cost" of backup power into an "asset" that generates income. Microsoft estimates additional revenue from BESS in data centers at $50-100/kWh per year.
Big Tech BESS projects
Deploys BESS in data centers in Belgium and the Netherlands. 100+ MWh LFP systems for peak shaving and grid services. Partnership of Fluence and Tesla. Goal: 24/7 carbon-free energy by 2030.
Microsoft
A pilot project to replace DSU with BESS in data centers in Ireland and Sweden. 36 MW*h Tesla Megapack. Total Investment in Iron-Air Form Energy batteries for a 100-hour reserve.
Amazon (AWS)
The largest corporate buyer of renewable energy. BESS integration in 50+ data centers for grid-scale peak shaving. Own PV+BESS projects for 1+ GW.
Meta (Facebook)
BESS deployment in Luleaa (Sweden) and Clonee (Ireland). Focus on thermal energy storage for cooling + LFP BESS for electricity. The goal is net-zero operations by 2030.
BESS architecture for data centers
The data center needs a specific BESS architecture, different from a typical C&I or utility-scale solution:
- 2N redundancy — duplication of each component (two independent BESS chains, each capable of supporting 100% load). The failure of one circuit does not affect the operation of the data center.
- Modular architecture — capacity scaling from 500 kW to 50+ MW through the addition of standard container units. A typical step is a 1-2 MW*h container.
- Hot-swap capability — the possibility of replacing a separate battery rack without shutting down the entire system. It is critical for a service without downtime.
- Dual-path power — BESS is integrated as a parallel power path: Grid → ATS → BESS → PDU → IT Load, of automatic switching <4 ms (online double-conversion topology).
Peak Shaving for hyperscalers
The most profitable use case of BESS in data centers — reduction of peak consumption from the network. Hyperscalers pay for two components: energy (kWh) and power (kW). Capacity is the maximum peak consumption for the month and determines the tariff. BESS "cuts" peaks, reducing contracted capacity by 20-40%.
Example: data center 10 MW of peaks up to 14 MW (during AI model training). Without BESS, it pays for 14 MW of capacity. BESS 4 MW / 16 MW*h covers peaks, reducing contracted capacity to 10 MW. Savings: $500K-1M per year on demand charges alone.
Ukraine as a data center market
Despite the war, the Ukrainian data center market has strategic advantages for development:
- Cold climate (average annual temperature 7-9°C) — natural cooling of servers 8-9 months a year, PUE <1.3 without mechanical cooling
- Educated IT personnel — 300K+ developers, strong DevOps/SRE culture
- Relatively cheap electricity — $0.06-0.10/kWh for industry (compared to $0.15-0.25 in Western Europe)
- Strategic location — at the intersection of network routes between Europe and Asia
Challenge: grid stability. BESS becomes not just an option, but a mandatory component for any data center in Ukraine. 500 kWh — 10 MWh systems allow you to ensure Tier III reliability even with an unstable network.
"A data center without a BESS in 2026 is like a data center without air conditioning in 2010. It's technically possible, but no serious customer will consider such a facility."
Power Density and cooling
Modern AI GPU clusters have a power density of 40-80 kW per rack (compared to 5-10 kW for traditional servers). This creates a challenge not only for power supply, but also for cooling. BESS can be integrated with the thermal storage system:
- Chilled water storage — BESS charges refrigeration units at night (cheap electricity), accumulating cold in tanks. During the day, cooling takes place in the tanks without additional load on the network.
- Liquid cooling integration — for 40+ kW rack GPUs, liquid direct cooling (CDU) can run from the BESS during peak hours, reducing the overall load.
- Free cooling optimization — AI EMS optimizes the transition between free cooling and mechanical cooling, taking into account the weather forecast and tariffs.
Frequently Asked Questions
What is the minimum BESS capacity for a Tier III data center?
Can BESS completely replace diesel generators?
What are the fire safety requirements of BESS in the data center?
How does BESS affect data center PUE?
What is the payback of BESS for the Ukrainian data center?
BESS for your data center
BESS calculation for the data center: UPS replacement, peak shaving, TCO analysis.