Sizing a UPS for a data center starts at the rack and works upward, not at the nameplate of a power module: total the IT load from rack density, convert it to total facility power with the site PUE, then choose a redundancy architecture and derate for heat and harmonics. This order is what keeps the result from being either undersized at peak or wildly overspent on idle capacity.
The only number that matters initially is the load the servers actually draw. Multiply rack count by average watts per rack: two hundred racks at eight kilowatts each is one point six megawatts of IT load. Do not size the UPS from the rack power-strip rating or the breaker label, because those reflect worst-case provisioning, not real draw. Measure at the rack power distribution unit, or take the design density and apply a realistic utilisation factor, typically 60 to 80 percent, so the estimate reflects what the hall will pull in normal operation rather than the day every cabinet is fully populated.
The UPS usually feeds more than the servers. Cooling, pumps and some lighting sit on the critical bus, and their share is captured by the power usage effectiveness ratio. At a PUE of 1.4, one point six megawatts of IT load implies about two point two four megawatts of total facility power on the protected bus. Sizing the UPS to the IT figure alone would leave the cooling load uncovered during an event, which is a classic and expensive mistake. Use the site PUE, not an industry average, because a free-cooled hall and a legacy raised-floor hall differ by twenty percent or more.
The core relation is straightforward: UPS capacity in watts equals IT load in watts multiplied by PUE, multiplied by a derating factor for temperature and harmonics, then divided by the target loading headroom. As numbers: 1,600,000 watts of IT load times 1.4 PUE equals 2,240,000 watts total. Apply a 0.95 derating for a warm aisle and mild harmonic content to reach about 2,128,000 watts, then size so this sits at roughly 80 percent of rated capacity to leave growth headroom, giving a required rating near 2,660,000 watts, about 2.66 megawatts. In kVA at a 0.9 output power factor that is roughly 2.95 megavolt-amperes, so a practical configuration is five 450 kilovolt-ampere modules in a 4 plus 1 arrangement delivering 1.8 megavolt-amperes of usable capacity with one module spare.
Single module, N, is cheapest and acceptable only for a non-critical hall. N plus 1 adds one spare module so any single fault never reduces capacity, and it is the common choice for enterprise and training workloads. Two N, sometimes written 2N, duplicates the entire power train so either side can carry the full load, which is the standard for financial, healthcare and regulated environments. Two N plus 1 adds a spare on top of full duplication for the largest sites. The architecture, not the module size, drives most of the cost, so decide it from the downtime tolerance of the workloads before quoting capacity.
Inverters lose rated capacity as ambient temperature rises, and a hall running at 35 degrees Celsius can shed several percent versus the 25 degree Celsius nameplate. Harmonic current from switched-mode supplies raises the thermal load on the inverter and the distribution gear, so a 0.95 derating is a reasonable starting allowance for a dense hall. Leave explicit space for growth: AI and high-performance clusters rarely stay at their first-phase load, and a UPS sized to the watt with no headroom forces a forklift upgrade at the worst moment. Targeting 70 to 80 percent loading at day one keeps both margins available.
Runtime is a business decision, not a fixed rule. Where a generator starts reliably, five to ten minutes of battery covers start and stabilisation, and lithium iron phosphate supports fast recharge that shortens the exposed window. Where there is no generator, runtime is whatever the site must survive unassisted. Either way the UPS must handshake cleanly with the generator: online double-conversion accepts the wide input frequency swing during generator loading, while a line-interactive unit may reject the source and drain its battery. Confirm the transfer and reconnection logic before commissioning.
Data-center halls use three-phase distribution, commonly 415 volt line-to-line with 240 volt line-to-neutral, and the UPS must support that natively without an added transformer stage that adds loss and floor space. Check the neutral handling, because modern loads are not perfectly balanced and a weak neutral arrangement causes nuisance trips. Higher-density halls are moving toward raised distribution voltage or direct-current buses, so confirm the unit matches the intended scheme rather than forcing a retrofit later. Our data-center power configurations are outlined at https://www.upsboss.com/data-center/, and the supported modules are listed at https://www.upsboss.com/products/.
Key takeaway: size a data-center UPS bottom-up from rack load and PUE, choose redundancy from downtime tolerance rather than cost alone, and derate for heat and harmonics so the result holds at peak instead of on paper.
Why not just size the UPS to the IT load?
Because the protected bus also carries cooling and supporting infrastructure, and PUE captures that share. Sizing to IT alone leaves the cooling load uncovered during an event, which can defeat the whole design.
Is N plus 1 enough for a data center?
For most enterprise and training workloads yes, because a single module fault never reduces capacity. Regulated, customer-facing or healthcare halls usually specify 2N, where either full train can carry the load alone.
How much runtime should I plan for?
If a generator is present and reliable, five to ten minutes covers start and stabilisation. Without a generator, runtime is whatever the site must endure unassisted, and that is a business decision rather than a technical constant.
Contact: Frank Zhang
Phone: +86-135 5688 8641
Email: frank@upsboss.com
Add: Jufeng Road, Guangming Street, Guangming District, Shenzhen City, Guangdong Province, China
