Sizing a data center backup generator is not the same problem as sizing a generator for a factory or an office building. The IT load is a high-power-factor, non-linear electrical load with harmonic content that interacts with UPS systems, cooling infrastructure, and step-load transients in ways a single "kVA = kW ÷ 0.8" calculation cannot capture. Get the sizing wrong by 15-20% and the consequences are not theoretical — they show up the first time you have a real utility outage at peak summer ambient.
This guide is written from a manufacturer's perspective. Over the past decade ASO Genset has delivered diesel generators for critical-load applications including hospitals (Vietnam), industrial continuous-process plants (Indonesia), marine offshore (North Sea), and most directly relevant to this guide, a 5 MVA (1,250 kVA × 4, ≈ 4 MW at 0.8 PF) Tier III-class data center backup project in Riyadh operating at 50°C peak ambient. We will walk through the actual sizing math — kW vs kVA, PUE, redundancy topologies, transfer-window requirements, tropical derating — and then share the top 6 sizing mistakes we see in real data center RFPs.
Table of Contents
- Why Data Center Generators Are Different from Standard Backup
- kW vs kVA — The #1 Confusion in Data Center RFPs
- Calculating Total Generator Capacity from IT Load (Step by Step)
- Redundancy Topologies: N, N+1, 2N, 2N+1 and Common Design Intent
- Transfer-Window & Step-Load Acceptance Requirements
- Tropical Data Center Considerations (50°C, Humidity, Dust)
- Top 6 Sizing Mistakes We See in Data Center RFPs
- ASO Critical Load Reference Cases (Saudi 5MVA + Vietnam + Indonesia)
- Specification Checklist for Data Center Generator RFQs
- Frequently Asked Questions
Quick Answer
Data center backup generator sizing follows a specific sequence: (1) start with the actual IT load in kW; (2) multiply by your facility's Power Usage Effectiveness (PUE), typically 1.4-1.8, to capture cooling and infrastructure load; (3) divide by the alternator power factor (typically 0.8) to convert kW to kVA; (4) add a 20-25% margin for transient step loads, future growth, and harmonic distortion from non-linear UPS loads; (5) finally, apply site-specific derating for ambient temperature and altitude based on the relevant OEM derating curve. For a 1 MW IT load at PUE 1.5 in a 50°C ambient, the required N capacity can reach approximately 2,750-2,900 kVA of 25°C-reference nameplate capacity after margin and derating. In a 3+1 N+1 configuration, this typically translates into four units in the 1,000 kVA class, so that three running units can carry the required N load at site conditions while the fourth remains standby. Sizing on nameplate rating at 25°C reference — and skipping PUE, harmonics, or site derating — is the most common (and most expensive) mistake in data center RFPs.
Why Data Center Generators Are Different from Standard Backup
A standard commercial backup generator is sized against a static load schedule: lights, HVAC, a few motors, maybe a freight elevator. The load profile is predictable, the power factor is near 0.85-0.9, and a 10-15% sizing margin is comfortable. A data center is fundamentally different on every dimension.
The Load Profile Is Non-Linear and High Power Factor
Server power supplies, UPS rectifiers, and switching power converters draw current in non-sinusoidal pulses. The result is harmonic distortion (typically in the 5-15% THD range, depending on UPS technology) that an alternator must handle without overheating. Modern data centers also often run at very high effective power factor (0.95+) — the load looks "clean" on paper, but the harmonic content stresses the alternator differently than traditional inductive loads.
Practical consequence: alternators for data center applications are often specified one frame size larger than the kVA math alone would suggest, with low-reactance design and appropriate end-winding capacity to handle harmonic heating without insulation degradation over the design life.
Cooling Is a Major Share of Total Load
The IT load you see on the rack is only part of the total electrical demand. Cooling (CRAC units, chillers, pumps), UPS losses, lighting, and auxiliary infrastructure can add 40-80% on top — captured by the facility's PUE metric. A 1 MW IT load at PUE 1.5 is actually a 1.5 MW electrical load on the generator. Skip this and you've undersized by roughly 33% on day one.
Outage Tolerance Is Very Low
A factory can ride through a 30-second blip. A data center cannot. The generator must restore emergency power within a tight transfer window — typically under 10 seconds for NFPA 110 Type 10 systems — with no measurable disruption to the UPS bus. The specific step-load and voltage/frequency recovery requirements depend on the project's load profile, UPS technology, and ATS sequence, and should be verified at site ambient during commissioning.
Compliance Stacks Up
Data center generators in many jurisdictions must comply with NFPA 110 (in the US), local AHJ requirements, and sometimes additional standards. Large hyperscale and colocation operators often add project-specific requirements on top of local code and international standards, covering items such as harmonic tolerance, step-load profile, fuel system redundancy, and remote monitoring integration. For the full NFPA 110 testing requirements that apply to most critical-load installations, see our NFPA 110 generator testing requirements guide.
kW vs kVA — The #1 Confusion in Data Center RFPs
If we had to identify the single most common mistake in data center generator RFPs, it would not be a sophisticated engineering error. It would be the simple confusion between kilowatts (kW) and kilovolt-amperes (kVA).
Definitions in Plain English
kW (kilowatts) = real power. This is the actual work the generator does — the power your IT load, cooling, and lights actually consume.
kVA (kilovolt-amperes) = apparent power. This is the total electrical capacity the generator must supply, which includes both real power and reactive power that flows back and forth between the load and the source without doing useful work.
Power Factor (PF) = kW ÷ kVA. This is the ratio between what the generator actually delivers as useful work and what its windings must physically carry. A typical generator is rated at 0.8 PF — meaning a 1,250 kVA generator can deliver 1,000 kW of real power at that PF.
Why Specs Are in kVA but Loads Are in kW
Generator manufacturers rate equipment in kVA because that's what the alternator's physical capacity actually is — the current the windings can carry without overheating. Data center facility teams measure loads in kW because that's what the utility bills them for and what the IT equipment consumes.
When these two numbers meet in an RFP without clear conversion, mistakes happen:
| Common Mistake | Typical Result |
|---|---|
| Requesting "1 MW generator" without specifying kW or kVA | Suppliers quote 1,000 kVA (≈ 800 kW at 0.8 PF) when you need 1,000 kW real power |
| Comparing two bids where one quotes kW and the other quotes kVA | The "cheaper" bid may actually be undersized by ~20-25% in real power terms |
| Sizing on IT load kW directly without applying PUE or PF | Generator capacity short of facility electrical demand by 40-60% on day one |
| Assuming PF 0.8 when modern data center load runs at PF 0.95+ | Alternator oversized for kVA but mismatched for actual load profile and harmonics |
The Conversion You Must Get Right
For a data center, the conversion sequence is:
Step 1: Total electrical load in kW = IT load (kW) × PUE
Step 2: Required generator kVA = Total kW ÷ Power Factor (typically 0.8)
Step 3: Apply margin (typically 20-25%) for transients and growth
Step 4: Apply site derating per OEM derating curve (ambient temperature, altitude)
We work through this in detail in the next section, but the principle is: always specify both kW and kVA in any data center generator RFP, and require bidders to confirm both numbers against your site conditions.
Calculating Total Generator Capacity from IT Load (Step by Step)
Here is the full sizing calculation, with a worked example for a representative data center. Note: all numerical ranges below are typical industry approximations; final values for any specific project depend on the OEM derating curves, the alternator family selected, and the project's load profile.
Step 1: Identify the IT Load
Start with the actual maximum sustained IT load at the rack level, measured in kW. This is what your servers, network gear, and storage consume at full operational load. Be honest about peak vs. average — generator sizing must address peak.
Example: 1,000 kW IT load (representing roughly 250 server racks at 4 kW per rack, a common high-density configuration).
Step 2: Apply PUE to Capture Total Facility Load
PUE (Power Usage Effectiveness) is the ratio of total facility electrical load to IT load. Cooling, UPS losses, lighting, and auxiliary infrastructure all show up here.
| Facility Type | Typical PUE Range | Notes |
|---|---|---|
| Highly-optimized hyperscale | 1.1 - 1.3 | Free cooling, optimized airflow, often in cool climates |
| Modern colocation (typical) | 1.4 - 1.6 | Tier III-class designs, mechanical cooling required |
| Tropical climate facilities | 1.6 - 2.0 | Common when relying on mechanical cooling without advanced free-cooling, liquid cooling, or highly optimized airflow strategies |
| Legacy enterprise data center | 2.0 - 2.5 | Older cooling infrastructure |
Example (continued): 1,000 kW IT load × PUE 1.5 (modern colo) = 1,500 kW total facility electrical load.
Step 3: Convert kW to kVA Using Power Factor
Generators are rated at a specific power factor — typically 0.8 lagging for traditional designs, sometimes higher for data center-specific units. The kVA rating is what the alternator windings can physically carry; the kW rating is the real power delivered at that PF.
Required kVA = Total kW ÷ generator PF
Example (continued): 1,500 kW ÷ 0.8 PF = 1,875 kVA required.
Step 4: Add Margin for Step Loads, Growth, Harmonics
Data center loads are not steady-state. UPS rectifiers create transients during transfer events; cooling compressors start in step-load fashion; future IT capacity additions need headroom. A common industry margin is 20-25% above the calculated base, though the exact figure depends on the load profile and projected growth.
Example (continued): 1,875 kVA × 1.20 margin = 2,250 kVA design capacity (or 1,875 × 1.25 = 2,344 kVA for a more conservative bound).
Step 5: Apply Site Derating
The kVA you just calculated is at 25°C ISO reference conditions. Real sites are not at 25°C. Actual derating depends on the engine model, alternator frame, cooling package, and OEM derating curve — generic numbers are starting points, not procurement specs. For ambient temperature and altitude derating math typical of tropical installations, see our tropical climate diesel generator selection guide — at 50°C ambient (typical Middle East summer), usable alternator output is commonly 15-25% lower than 25°C nameplate, depending on the specific equipment and configuration. Applied to our example, a 2,250 kVA design capacity may require a nameplate of approximately 2,750-2,900 kVA at 25°C reference.
Step 6: Apply Redundancy
The 2,750-2,900 kVA is your "N" requirement — the total capacity needed when all generators run at site conditions. For N+1 redundancy, each unit must be sized so that, with one unit offline, the remaining N units can carry full load. We cover redundancy topology choices in detail in the next section.
Worked Example Summary
1,000 kW IT load → ×1.5 PUE = 1,500 kW total → ÷0.8 PF = 1,875 kVA base → ×1.20 margin = 2,250 kVA design → site derating at 50°C (illustratively ≈22%; OEM-dependent) → ~2,900 kVA total nameplate requirement → in a 3+1 N+1 topology, this translates into four units in the 1,000 kVA class each. Specific per-unit ratings depend on the OEM derating curve and the acceptable load factor at peak ambient. The key takeaway: once PUE, PF, margin, derating, and N+1 redundancy are accounted for, the installed generator nameplate can be roughly 3× the IT load number — which is why "1 MW IT load needs a 1 MW generator" is a dangerous shortcut.
Redundancy Topologies: N, N+1, 2N, 2N+1 and Common Design Intent
Generator redundancy is not a "more is always better" decision. It is a deliberate trade-off between availability requirements, capital cost, operational complexity, and floor space. Each topology has a typical design intent, though the actual reliability tier of a facility depends on much more than generator topology alone.
The Four Standard Topologies
| Topology | Configuration | Indicative Availability | Commonly Associated Design Intent |
|---|---|---|---|
| N | Just enough capacity, no spare | ~99.671% | Often seen in basic backup / Tier I-style designs |
| N+1 | One spare unit on top of N | ~99.741% | Common in redundant capacity component designs; may support Tier II/III depending on distribution paths and overall facility design |
| 2N | Two complete independent systems | ~99.982% | Often used where concurrent maintainability or higher fault tolerance is required |
| 2N+1 | Two systems plus one extra spare | ~99.995% | Hyperscale, financial, and mission-critical designs beyond minimum redundancy |
Availability percentages above are commonly cited industry approximations. Actual Uptime Institute Tier certification (and equivalent ratings under TIA-942 or other frameworks) depends on the complete facility topology, distribution paths, mechanical infrastructure, maintainability, operations, and testing regime — not generator topology alone. A facility with N+1 generators may or may not qualify as Tier III depending on the rest of its design.
The N+1 Calculation Trap
"N+1" sounds straightforward but contains a subtle trap we see in roughly half of data center RFPs we review. The error: sizing each unit at "(total load) ÷ (N+1) units" instead of "(total load) ÷ N units, plus one additional spare unit of the same capacity."
Wrong approach: 3,000 kVA total site requirement, 4 units in "N+1" → 750 kVA per unit. If one fails, the remaining 3 units carry 1,000 kVA each — well over their 750 kVA rating, and they will trip on overload.
Correct approach: 3,000 kVA total site requirement, N+1 = (3 units required to carry load) + 1 spare = 4 units total, each rated at 1,000 kVA. If one fails, the remaining 3 carry exactly 1,000 kVA each — their full rating. The +1 unit is genuinely spare.
The Derating Compounding Effect
Now combine the N+1 calculation with site derating. In Riyadh at 50°C, a 1,250 kVA nameplate generator may deliver something in the range of 950-1,050 kVA usable, depending on the specific equipment and configuration. If your N+1 is sized on nameplate ratings, you may have N+0 in practice — the moment ambient hits peak summer and one unit goes down for maintenance, you are running at or above each unit's rating instead of the planned load factor. This is precisely how the Riyadh project below was sized: starting from the derated capacity, not the nameplate.
2N vs N+1: When Does Each Make Sense?
N+1 is a common choice for many modern colocation designs: cheaper, simpler, and sufficient availability for most enterprise workloads when paired with appropriate distribution and operations. 2N becomes appropriate when:
- The design requires concurrent maintainability (the ability to take any single component offline without affecting load)
- Higher Tier certification is the design intent
- The workload value justifies the typical 40-60% capital cost premium of full duplication
- The end customer specifies it (often the case with hyperscale and large enterprise contracts)
2N+1 is rare — generally reserved for hyperscale build-outs and high-value financial operations where any downtime translates to immediate revenue loss.
One downstream consequence of 2N and 2N+1 over-provisioning that is rarely flagged at the design stage: each generator routinely runs at very low load during the monthly no-load test cycle that data centers contractually require, and that combustion-chamber under-loading is the structural cause of wet stacking. Data center wet stacking is less common than marine wet stacking because the test cycles are shorter, but the same prevention discipline applies — periodic load bank testing, not no-load idling. For the full diagnostic and prevention framework (developed for marine vessels but mechanistically applicable to any low-load generator), see our wet stacking causes and prevention guide.
Sized the generators — now design the redundancy? Sizing determines the value of N (the minimum capacity to carry the IT load); redundancy design determines what sits beyond N to deliver the target Uptime Tier. For a full walkthrough of data center generator redundancy design — including the Uptime Tier standard mapping, N+1 vs 2N architectures, DCC vs ESP duty ratings, 12-hour vs 72–96-hour fuel autonomy, PMG excitation requirements, and a 4-step tier decision framework — see our companion pillar with downloadable 3-page design checklist PDF.
Transfer-Window & Step-Load Acceptance Requirements
A data center generator must do something a standard backup generator never has to do: restore emergency power within a strict transfer window, with step-load acceptance verified against the project's specific load profile, UPS behavior, ATS sequence, and site ambient conditions.
The 10-Second Standard (and Where It Comes From)
The 10-second figure is rooted in UPS battery runtime economics. Modern data center UPS systems typically carry 5-10 minutes of battery runtime to ride through utility outages. The generator must be online and stable well before that battery reserve is consumed. NFPA 110 defines this as a "Type 10" system — emergency power restored within 10 seconds — and this is the standard most North American data centers and many international tropical facilities follow.
What "Restoration Within Transfer Window" Actually Requires
Whether the project specifies sub-10-second transfer or a different design window, the practical engineering requirements are similar:
- Block heater on the engine. Engine oil and coolant kept at typical operating range (often ~40-50°C) continuously, so the engine starts close to operating temperature rather than ambient cold. Important in tropical climates too — ambient may be 25°C overnight, below operating temp.
- Pre-energized excitation. The alternator must build voltage within 1-2 seconds of engine reaching rated speed.
- Engine governor tuned for step-load acceptance. Standard tuning may not handle the project's specific step load — tuning for transient response should be specified and verified at site ambient.
- Alternator capable of step-load acceptance. Some alternator designs cannot accept large step loads without voltage collapse. Specify the required step-load percentage and voltage/frequency recovery limits explicitly in the RFP.
- ATS tuned for the design transfer cycle. The ATS sequence — sense utility loss → command engine start → confirm voltage and frequency → transfer load — must complete within the design transfer window.
Where the Design Requires Full Block-Load Pickup
Some designs require full block-load pickup; others use step-load profiles where the UPS load is transferred first and mechanical loads are added in stages. Where full block-load pickup is required, the RFP should explicitly state:
- The required load step percentage (e.g., "100% block load" or "70% step 1, 30% step 2 after 5 seconds stabilization")
- Voltage and frequency recovery limits (e.g., voltage dip ≤ 20%, frequency dip ≤ 10%, recovery within 3 seconds)
- Test conditions, including site ambient temperature for commissioning verification
Step-Load Performance at Site Ambient (Critical for Tropical / Cold)
OEM step-load and recovery performance is almost always quoted at 25°C ambient. At 45°C+ ambient, the engine air intake is hotter, fuel atomization changes, and the alternator excitation system operates differently. Specify your step-load requirement at site ambient, and require the supplier to verify at that condition during commissioning — not only at the factory acceptance test in a temperate factory.
NFPA 110 Compliance for US Markets
If your data center is in a US jurisdiction or serves customers that require NFPA-compliant systems, NFPA 110 compliance is typically non-negotiable for the diesel emergency power system. The full testing and maintenance regime — weekly inspections, monthly operational tests, annual load bank testing — is detailed in our NFPA 110 generator testing requirements guide.
Tropical Data Center Considerations (50°C, Humidity, Dust)
The data center boom is now strongly weighted toward emerging markets in Southeast Asia, the Middle East, and parts of Africa — exactly the regions where ambient conditions push diesel generator performance to its limits. Tropical data center generator selection requires engineering that goes well beyond temperate-climate practice.
Why Tropical Deratings Compound for Data Centers
Three effects stack:
- Engine derating: typically several percent reduction in engine power at 50°C vs 25°C reference (the exact figure depends on the OEM derating curve)
- Alternator derating: a further reduction in alternator output at 50°C — often in the 15-22% range above 25°C reference
- Cooling load increase: higher PUE pushes total facility electrical demand up
For full derating math and tropical specification requirements, see our tropical climate diesel generator selection guide — including 50°C-rated radiator, anti-condensation heater, dust filtration, and salt-air protection requirements for Middle East, Southeast Asian coastal, and West African deployments.
Why N+1 Redundancy Means N+0 If You Get This Wrong
This is one of the most expensive mistakes we see in tropical data center generator projects. The procurement team specifies "N+1 with 1,250 kVA units" at 25°C ISO ratings. The site is in Riyadh at 50°C peak. Each unit's usable site capacity is closer to 1,000 kVA. The N+1 redundancy that looks good on paper becomes N+0 the moment summer hits — there is little real spare capacity, just enough to carry full load with all units running.
The fix is to size the redundancy on derated capacity from the start. We discuss this in the Riyadh case study below.
Tropical Data Center-Specific Spec Additions (Typical)
- 50°C-rated radiator, often with two-stage fans
- Anti-condensation heaters on alternator (humidity damage prevention)
- Heavy-duty two-stage air filtration (cyclonic pre-cleaner + primary)
- IP55+ control enclosure with conformal-coated PCBs
- Class H alternator insulation
- Engine room HVAC sized for adequate airflow at peak ambient (often several times normal airflow)
- Temperature-compensated battery charging (and a shorter battery replacement interval — often 2 years in high-ambient sites)
Top 6 Sizing Mistakes We See in Data Center RFPs
These are the six sizing errors we see most often when reviewing data center generator RFPs and bid responses — drawn from competitive bid analysis on customer projects we've quoted and won, and the ones we've lost to undersized cheap bids that the customer later regretted.
1. Sizing on IT Load Only (Forgetting PUE)
Pattern: RFP states "1 MW IT load, please quote a 1 MW generator." Suppliers quote ~1,000 kVA / 800 kW units. The actual facility electrical demand including cooling is 1.4-1.8 MW.
Result: Generator undersized by 40-80% relative to true facility demand. Unable to pick up full facility load during utility outage; UPS batteries drain; IT load drops.
Fix: Always size against total facility electrical demand (IT load × PUE), not IT load alone.
2. Using Nameplate kVA Without Site Derating
Pattern: RFP specifies "4 × 1,250 kVA, N+1" without referencing site ambient. Site is in Dubai (50°C peak).
Result: Effective site capacity is typically 950-1,050 kVA per unit, not the 1,250 kVA nameplate. Real redundancy may be N+0 rather than N+1.
Fix: Size on derated capacity at site ambient per the OEM derating curve. Specify required output at site conditions, then back-calculate the required nameplate.
3. Confusing kW and kVA in Bid Comparison
Pattern: Two bids — one quotes "1,000 kVA" and one quotes "1,000 kW." Buyer treats them as equivalent.
Result: The 1,000 kW bid is actually a ~1,250 kVA unit (at 0.8 PF). The "cheaper" 1,000 kVA bid is meaningfully undersized in real power. Customer picks cheap, regrets later.
Fix: Require both kW and kVA in every bid, with the rated power factor stated explicitly. For a structured way to evaluate competing data center generator bids, see our 25-point generator bid comparison checklist.
4. Sizing on Average Load Instead of Peak Transient
Pattern: IT operations team provides "average IT load = 800 kW." Sizing done against this number.
Result: Peak load during workload migrations, cooling compressor restarts, or end-of-quarter compute bursts can exceed 800 kW. Generator hits thermal limits or trips on overload.
Fix: Size on peak sustained load (e.g., the 95th percentile of measured load profile), not average. Add the standard 20-25% margin on top.
5. Inadequate Step-Load Specification
Pattern: RFP says "fast start" without quantifying. Supplier delivers a unit that takes longer than the project's intended transfer window or cannot accept the required load step cleanly.
Result: During a real outage with depleted UPS batteries, IT load may drop before the generator stabilizes.
Fix: Specify the required transfer window (e.g., "Type 10" per NFPA 110), the required step-load percentage, and the voltage/frequency recovery limits. Require these in writing in the supplier's quote and verified at commissioning at site ambient.
6. Ignoring 5-Year Growth Plans
Pattern: Data center built for current IT capacity with no margin for expansion. Three years later, IT load grows 40% and generators are at or near full capacity.
Result: Sustained operation near maximum load can accelerate maintenance intervals, thermal stress, and component wear, especially in high-ambient environments. Adding new generators may require shutting down portions of the data center for retrofit — defeating the purpose of redundancy.
Fix: Build sizing on Year 5+ projected IT load, not day-one load. The 20-25% margin already includes some growth headroom — but for fast-growing facilities, plan 40%+ headroom from the start.
Pattern Across All 6 Mistakes
Every one of these mistakes results in undersized generation capacity. We almost never see data center generators meaningfully oversized in the field — but we routinely see them undersized by 30-60% relative to true facility demand at site conditions. The economic asymmetry is severe: the cost of going one frame size larger at order time is typically 8-15% of unit price; the cost of being undersized during an outage — in SLA penalties, lost revenue, reputational damage, and recovery costs — can exceed the entire generator capital cost within a single incident.
The mistakes above all apply to land-based critical-load applications—data centers, hospitals, telecom hubs. Marine vessels follow a different sizing methodology entirely: there is no utility grid to fall back on, SOLAS Chapter II-1 governs emergency power, and class society review (ABS, DNV, LR, BV, CCS) adds its own margin requirements layered on top of the base sizing math. For vessel applications, see our marine generator sizing methodology guide—the underlying load list and kVA conversion logic is closer to N+1 data center sizing than most expect, but the regulatory layer is completely different.
ASO Critical Load Reference Cases
The following three deployments illustrate how ASO Genset approaches critical-load generator sizing across data centers, hospitals, and industrial continuous-process applications. Customer names are withheld for confidentiality; project parameters are described in terms agreed for public reference.
Hero Case — Riyadh Data Center, ~5 MVA Backup Power (2023)
| Parameter | Specification |
|---|---|
| Application | Tier III-class data center backup power, low-THD requirement |
| Configuration | 1,250 kVA × 4 units (3+1 N+1), total 5,000 kVA nameplate (≈ 4,000 kW at 0.8 PF) |
| Site conditions | 50°C peak / 38°C average, 40-60% RH, fine sand exposure |
| Derating strategy | Each 1,250 kVA unit was evaluated against its derated output at 50°C peak ambient. Under the project's design assumptions, three running units could carry the required critical load at site conditions, with the fourth unit serving as true standby redundancy. Sizing was done on derated output, not nameplate. Class H alternators, 50°C-rated radiators throughout. |
| Filtration | Heavy-duty two-stage filtration, sand-trap intake louvers, monthly cleaning protocol |
| Controls | Paralleling control tuned for derated operating band, remote monitoring, transfer window verified at site ambient during commissioning |
| Status | Units consistently meet the project's critical-load requirement at peak summer; zero unplanned downtime since commissioning |
Key engineering decision: sizing the units against 50°C-derated output, not 25°C nameplate. With four units in a 3+1 N+1 configuration, three running units carry the project's full critical load at derated capacity while the fourth provides genuine standby redundancy. Sizing on nameplate rather than derated output would have left the remaining three units significantly above their rated capacity in any single-unit failure at peak summer — effectively eliminating the redundancy buffer the customer was paying for. The project was scoped on the derated number from the start, which is why N+1 is real for this customer at peak summer.
Supporting Case — Ho Chi Minh Hospital Emergency Backup (2024)
A hospital required 800 kVA × 2 in N+1 configuration with sub-10-second ATS transfer for life-safety loads (ICU, surgical, life support). Site conditions: 38°C peak, 80%+ RH year-round. Spec stack included Class H alternators with anti-condensation heaters, IP55+ control panels with conformal-coated PCBs, and a quarterly insulation resistance testing program. Two years in service: zero unplanned downtime, transfer-window performance consistently verified, no degradation in alternator insulation resistance under sustained humidity.
Why this case applies to data center sizing: hospital critical-load and data center critical-load share much of the same engineering envelope — short transfer window, N+1 redundancy on derated capacity, and tropical-spec alternator protection. The 800 kVA × 2 configuration is conceptually similar to a smaller data center critical-load deployment (e.g., a ~600 kW IT load with appropriate PUE multiplier).
Supporting Case — Jakarta Coastal Industrial Continuous-Process (2023)
A coastal industrial facility outside Jakarta required 500 kVA × 2 with automatic synchronization for continuous-process critical loads (process control, instrumentation, key pumps). Site: 40°C peak, 85%+ RH, coastal salt aerosol. Three years on-site with no enclosure corrosion through the topcoat, no measurable degradation in alternator insulation resistance, and no radiator capacity loss.
Why this case applies to data center sizing: industrial continuous-process loads have similar transient profiles to data center loads — sudden step demands from cooling and process equipment, high reliability requirements, and the same humidity-driven alternator failure mode as a tropical data center.
Specification Checklist for Data Center Generator RFQs
Use this checklist when issuing an RFP for a data center backup generator project, or when reviewing competing bids. Each line item is one we have seen cause problems when omitted or vague.
| Category | Required Specification |
|---|---|
| 1. Site declaration | Peak ambient, average ambient, altitude, humidity, coastal distance, dust exposure. Bidder must provide OEM derating curve against these conditions. |
| 2. IT load declaration | Peak IT load (kW), facility PUE, projected Year 5 load growth. |
| 3. Required output at site | Required kVA AND kW at site conditions, with PF stated explicitly. |
| 4. Redundancy topology | N, N+1, 2N, or 2N+1 — with N calculated on derated capacity at site ambient, not nameplate at 25°C. |
| 5. Transfer-window & step-load performance | Required transfer window (e.g., Type 10 per NFPA 110). Step-load percentage and voltage/frequency recovery limits, verified against the project's actual load profile, UPS behavior, and site ambient conditions during commissioning. |
| 6. Alternator specification | Class H insulation, low-reactance design for non-linear loads, harmonic distortion tolerance appropriate to the project's UPS load profile. |
| 7. Radiator / cooling | Rated for site peak ambient. Two-stage fans common on 45°C+ sites. Coated fins for coastal/dust environments. |
| 8. Air filtration | Two-stage (cyclonic pre-cleaner + primary) recommended for dusty sites. Lifetime filter cost calculation for site dust load. |
| 9. Engine block heater | Coolant block heater to maintain typical operating temperature during standby. |
| 10. ATS / transfer switch | Transfer cycle aligned with design transfer window. UL-listed or equivalent. Service entry-rated if required. |
| 11. Paralleling controls | Synchronization tuned for derated operating band. Hot-season retuning included for tropical sites where applicable. |
| 12. Remote monitoring | Real-time status, alarming, integration with data center BMS/DCIM. |
| 13. Fuel system | Day tank + bulk storage sized for project autonomy requirement (commonly 24-72 hr). Desiccant breather, water separator, biocide plan. |
| 14. Testing & documentation | Factory load test at derated power, witness FAT option, transfer-window performance verified at commissioning at site ambient. |
| 15. Code compliance | NFPA 110 (US), IEC 60364, or local AHJ requirements. End-customer project-specific specs if applicable. |
Frequently Asked Questions
How do I convert IT load (kW) to required generator kVA?
Multiply IT load (kW) by your facility PUE (typically 1.4-1.8) to get total facility electrical load. Divide by the generator power factor (typically 0.8) to get base kVA. Add a 20-25% margin for transients and growth. Then apply site derating per the OEM derating curve for ambient temperature and altitude. For a 1 MW IT load at PUE 1.5, the required N capacity typically lands around 2,250 kVA at temperate sites; at 50°C ambient, it can rise to approximately 2,750-2,900 kVA of 25°C-reference nameplate capacity. In a 3+1 N+1 configuration, this translates into four units in the 1,000 kVA class — three running units carrying the required N load, with the fourth as standby — depending on the OEM derating curve.
What is the difference between N+1 and 2N redundancy for data center generators?
N+1 means "the number of units required to carry the full load (N), plus one spare unit." If one unit fails, the remaining N units carry full load. 2N means two completely independent generator systems — each capable of carrying the full load on its own. 2N is typically 40-60% more expensive than N+1 but provides concurrent maintainability and is often required for higher Tier designs and many hyperscale deployments. Final selection depends on the overall facility availability target and the cost-of-downtime business case.
Why does my data center generator need to restore power in under 10 seconds?
UPS batteries typically carry 5-10 minutes of runtime. The generator must be online and stable well before that battery reserve is consumed. NFPA 110 defines a "Type 10" system as one that restores emergency power within 10 seconds, which is the de facto standard most North American data centers and many international tropical facilities follow. Achieving this typically requires block heaters, pre-energized excitation, governor tuning for the project's load step profile, an alternator designed for the required step-load, and an ATS sequence tuned for the design transfer window — all verified at site ambient conditions during commissioning.
What PUE should I assume when sizing for a tropical data center?
Tropical facilities in Southeast Asia and the Middle East commonly run PUE 1.6-2.0 when relying on mechanical cooling without advanced free-cooling, liquid cooling, or highly optimized airflow strategies. Highly-optimized hyperscale facilities in temperate climates can run PUE 1.1-1.3 by using free cooling. Modern colocation in moderate climates typically runs 1.4-1.6. Use your facility's actual design PUE, not the industry average, and verify with your mechanical engineering team before sizing the generator.
Can I use the same generator for both hospital and data center critical-load applications?
Engineering-wise, yes — the requirements overlap heavily: short transfer window, N+1 redundancy, NFPA 110 compliance (where applicable), and high-quality power delivery to non-linear loads. The differences are mostly in spec details: hospitals emphasize voltage stability for medical imaging equipment, data centers emphasize harmonic tolerance for UPS rectifiers, and tropical data centers add 50°C ambient and humidity protection. A properly engineered critical-load generator can serve either application with appropriate spec selection.
How much fuel storage do I need for a data center backup generator?
Industry minimum is often 24 hours of full-load runtime; Tier III-class data centers typically specify 48-72 hours; hyperscale and financial applications often require 96+ hours or contracted fuel delivery within a defined window. As a rough planning estimate, a 5 MVA generator fleet running near 0.8 PF may consume roughly 1,000-1,200 liters per hour at high load, depending on engine model, load factor, fuel quality, and site conditions. A 72-hour reserve at that load level would require approximately 72,000-86,000 liters of diesel storage. Most modern facilities right-size storage based on the realistic outage scenario (typically 50-70% of full-load consumption averaged over outage duration) rather than worst-case continuous full-load.
What happens if I undersize a data center generator?
The generator typically works fine during commissioning and normal weekly tests at light load. The problem appears during an actual utility outage at peak summer or peak workload, when the unit cannot accept full block load within the design transfer window. UPS batteries deplete before the generator stabilizes, and IT load may drop — the exact outcome the backup investment was meant to prevent. The financial consequences vary by application, but SLA penalties, lost revenue, recovery costs, and reputational damage can exceed the entire generator capital cost within a single major incident.
Does this guide apply to marine generators on ships?
No—marine generator sizing follows fundamentally different rules. Vessels operate as closed electrical islands (no utility grid to fall back on), SOLAS Chapter II-1 governs emergency generator requirements, and the "big 5" class societies (ABS, DNV, Lloyd's Register, Bureau Veritas, CCS) each apply their own review factors to the sizing math. The load classification itself splits into hotel / propulsion auxiliary / mission-specific / emergency, which doesn't map cleanly to the data center IT-load + PUE model. For complete marine methodology, see our marine generator sizing guide.
Free Download: Data Center Generator Sizing Calculator (Excel)
Built from ASO's deployments across Saudi Arabia, Vietnam, and Indonesia
- IT load → total facility kVA calculator with PUE and PF inputs
- Site derating calculator (temperature + altitude) with OEM curve placeholder
- Redundancy topology comparison (N, N+1, 2N, 2N+1) with capital cost ratios
- 15-point specification checklist
- Top 6 sizing mistakes diagnostic reference
- RFP boilerplate language for kW/kVA/site-derating clarity
No spam. Excel emailed instantly. Use it with any supplier, including ASO.
Related Reading
- Tropical Climate Diesel Generator Selection Guide — Detailed engineering for 50°C+ ambient, humidity, salt air, dust environments common to Middle East and Southeast Asian data centers.
- NFPA 110 Generator Testing Requirements Guide — Testing and maintenance regime for critical-load diesel emergency power systems serving data centers and healthcare facilities.
- 25-Point Diesel Generator Bid Comparison Checklist — Commercial framework for evaluating competing generator bids side-by-side, including data center-specific RFP requirements.
- ABS vs DNV vs CCS Marine Generator Classification — For shipowners and offshore platform operators where data center-style critical-load requirements meet marine class society certification.
Contact
Specifying or Reviewing a Data Center Generator Project?
ASO Genset designs and manufactures diesel generators for critical-load applications across data centers, hospitals, industrial continuous-process, and marine offshore deployments — including the 1,250 kVA × 4 Tier III-class data center backup project in Riyadh referenced in this guide. We build to international critical-load specifications: Class H alternators with low-reactance design for non-linear loads, transfer-window performance verified at site ambient, 50°C-rated cooling for tropical sites, and N+1 / 2N redundancy sized on derated capacity rather than nameplate. Our quotes itemize critical-load features as line items — alternator class, cooling rating, ATS specification, monitoring — so you can see exactly what you are paying for and compare us against any other bidder using the 15-point checklist above.
Whether you are scoping a new data center, reviewing competitor bids on a colocation build-out, or troubleshooting an existing critical-load installation that is not meeting its transfer-window target, send us your site conditions (peak ambient, altitude, PUE) and load profile (IT load kW, projected Year 5 growth, redundancy topology required). We will respond with a sizing calculation, recommended specification, and line-item quotation broken down by unit — typically within two business days.
