A data centre build needs megawatt-class power months before its permanent electrical plant is released — for fit-out, for starting individual systems, and above all for staged commissioning under load. Rented, prime-rated diesel generators carry that load while the utility feed is pending and the permanent gensets are still under test themselves.
TL;DR
- Commissioning proves a data centre under load before any IT arrives — load banks stand in for the servers, and the power behind the test is very often temporary generation
- Industry practice stages the work in five levels: factory tests (L1) through integrated systems testing, IST (L5), where everything runs together and failures are simulated
- Resistive load banks prove the engine (kW); reactive banks prove the alternator and transient response — mission-critical sites are exactly where that difference matters
- Commissioning load is variable and partly unknown — prime-rating territory, not standby
- Above any single set’s capacity, the answer is a synchronised multi-set bank — the fleet runs 25 to 1250 kVA, with the 1010-and-above band positioned for data-grade loads and synchronised banks
Why a data centre needs power before it has power
The paradox of a data centre build is that the thing being constructed is the power system — utility intake, transformers, switchboards, permanent gensets, UPS, distribution. It cannot power its own proof. Yet from mid-construction onward the site consumes serious electricity: fit-out trades, BMS and fire-system installation, first runs of chillers and air handlers, and then the commissioning programme itself. Three things push that load onto temporary generation:
- The utility feed usually lands late. Substation works, metering and release formalities tend to conclude near the end of the programme — while commissioning needs sustained load months earlier.
- The permanent gensets are themselves items under test. They arrive with their own factory and site acceptance steps, and running the site’s construction load on them consumes engine hours and muddies the acceptance baseline before handover.
- Commissioning deliberately breaks the power chain. Blackout simulations and changeover tests take the plant under test dark on purpose — and the trades still working in the building need a supply that stays up while that happens.
So a temporary DG installation — sets, cabling, distribution and changeover — becomes part of the construction infrastructure, sized to the stage schedule rather than to the finished facility.
The five commissioning levels, briefly
Data-centre commissioning is commonly staged in five levels. The definitions below follow CxPlanner’s published commissioning guide (terminology and tag colours vary a little between commissioning agents, but the structure is standard industry practice):
| Level | Common tag | What happens |
|---|---|---|
| L1 | Red | Factory testing of components and systems before delivery to site |
| L2 | Yellow | Delivery and pre-installation checks on site — inspection, specification verification |
| L3 | Green | Pre-commissioning: functional checks and initial start-up of individual systems |
| L4 | Blue | Functional performance testing of individual components and systems against specification |
| L5 | White | Integrated systems testing (IST): everything operating together, including simulated failure scenarios such as blackout tests |
Read that table as a load curve. L1 happens at the factory; L2 needs little more than tools and lighting. L3 is where real electrical load appears on site, as individual systems start up for the first time. L4 sustains it, system by system, for hours at a stretch. And L5 is the peak: the whole facility running at design load simultaneously — chillers, pumps, air handling, UPS, controls — while the commissioning agent pulls breakers to prove the site rides through. Because the IT equipment is not installed yet, its heat and electrical demand are simulated by load banks.
Whoever supplies temporary power for this programme is therefore not quoting “a generator for a construction site” — they are quoting a staged capacity plan that ramps from fit-out load to full design load, with witnessed test windows and zero tolerance for a mid-test trip.
What load bank testing actually proves
A load bank is a machine built to consume electricity in a controlled, measurable way — the stand-in for the servers, and the instrument that proves the electrical chain. The distinction that matters is between resistive and reactive load — drawn here per Consulting-Specifying Engineer’s genset-commissioning guidance and standard load-bank practice:
| Load bank type | What it applies | What it proves |
|---|---|---|
| Resistive | Real power (kW) at unity power factor | The engine (prime mover) side, plus load-sharing and load add/shed controls |
| Reactive (with resistive) | kW plus kVAR — operation at the machine’s rated power factor | The alternator, voltage regulation and transient response under loads that resemble the real installation |
The gap between the two columns is the gap between “the engine ran fine” and “the site will ride through”. (A sustained full-load run also works the engine’s fuel, exhaust and cooling systems — general practice, not the cited guidance.) A genset is nameplated in kVA at 0.8 power factor, so a resistive-only test at unity power factor never exercises the alternator at its rated duty. And the transient story is worse than intuition suggests: per the same guidance, reactive testing produces roughly 25% greater voltage dips than resistive-only testing — dips that expose sensitive solid-state controls and power supplies a resistive test would have waved through. That is why the article lists data centres, healthcare and life-safety systems as the applications where reactive testing matters most.
One thing worth being straight about: load banks are normally specified and brought by the commissioning agent or a specialist test house — they are the measuring instrument, and independence is part of their value. Our side of the ecosystem is the generation feeding the test, the panels and cabling that connect it, and the discipline of running rented engines at sustained high load for witnessed windows. Load testing is not foreign territory for a fleet that load-tests and recommissions every machine after major overhaul work as standard practice.
Commissioning load is prime-rating territory
Which rating should the rented sets carry? Commissioning duty is load that varies hour to hour and stage to stage, for total hours nobody can promise in advance, across weeks of sustained daily running. That profile fails the standby (ESP) definition twice over — ESP is capped at 200 hours a year and reserved for genuine utility failure. Prime (PRP) — variable load, unlimited hours — is the only rating that honestly covers a commissioning programme, which is one more reason this fleet quotes prime power on every capacity page.
Two sizing cautions carry extra weight here. First, step loads: commissioning applies load in deliberate blocks — a load bank stepping up in increments, a chiller starting direct-on-line — so the sets need transient headroom, not just enough kVA on paper. Second, the load is electronically ugly: UPS rectifiers and VFDs dominate a data centre’s demand, and the sizing guide’s warning about UPS- and VFD-heavy sites — harmonics heating the alternator — applies in full. Send the commissioning agent’s load schedule with the enquiry, and get the sizing answered before mobilisation rather than after the first tripped test window.
MW-scale loads run on synchronised banks
The largest single set on the rental ladder is 1250 kVA. Data-centre commissioning loads routinely exceed any single machine — which is why the fleet’s top band, 1010 kVA and above, is positioned for exactly this duty: data-grade loads and synchronised banks. Sets paralleled onto one bus behave as one large, redundant machine: capacity scales in steps, sets are added or dropped as the stage load ramps, and a single-engine problem during a witnessed test costs redundancy rather than the test.
Synchronisation is panel engineering as much as genset engineering, and it is a delivered capability here, not a brochure line. The UPS and LT panels practice supplies and integrates synchronising and PCC panels with Woodward or equivalent controllers, AMF and changeover panels, and integration with existing gensets and switchgear of any brand. The delivered proof: at a Jabalpur bottling plant, a new 1500 kVA set was paralleled with the plant’s existing 1250 and 500 kVA units — three generators of different vintages sharing one bus — and at a food plant, a 1010 kVA Cummins was synchronised to existing CAT sets. Every rental, at any scale, includes delivery, commissioning and scheduled B-checks as standard scope.
The discipline IT campuses actually test
We will not dress up our project list: the published record here is IT-campus and institutional power work, not hyperscale data-centre construction. But the discipline those builds demand — staged work on live critical infrastructure, cover verified before anything is touched, load testing before anything is handed back — is what the record shows. At IBM Bangalore, four 1010 kVA Cummins sets protecting a live campus went through D-check top overhauls one at a time: each machine released only after the campus’s remaining standby cover was verified, and every set returned through load testing into auto standby. The campus was never left unprotected. Commissioning programmes run on method statements and stage gates; a power vendor whose habits are verify-cover-first and load-test-last fits that world without retraining.
If your build is in the south, the Hyderabad page carries the published track record there — anchored by a 1500 kVA turnkey DG installation with a 30-metre exhaust for Central Bank of India at Koti, delivered as supply, installation, testing and commissioning under one order.
Specifying temporary power for a commissioning programme
The enquiry that gets a useful quote carries six things:
- The stage schedule — which commissioning levels, in which months, with the witnessed-test windows
- Peak load and largest step — the commissioning agent’s load schedule in kW and kVA, and the biggest single block applied
- Run profile — hours per day and total weeks, so the duty lands correctly on the prime rating
- Test participation — whether the sets feed the load banks, carry the construction supply during blackout tests, or both
- Site interfaces — cable runs, panel and changeover requirements, any existing switchgear to integrate with
- Who brings the load banks — usually the commissioning agent; the generation plan is built around their test plan, not the other way round
Send that on WhatsApp and sizing comes back the same day. Browse the ratings — every set from 25 to 1250 kVA with its full Cummins specification — on the rental fleet page.
Sources: CxPlanner, “Data center commissioning: The 5 levels of testing” — https://cxplanner.com/data-centers/resources/data-centers-level-testing — for the L1–L5 stage definitions, tag colours and the L5 IST description including blackout tests. Consulting-Specifying Engineer, “Commissioning, testing gensets using resistive/reactive load banks” (HPS Loadbanks, 13 November 2013) — https://www.csemag.com/commissioning-testing-gensets-using-resistive-reactive-load-banks/ — for what resistive and reactive load banks each prove, the ~25% transient-dip comparison, and the mission-critical applications note. Rating definitions (ESP 200-hour cap, PRP variable-load unlimited hours) per ISO 8528-1 as published in our prime vs standby guide. Fleet range, band positioning, panel scope and all project facts — IBM Bangalore, Jabalpur, the food-plant synchronisation and Central Bank of India Koti — are as published on this site’s rental, panels, projects and location pages. Commissioning-level terminology varies between agents; your commissioning plan governs.