QA/QC in data center construction projects.
Data center projects fail QA/QC the same way, in every market — a strong design, a competent GC, aggressive schedules, and a commissioning programme squeezed into the last six weeks. This is a practical framework for owner-side QA/QC on hyperscale, colocation and enterprise builds across the US and Canada: what to inspect, when to inspect it, and what evidence has to be in the turnover binder before you accept the site.
Data centers are unforgiving. A missed MEP acceptance test, an untested ATS transition, an unlogged Level 4 fault scenario — these do not get discovered at handover. They get discovered when a customer's workload trips at 3am and the SLA credits start compounding.
The programmes that turn over cleanly share four traits: a schedule of quality events written into the master schedule (not stapled on at 80% complete), factory witness testing that has actually been witnessed, five levels of commissioning executed in sequence with owner sign-off at each gate, and a turnover package that resolves every ITP row to retrievable, indexed evidence. This guide walks through the framework we run for owners across the US and Canada.
- 01Data center QA/QC lives on Levels 1–5 commissioning — not on a stand-alone inspection programme.
- 02Bake the schedule of quality events into the P6 master schedule at design freeze; retrofitting it after 60% complete never recovers.
- 03Factory witness tests (FWT) at OEM works catch the defects site testing cannot — skip them and you accept every defect at receiving.
- 04Level 5 Integrated Systems Testing (IST) is the acceptance test — everything before it is preparation, everything after it is punch list.
- 05Turnover is owned by the owner's commissioning agent, not the GC — separation of duties is what makes the acceptance credible.
What makes data center QA/QC unlike other MEP construction
Data centers concentrate risk in a way no other commercial construction category does. A hospital MEP failure is contained by redundant clinical protocols; a manufacturing plant failure loses a shift; a data center failure loses a Fortune 500 tenant's production workload — and the SLA credits, tenant chargebacks and reputational damage that come with it. The QA/QC programme has to be calibrated to that risk asymmetry, not to the construction cost.
Three properties define the difference. First, systems interdependence: cooling depends on power depends on controls, and every fault scenario has to be tested against the actual as-built topology, not the design intent. Second, live-load acceptance: unlike a warehouse or office fit-out, a data center is accepted while it is powered and, on many projects, while adjacent halls are already carrying tenant load. Third, the acceptance criteria are not local building code — they are the design's own reliability targets (Tier III concurrent maintainability, Tier IV fault tolerance) verified through end-to-end failure testing, not visual inspection.
Five workstreams that make data center QA/QC work
These workstreams run in parallel from design freeze to owner acceptance. Skip any one and the programme collapses into inspection-by-exception at handover.
1. Design QA & constructability review
Independent review of drawings, single-line diagrams, SOOs and control narratives before IFC. Catches redundancy path errors, mis-sized ATS, missing bypass provisions and control philosophies that will not survive Level 5 IST.
2. Schedule of quality events (SoQE)
A living register — tied to the P6 master schedule — of every FAT, FWT, ITP hold point, Level 1–5 activity and inspection required, sequenced against the construction milestones. Updated weekly; reviewed at project controls.
3. Factory & site inspection (ITPs, FWT, SAT)
Owner-side ITPs for switchgear, UPS, chillers, generators, PDUs, RPPs and busway. Factory Witness Testing (FWT) at OEM works before shipment; Site Acceptance Testing (SAT) at receiving; installation inspection to code and specification.
4. Commissioning Levels 1–5
The core of the programme: L1 pre-installation checks, L2 installation verification, L3 pre-functional & start-up, L4 functional performance testing (FPT), L5 Integrated Systems Testing (IST). Each level gated by owner sign-off.
5. Turnover, training & warranty
As-built drawings, O&M manuals, warranty registers, spare parts inventory, redline drawings reconciled, MOP/SOP library, operator training records, and a live tenant onboarding runbook — assembled continuously, not at handover.
6. Governance & independent oversight
Weekly quality steering with GC, MEP subs, Cx agent, owner's engineer. RACI for every ITP hold. Independent third-party audit of the Cx agent's own records on a rolling basis — nobody grades their own homework.
Levels 1–5 explained: what happens at each gate
Level definitions vary across the industry. The framework below matches how ASHRAE Guideline 0, BCxA best practice and the major hyperscale owners actually run programmes in the US and Canada — with sign-off at every gate.
| Level | Scope | Typical activities | Owner sign-off gate |
|---|---|---|---|
| Level 1 — Factory | OEM works testing before shipment. | FAT scripts for switchgear, UPS, chillers, generators; heat runs; short-circuit and dielectric tests; software burn-in; witness by owner Cx agent or delegated TPI. | FAT report signed and lot-released before shipment. |
| Level 2 — Installation | Verify equipment installed per approved drawings, specifications and manufacturer's instructions. | Anchorage, torque, seismic bracing, cable pulling tension logs, terminations, labelling, clearances, arc-flash boundaries, fireproofing at penetrations. | Installation punch clear; ready for start-up. |
| Level 3 — Pre-functional / start-up | Energise, calibrate, tune, verify safeties. | Insulation resistance tests, phase rotation, breaker settings loaded per short-circuit study, controls point-to-point, BMS graphics verified, chilled water flushing and passivation, refrigerant leak checks. | Pre-functional checklists signed; ready for FPT. |
| Level 4 — Functional performance (FPT) | Test each system's designed sequences of operation individually. | ATS transfers, generator loading and unloading, UPS transfers to bypass and back, chiller staging, DX changeover, VFD ramp response, alarm generation and reset. | FPT scripts signed; deficiencies logged. |
| Level 5 — Integrated Systems Testing (IST) | Test the whole facility as a system under simulated failure scenarios and full design load. | Utility failure, generator start and pickup, cooling loss and recovery, dual-path failure, N+1 to N transition, black-building start, fire alarm interfaces, sustained load bank testing (often 24–96 hours). | IST acceptance report; Certificate of Readiness. |
The ITP topology every data center project needs
A single project-wide ITP does not exist for data centers — the equipment count and vendor list is too large. Instead, run a hierarchy of ITPs indexed by a master ITP register.
- Master project ITP register — one row per equipment package, subcontractor and building system, indexed to sub-ITPs.
- Equipment ITPs (per OEM package) — switchgear, UPS, PDU, RPP, chiller, CRAH/CRAC, generator, ATS, static switch, busway. Includes FAT, FWT, SAT and site installation.
- Discipline ITPs (per subcontractor) — electrical rough-in, medium-voltage cable, grounding & bonding, chilled water piping, refrigerant piping, BMS/EPMS controls, fire suppression, life safety.
- Envelope & structural ITPs — slab, waterproofing, roofing, fire-rated barriers, containment integrity, raised floor & tile loading verification.
- Commissioning ITPs — one per Level (L1–L5), each with pass/fail acceptance criteria referenced to the specification.
“The turnover binder does not prove the building works. It proves someone verified the building works, with evidence, on a specific date. Get that distinction right and the entire QA/QC programme falls into place.”
QA/QC coverage matrix for critical MEP equipment
Every project varies; treat this as the minimum viable coverage per critical equipment category. Waive a row only with a documented risk acceptance from the owner.
| Equipment | Design review | FAT / FWT | SAT / L2–L3 | L4 FPT | L5 IST role |
|---|---|---|---|---|---|
| Medium-voltage switchgear | Short-circuit study, arc-flash, protection coordination | Full FWT witnessed at OEM: primary injection, dielectric, mechanical operation | Insulation resistance, contact resistance, relay settings loaded per study | Trip curves verified, interlocks tested | Fault clearing under simulated utility loss |
| Uninterruptible power supply (UPS) | Topology, redundancy path, battery runtime calc | FWT with full battery discharge, transfer to bypass, THD readings | String voltage, IR test on batteries, breaker settings | Transfer to bypass and back under load, harmonic verification | Loss-of-utility ride-through; battery run to setpoint |
| Standby generator | kW/kVA sizing, block-load acceptance, fuel autonomy | Load-bank FAT at OEM (min. 4 hours full load), sound and emissions | Voltage, frequency, oil, coolant, cranking, transfer verification | Loaded start within design window (typically 10 sec), paralleling | Facility-wide black-building start with full load pickup |
| Automatic transfer switch (ATS) | Open/closed/soft-transition selection, bypass provisions | FWT closed-transition timing to spec | Timing curves, seating pressure, phase-check on both sources | Transfer on simulated source loss, retransfer with time delay | Multiple simultaneous transfers across the facility |
| Chiller plant | Redundancy (N+1 / 2N), chilled water reset strategy | Chiller FAT: performance curve, sound, vibration, refrigerant charge | Flushing, passivation, pressure test, refrigerant leak test | Staging, unloading, low-load operation, alarm generation | Loss of chiller, staging recovery, thermal ride-through |
| CRAH / CRAC units | Airflow strategy, containment integration | AHRI certified performance, EC fan curves | Belt tension, condensate, filters, VFD parameters | Setpoint tracking, differential pressure control | Loss of unit, recovery within thermal envelope |
| BMS / EPMS controls | Sequences of operation, alarm philosophy, cybersecurity posture | Software FAT with simulated I/O | Point-to-point verification, graphics, trend logs, historian | Every sequence tested, every alarm generated and cleared | Whole-facility sequences under scripted failure scenarios |
| Fire alarm & suppression | Detection strategy, agent selection, interface to shutdowns | Detector calibration certificates, agent quantity verification | AHJ inspection, matrix verification, room integrity (fan test) | Every device tested; every interface actuated | Facility-wide matrix tested during IST |
Integrated Systems Testing: what a credible IST looks like
IST is the only test in the programme that verifies the facility as a system. Everything else is component-level. A credible IST is scripted end-to-end, executed against sustained design load (typically load banks representing the design IT load plus cooling loads), witnessed by the owner, the tenant (if named) and the AHJ where required, and repeated for every major failure scenario in the design basis.
The single most common IST failure is not a technical failure — it is a scoping failure. The scripts are written by the Cx agent, reviewed by the owner's engineer, and approved by the owner before the test window opens. If the scripts arrive on the morning of the test, the test is not credible and the acceptance is not defensible.
Minimum IST script coverage
The turnover package that survives tenant acceptance
The turnover binder is the accepted evidence layer. Every entry is dated, revision-controlled, and traceable to the ITP row or Cx script it satisfies.
As-built documentation
Redlined drawings reconciled to CAD/Revit, single-line diagrams updated to as-built, coordination models delivered, control narratives revised.
Test reports & evidence
FAT, FWT, SAT, L1–L5 scripts with pass/fail evidence, deficiencies logged, load bank logs, thermal images, IR reports, insulation & continuity tests.
O&M manuals
OEM manuals per equipment, indexed, PDF-searchable; maintenance schedules per manufacturer; consolidated spare parts list with reorder codes.
MOP/SOP library
Method of Procedure and Standard Operating Procedure per critical activity — utility switching, generator paralleling, UPS bypass, chiller changeover — approved by the owner's operations team.
Warranty register
One row per warranty (equipment, workmanship, roof, structural), effective date, expiry, contact, renewal terms. Living document owned by owner ops.
Training records
Operator training records per system, attendance logs, competency assessments, hands-on drills witnessed by the Cx agent.
Punch list closure
Every open punch item resolved with evidence — photo, retest, signed acceptance — before Certificate of Substantial Completion is issued.
Tenant onboarding runbook
Escort procedures, badging, MOP approval flow, change management interface between owner ops and tenants — ready before first tenant walkthrough.
Certifications & permits
Building permit sign-offs, AHJ acceptances, fire marshal approvals, electrical certificate, elevator, life-safety certificates.
Where data center QA/QC programmes most often fail
The Cx agent reports to the GC
The commissioning agent must be contracted by the owner, not the general contractor. When the GC pays the Cx bill, the deficiency log gets short before the sign-off gets long.
FWT waived to protect schedule
'The vendor's FAT is comprehensive, we don't need to witness' is the sentence you hear ninety days before a UPS lands with a defective battery string. Do not waive FWT on any critical MEP equipment.
IST compressed into the last week
IST requires load banks, tenant coordination, AHJ presence and days of continuous testing. Compressing it produces scripts that skip failure scenarios. Book the IST window at design freeze.
As-built drawings 'to follow'
Turnover binders shipped with 'as-built drawings to follow' means the operator inherits a facility with no accurate topology. Withhold retention until as-builts are reconciled.
MOPs written after the incident
MOP/SOP libraries built retrospectively are libraries of what was tried, not what was proven. Approve the MOP set before Level 5 begins so the IST executes against them.
Punch closed by the same party that opened it
Separation of duties applies here too. Punch closure verified by a party independent of the one that executed the corrective action — otherwise the punch list becomes a to-do list.
US and Canadian codes and standards that shape data center QA/QC
The design and acceptance framework is code-driven in both jurisdictions. QA/QC scripts should cite the specific edition and clause — never 'per code' — and the project specification where it is more stringent.
- NFPA 70 (National Electrical Code) and CSA C22.1 (Canadian Electrical Code) — installation, grounding, arc-flash labelling.
- NFPA 70E — arc-flash and electrical safe work practices during commissioning and operations.
- NFPA 72 — fire alarm systems; NFPA 75 & NFPA 76 — protection of information technology equipment and telecommunications facilities.
- NFPA 2001 — clean agent extinguishing systems; NFPA 25 — inspection, testing and maintenance of water-based systems.
- ASHRAE Guideline 0 and ASHRAE 90.4 — the commissioning process and energy standard for data centers.
- ASHRAE TC 9.9 thermal guidelines — envelope for IT equipment intake temperatures during IST.
- Uptime Institute Tier Certification (Design, Constructed Facility, Operational Sustainability) — the acceptance model most owners align to; drives IST script coverage.
- TIA-942 — data center telecommunications infrastructure standard; often referenced for pathway, spaces and cabling QA.
- Provincial and state AHJ requirements — building, fire and electrical permits and inspections in each jurisdiction.
- Local utility interconnection standards — service entrance switchgear, metering, and testing requirements per utility.
Independent QA/QC oversight for data center projects.
Owner-side QA/QC on a data center is not a checkbox activity. It is the mechanism that catches the missing bracket in a switchgear lineup, the mis-tuned VFD parameter on a CRAH unit, the incorrectly loaded relay setting that would clear a downstream fault too slowly. Every one of those catches — before energisation, before IST, before the first tenant — pays for the entire QA/QC programme many times over.
Independent third-party inspectors and lead auditors sit outside the GC's schedule pressure and outside the Cx agent's contract with the delivery team. We prepare and review ITPs, witness factory and site acceptance testing, audit the Cx agent's records against the specification, and verify the turnover package survives tenant acceptance. That is what defensible owner-side QA/QC looks like in a data center build.
Questions we get on this topic
What is QA/QC in data center construction?
QA/QC (Quality Assurance / Quality Control) in data center construction is the integrated set of design reviews, inspection and test plans, factory and site acceptance testing, five levels of commissioning (L1 factory through L5 Integrated Systems Testing), and turnover verification that confirms the facility meets the design, code and reliability requirements before tenant acceptance.
What is the difference between commissioning and QA/QC in a data center?
Commissioning (Cx) is the structured verification that systems perform per the design intent — Levels 1 through 5 of testing. QA/QC is the wider programme that includes commissioning plus design QA, ITPs at equipment and discipline level, factory witness testing, installation inspection, punch closure and turnover documentation. Cx is the acceptance engine; QA/QC is the framework that ensures Cx has something credible to accept.
What are Levels 1 to 5 of data center commissioning?
Level 1 — Factory testing (FAT) at the OEM works before shipment. Level 2 — Installation verification against approved drawings and specifications. Level 3 — Pre-functional and start-up: energise, calibrate, verify safeties. Level 4 — Functional Performance Testing (FPT): each system tested against its sequence of operation individually. Level 5 — Integrated Systems Testing (IST): the whole facility tested as a system under failure scenarios and sustained design load. Owner sign-off is required at each level gate.
What is Integrated Systems Testing (IST) in a data center?
IST is the acceptance test for the facility as a system. It executes scripted failure scenarios — loss of utility, loss of chiller, loss of UPS module, black-building start — against sustained design load (typically load banks) with continuous data logging. It is the only test in the programme that verifies the interactions between systems, not the systems in isolation. IST typically runs 24 to 96 hours and is witnessed by the owner, the tenant where named, and the AHJ where required.
Who owns QA/QC on a data center project — the GC or the owner?
The owner. The GC executes their own QA/QC as part of construction management, but the acceptance framework — ITP review, factory witness testing, commissioning oversight, turnover verification — must sit with the owner or the owner's engineer. The commissioning agent is contracted by the owner, not the GC, so the incentive to close deficiencies before sign-off is not compromised.
How much should QA/QC cost on a data center project?
For hyperscale and colocation builds in the US and Canada, budget 4–6% of MEP cost for combined third-party commissioning, owner's engineer QA and independent inspection. Programmes below 3% consistently miss critical defects that show up post-turnover; programmes above 6% typically indicate scope duplication rather than better coverage.
How does Uptime Institute Tier Certification affect QA/QC scope?
Tier Certification sets the design reliability target — Tier III requires concurrent maintainability, Tier IV requires fault tolerance — and drives the IST script coverage needed to verify the target. Tier IV programmes require significantly more dual-path failure scenarios and simultaneous-failure testing than Tier III, which expands Level 5 script length and load bank duration.
When should the QA/QC programme start on a data center project?
At design freeze, not at construction start. The Schedule of Quality Events (SoQE) — every FAT, FWT, ITP hold point, L1–L5 activity — is baked into the P6 master schedule as milestones with predecessors and durations. Programmes that start QA/QC at 30–50% construction complete are already recovering; programmes that start after 60% are producing an audit trail, not preventing defects.
What is the difference between FAT, FWT and SAT?
Factory Acceptance Testing (FAT) is the OEM's own test at their works. Factory Witness Testing (FWT) is the same test with the owner or owner's delegate present, witnessing and signing. Site Acceptance Testing (SAT) is testing at the receiving site after installation and start-up. FAT alone accepts the vendor's word; FWT plus SAT is the defensible standard for critical MEP equipment on a data center.
Does QA/QC apply to colocation and enterprise data centers the same way as hyperscale?
The framework is the same; the scale and MOP/SOP maturity differ. Hyperscale owners typically bring standard commissioning specifications, MOP libraries and Cx scripts across projects. Colocation and enterprise owners often build these from scratch on the first facility, which is where an independent QA/QC partner adds the most value — the framework is portable across projects even when the owner is not repeating a build.
Planning a data center build in the US or Canada?
Our QA/QC leads and commissioning authorities support owners on hyperscale, colocation and enterprise data center projects — from design QA and ITPs through Levels 1–5 to owner acceptance and turnover.