Field testing, verification and operations for liquid cooling
In a fifteen-year FM loss review cited by Swiss Re Institute, about 24 % of total data-centre loss cost was liquid-related (all liquid categories, not a direct-to-chip failure rate). This is the service that addresses that risk on site, across North America: written criteria before the work, records at the end, and a maintenance programme for the cooling loop.
A liquid-cooled hall can be built by five suppliers and commissioned by none of them. Every party can prove its own segment and the seam between them remains untested. This service takes the seam: the loops, manifolds, hoses and connectors from the facility water interface to the serviceable unit, and the records that say who tested what, against which criterion, and where the evidence sits.
We work on systems we supplied and, more often, on systems we did not. A multi-vendor site is the one that most needs somebody accountable for the interfaces.
Five lines, engageable separately
Commissioning
Four gates, criteria agreed before the work, a turnover package at the end. Engageable gate by gate, including for loops somebody else built.
Annual operations
Monthly pipework inspection, fluid sampling against the commissioning baseline, filter differential trending, quick-disconnect cycle records, annual re-verification.
Repair and response
Call-out, labour and materials. Live-system repair is arranged through pre-qualified specialists under framework agreement, because the capability is scarce and cannot be found on the day.
Factory and receiving inspection
Gate one executed at the manufacturer, including in China: pressure test, cleanliness release, torque records, quick-disconnect register, document check.
Advisory and records
Rebuilding a turnover package for a system already in service, gap assessment against acceptance criteria and insurer requirements, subcontractor qualification review.
Commissioning: four gates
A gate is a point of no return. Past it, the cost of going back rises sharply. The acceptance criteria for each gate are written and approved before that stage of work begins, never after it.
Before release from the factory
- Pressure test of every prefabricated assembly, with serials
- Cleanliness release of wetted surfaces, verified by sample
- Torque record for every threaded and flanged joint, with the wrench calibration certificate
- Quick-disconnect register opened, with the mate cycles made during factory testing recorded
Before the loop is filled
- Segmented pressure test, zone by zone, with the pressure–time record
- Flush to a stated end point, measured rather than observed
- Filter differential baseline recorded before fill
- Fluid concentration, pH, inhibitor and conductivity verified against specification
Before compute is energised
- Leak detection and automatic isolation proven by fault injection, zone by zone
- Flow and differential measured at every rack, not only the one nearest the pump
- CDU failover under load: single pump, single unit, primary-side interruption
- Telemetry proven to reach the system that has to act on it
Before handover
- Turnover package assembled and indexed against the boundary structure
- Operating baselines recorded at a stated load
- Open items listed, owned and dated, accepted in writing by the owner
- Warranty boundary written per interface, with a start date
Operations: three tiers
The lowest tier is not a marketing floor. It is built around the practices property insurers publish for liquid-cooled halls: periodic inspection of the pipework in the equipment room, monitoring of coolant quality against its design range, and an operating team trained and authorised to isolate a leak at the nearest valve. Proposed frequencies are confirmed against OEM requirements, site risk and the contract or insurance conditions actually adopted; they are not universal insurer requirements.
Inspection, sampling and records
- Monthly pipework inspection to the CDU or heat exchanger connection
- Quarterly fluid sampling and laboratory analysis
- Filter differential logged against baseline
- Quick-disconnect register maintained
- Annual report
Routine care plus annual functional re-verification
- Everything in Tier 1
- Annual leak detection and isolation proven by fault injection
- Annual isolation drill with the operating team
- Annual flow balance re-measurement
- Root cause analysis and record update after every event
For high-density, high-value or multi-vendor sites
- Everything in Tier 2
- Annual re-verification: sampled repeat of the Gate 3 tests
- Baseline update and change management support
- Priority scheduling
- Scarce-capability partners held under framework agreement
Attendance commitments are priced separately from the tier. Four hours to site and next business day to site are different businesses, and bundling them hides which one you are buying.
How we grade a fault (NextGenergy service levels L0–L3)
Liquid cooling cannot borrow the IT severity scale. The question is not how many nodes are affected. It is whether the system is causing, or about to cause, asset damage.
| Level | Definition | First principle |
|---|---|---|
| L0 | Liquid is escaping in the white space and automatic isolation has not contained it | Containment unsuccessful: activate the approved emergency procedure. Authorised isolation and electrical actions come before diagnosis. |
| L1 | A leak has been detected and isolated, or a CDU has failed and the loop has lost cooling | Protect the IT through the approved thermal response, then restore cooling to the loop. |
| L2 | A single point is weeping without spreading, a standby pump has taken over, or a rack is outside the flow tolerance | Assess the impact and schedule corrective work within the approved operating envelope. |
| L3 | Fluid chemistry drifting from baseline, filter differential climbing, a quick disconnect approaching its cycle limit | Investigate the trend and plan the intervention before limits are exceeded. |
These are NextGenergy's service levels, not an industry standard; at the start of each project they are mapped level by level to the owner's own incident classes. Response and attendance times are agreed per contract against this scale, and are stated in the service agreement rather than advertised.
How the work is delivered
We hold the criteria, the coordination and the records. We bring the acceptance criteria, drawn from OCP, ASHRAE and the equipment manufacturers' requirements and set out in writing for the owner's engineer to approve, then hold every result against them. We sequence the work and assemble the turnover package. The owner or their authorised representative accepts the delivery.
We take projects anywhere in North America. Projects are coordinated from Toronto. For each project we engage qualified, licensed local partners for the site work, and attendance and response times are written into the contract rather than advertised. Records review, data review and procedures are supported remotely.
Site work is performed by qualified partners. Wherever the site is, we qualify against local certified pipework capability rather than duplicate it: certificate scope checked for shop versus field, code class checked against the one the project actually falls under, welder qualifications checked against the people attending rather than the roster, and insurance checked for pollution liability, because many general liability policies exclude pollution, including glycol releases; the actual policy wording is checked on each project.
Scarce capability is secured in advance. Live-system repair, sealing a leak under pressure or tapping a line in service, is held by a handful of firms in any one region. That is arranged under framework agreement before it is needed, not on the day it is.
Professional engineering is carried out by licensed engineers. Work that falls within professional engineering, such as the design of building systems, an engineering opinion or a sealed document, is carried out by licensed professional engineers, or firms holding the authorization that the province or state where the work is done requires (in Ontario, a Certificate of Authorization from Professional Engineers Ontario), engaged by the owner. Qualification questions are settled before the quote, not worked around at delivery.
What you receive
A turnover package that answers three questions for every leak point on the thermal path: was it tested, against what, and where is the record. A point that cannot answer all three is an open item, not a pass.
The same records define the warranty boundary between the parties, and they are the evidence an insurer asks for. A documented leak detection and inspection regime is one of the few things that visibly changes how a liquid-cooled hall is underwritten.