Liquid cooling data centre deployment and due diligence

High-density AI and HPC deployments require verifying facility readiness across primary chilled-water loops, coolant distribution units, slab loading and thermal demarcation before entering into a lease. When preparing a technical brief to match against capable halls and compare operator proposals side by side, buyers should specify:

• Target metropolitan market and country.
• Total power requirement and planned rack densities in kW.
• Hardware delivery milestones and deployment phases.
• Preferred cooling architecture, such as direct-to-chip cold plates, rear-door heat exchangers, or immersion tanks.

Assessing a colocation facility begins by separating central mechanical plant from white-space cooling circuits. Liquid cooling separates primary facility water from secondary technology cooling circuits using heat exchangers.

According to the OCP Data Center Facilities project, the Coolant Distribution Unit (CDU) serves as “the interface between facility-water systems and Technology Cooling Systems (TCS) serving cold plates, door heat exchangers, and immersion-cooled IT”. The primary facility water loop circulates water from chillers or dry coolers to the CDU, while the secondary loop circulates treated fluid to server hardware.

Research from IEA 4E EDNA notes that direct-to-chip single-phase liquid cooling “is the most mature and widely deployed currently, commanding approximately 43% of the liquid cooling market by revenue. Immersion cooling is though the fastest-growing segment”.

Key diligence questions for operators:

• Does the primary water loop supply liquid-cooled rows exclusively, or share pipework with air handlers?
• Can technicians isolate a single row header without interrupting flow to adjacent racks?
• What pipework materials are installed, and do they match CDU heat exchanger metallurgy?

Adopting liquid cooling does not remove the need for data hall air handling. Direct-to-chip cold plates extract heat from high-wattage processors, but adjacent electronics reject heat directly into ambient room air.

Technical documentation from the Open Compute Project establishes that “A typical cold plate cooling system may capture 70%-80% of the ITE heat… TCS capacity requirements for immersion would be closer to 100%”. In cold plate setups, the remaining heat dissipates into the room from memory modules, voltage regulators, and network transceivers. If airflow fails, ancillary components risk thermal throttling.

Key diligence questions for operators:

• Are room air handlers sized to remove twenty to thirty percent of total rack power as sensible heat?
• What containment structures are installed, and do partition seals withstand elevated static pressure?
• Do air circulation fans stay powered by uninterruptible power systems during utility transfers?

Operating temperatures dictate chiller dependency and efficiency. Buyers must verify whether the facility water loop maintains contracted supply temperatures during summer conditions without derating performance.

Guidelines from the Open Compute Project confirm that water-based TCS loops “typically operate <49 °C (120 °F) and are not expected to be >66 °C (150 °F)”, with temperatures above 66 °C requiring a wetted-materials review to prevent polymer degradation. Technical documentation from Dell Technologies recommends a 20 °C supply for rear-door heat exchangers and 32 °C for CDUs, site-dependent.

Immersion architectures require different physical tolerances. The Open Compute Project immersion requirements mandate that solutions handle facility water supply partial vacuum to 50 kPa absolute and pressure to 1000 kPa gauge, whilst supporting glycol mixtures to 50%.

Key diligence questions for operators:

• What was the peak primary water supply temperature recorded over the past three summers?
• Does the facility use dry coolers, cooling towers, or mechanical chillers during heatwaves?
• How does the operator prevent condensation if primary supply water temperatures fall below room dew point?

The physical boundary between tenant equipment and facility plant sits at the Coolant Distribution Unit or the row manifold. Determining who owns, operates, and services the CDU is essential to avoid operational conflicts. Technical specifications from Dell Technologies indicate that customers “are responsible for validating the selected CDU, secondary fluid network, and facility infrastructure”, with maintenance schedules set by the CDU vendor.

Operators deploy CDUs as centralised perimeter units distributing fluid through shared floor headers, or as dedicated in-row or in-rack units that isolate chemistry and pressure control while consuming white-space floor area.

The Chilldyne CDU-1500 specification, published by Tecomas, describes “negative pressure technology” designed to draw fluid through cold plates under sub-atmospheric pressure to mitigate leak risks.

Key diligence questions for operators:

• Does the operator or the tenant provide, service, and warrant the CDU?
• What pump redundancy is installed within the CDU, and can failed pumps be swapped online?
• Are secondary manifold connections equipped with dry-break quick-disconnect couplings?
• Where does contractual liability fall if a leak in facility pipework damages tenant hardware?

Secondary fluid quality directly impacts processor life and cooling efficiency. Inadequate water treatment causes biological fouling, galvanic corrosion, and micro-channel blockages inside cold plates.

Maintenance procedures from Dell Technologies state that coolant chemistry testing is required “a minimum of one test every 6 months… on each Secondary Fluid Network”. Regular testing verifies that inhibitor packages and fluid properties remain within manufacturer limits.

Key diligence questions for operators:

• Who collects fluid samples, and which accredited laboratory conducts chemical testing?
• What corrosion inhibitors, biocides, and glycol ratios does the operator require or permit?
• What micron rating is specified for inline filters, and are dual-filter bypass arrangements installed for online swaps?

Liquid cooling adds substantial weight from fluid manifolds, cold plates, and dense server components. Buyers must verify whether slab and raised-floor load ratings match the planned equipment. Pipe routing also requires mechanical segregation from electrical infrastructure.

Key diligence questions for operators:

• What are the point-load and distributed load ratings for the concrete slab or raised floor?
• Can the delivery path from loading bays to the white space support rolling weights of heavy rack cabinets?
• How does the operator physically segregate coolant pipework from electrical busways and power cables?
• What containment bunding, drainage sumps, and moisture-sensing ropes protect the white space beneath CDUs?

Liquid cooling systems rely on continuous electrical power to circulate fluid. If circulation pumps stop, server silicon temperatures rise quickly, triggering emergency shutdowns. Buyers must examine how mechanical pumping loops integrate with facility power paths.

Key diligence questions for operators:

• Are CDU pumps dual-corded and connected to separate uninterruptible power supply circuits?
• Do secondary circulation pumps maintain continuous flow during the mechanical transition to emergency backup generators?
• What automated controls isolate valves or shut down compute nodes if a sudden pressure loss signals a major fluid breach?

Colocation contracts for liquid cooling must move beyond standard room temperature and power availability metrics. Buyers and operators must establish measurable thermodynamic and hydraulic commitments.

Key diligence questions for operators:

• What specific fluid flow rate, supply temperature, and differential pressure commitments are written into the service level agreement?
• What are the agreed remedy periods and compensation terms if supply water temperatures exceed contracted ceilings during peak weather?
• Where does contractual liability fall if a leak in facility-managed pipework damages tenant server hardware?

Most public directories do not show which halls have primary chilled-water loops, floor loading and CDU demarcation ready for high-density AI and GPU racks on your delivery date. Data Centre Axis tracks 2,004 project signals across Australia (215) and the UK (1,789), maintains 115 facility profiles and 44 operator profiles across Australian data centres and UK data centres, and covers major US and Canadian colocation markets.

What you get when you work with us:

• Find capacity (/colocation/, /colocation/wholesale/): Liquid-cooled colocation, wholesale and AI-ready capacity sourced against your MW or kW load, rack density, cooling topology and target live date, with operator proposals compared side by side.
• Find tenants and offtakers (/colocation/anchor-tenants/, /colocation/offer-capacity/): Anchor tenants, pre-lets and high-density AI and HPC customers for operators commissioning liquid-cooled halls or new builds.
• Find a site: Off-market powered land and sites screened on firm grid headroom and water availability for developers and operators building high-density campuses.