Quick Answer: Data centre cooling removes heat from servers and networking equipment to keep hardware operating safely, prevent performance throttling, and reduce the risk of outages or premature component failure.
As AI, GPU, and high-performance computing workloads push rack densities far beyond traditional enterprise deployments, cooling has become a critical factor when choosing a colocation provider. Modern facilities use a combination of air cooling, liquid cooling, and, in colder climates like Canada, free cooling to improve efficiency.
Most IT decision-makers examine uptime SLAs, power redundancy, and carrier connectivity when evaluating a colocation provider. Cooling rarely gets the same level of scrutiny, and that oversight has become increasingly costly as workload densities climb.
When a cooling system fails, servers don't gradually slow down over hours. Temperature thresholds are breached in minutes, triggering hardware throttling, unexpected shutdowns, and in sustained thermal events, permanent equipment damage.
The challenge is compounded by the pace at which enterprise workloads have changed. AI inference, GPU-accelerated analytics, and high-performance computing have pushed rack densities far beyond the planning assumptions built into older data centre designs.
A facility that was adequate for standard compute three years ago may not support what your infrastructure team is deploying today, and it almost certainly won't support what you're planning two years out. Choosing a colocation provider without interrogating the cooling architecture is one of the more common ways enterprises end up in a contract with a facility their workloads will outgrow.
Data centre cooling is the continuous process of removing heat generated by IT equipment to maintain safe operating temperatures throughout a facility. It operates across multiple layers simultaneously, managing thermal load at the chip level, the rack level, and across the room.
The challenge is that heat generation in modern facilities is not uniform. A single high-density rack running GPU workloads in one aisle can generate more heat than an entire row of standard compute.
Every server in a data centre converts electrical energy into two outputs: computational work and heat.
Processors, memory, storage, and power supply units all contribute to thermal output, and that output scales with workload intensity. GPUs used in AI training and inference are particularly concentrated heat sources, generating thermal loads at the chip level that air movement alone struggles to manage efficiently at high densities.
Rack densities are another part of the problem here.
A decade ago, a typical enterprise rack drew 2-5 kW. Standard enterprise compute today commonly runs at 10-15 kW per rack. AI and high-performance computing deployments push 30-60 kW per rack, with next-generation GPU configurations targeting even higher.
Industry research consistently estimates that cooling accounts for approximately 30-40% of a data centre's total energy consumption, which reflects just how central these systems are to facility operations. When cooling degrades, the consequences for tenants compound quickly:
For organisations running mission-critical applications, even partial cooling degradation that doesn't formally breach an uptime SLA can quietly degrade application performance in ways that take time to diagnose.
This is precisely why cooling redundancy deserves serious attention during procurement, not just the headline uptime percentage a provider advertises.
Facilities don't all cool the same way, and the method a data centre uses has direct implications for the types and densities of workloads it can support. Three primary approaches are in active use across enterprise colocation environments today, each with distinct trade-offs around density support, energy efficiency, and infrastructure cost.
Air cooling is the most widely deployed method in enterprise data centres. It works by circulating chilled air through server racks to absorb and remove heat, using Computer Room Air Conditioning (CRAC) or Computer Room Air Handling (CRAH) units that draw warm exhaust air, cool it, and redistribute it throughout the facility.
Most modern facilities pair this with hot and cold aisle containment, a configuration that physically separates rows of cool air intake from hot exhaust rows using barriers, preventing mixing and improving thermal efficiency significantly.
ASHRAE's thermal guidelines for data processing environments recommend server inlet temperatures between 18°C and 27°C for optimal hardware performance, a range that well-configured air cooling systems maintain reliably. The limitation surfaces at density.
Air is a relatively poor conductor of heat compared to liquid, and as rack densities climb past 15-20 kW, airflow requirements, fan energy, and pressure management all scale poorly.
Data centre air cooling remains the right solution for the majority of standard enterprise workloads, but it has a practical ceiling for AI and high-performance compute at the densities those workloads now demand.
Liquid cooling addresses the density ceiling of air by bringing coolant directly to heat-generating hardware. The two primary approaches in active enterprise deployment are direct-to-chip cooling and immersion cooling, and they differ considerably in both application and upfront cost.
Direct-to-chip cooling uses cold plates mounted on processors, with coolant circulated through pipes to carry heat away from chip surfaces to a heat exchanger. This integrates with existing rack infrastructure and supports densities well above what air cooling manages alone, making it the more accessible entry point for organisations building AI-capable environments.
Immersion cooling takes the density ceiling further: servers are submerged in a bath of non-conductive dielectric fluid that absorbs heat directly from all components simultaneously. It achieves the highest thermal efficiency of any current method, but it requires purpose-built infrastructure and hardware configurations that carry a significant cost premium relative to air-cooled deployments.
Free cooling, sometimes called economisation, reduces or eliminates mechanical refrigeration by using ambient outdoor air or cold water sources to manage thermal load.
When outdoor temperatures are low enough, facilities draw cool outside air through heat exchangers to pre-cool or fully handle the thermal load without running energy-intensive mechanical chillers. In warmer climates, this is viable for only a fraction of the year. Canada's climate changes the calculation considerably.
Natural Resources Canada's Best Practice Guide for Data Centres identifies free cooling as a priority efficiency measure for Canadian operators, citing the country's extended cold seasons as a structural operational advantage.
For enterprise buyers considering colocation in cities like Toronto, Calgary, Edmonton, and Ottawa, Canadian facilities can use economisation for a meaningful portion of the year, reducing mechanical chiller use and improving overall energy efficiency.
Redundancy ratings are one of the most important signals in a facility's technical specification, yet they are frequently overlooked during procurement or treated as equivalent when they are meaningfully different.
A facility with inadequate cooling redundancy is not just less resilient during equipment failure; it is also constrained in its ability to service cooling units without creating risk windows for live workloads.
In cooling redundancy notation, N represents the number of cooling units required to sustain the facility's full thermal load under normal operating conditions. An N+1 configuration adds one unit beyond that minimum, so if one unit fails or requires maintenance, the remaining capacity absorbs the full load without interruption.
N+2 adds two additional units, providing a greater fault tolerance margin and enabling planned maintenance with significantly more flexibility and less operational risk.
The Uptime Institute's Tier classification system ties cooling redundancy directly to facility tiers. Tier III certification requires concurrently maintainable cooling systems, meaning any single cooling component can be serviced without taking the system offline.
Tier IV requires fault-tolerant systems where any single cooling failure has no impact on the live environment. For enterprise buyers, Tier III certification is a meaningful baseline: it confirms that a provider's cooling maintenance schedule doesn't require opening a planned risk window for your infrastructure.
Power Usage Effectiveness (PUE) measures how much of a facility's total energy reaches IT equipment versus being consumed by infrastructure like cooling, lighting, and power distribution. PUE is calculated by dividing total facility power by IT equipment power alone.
A PUE of 1.0 is theoretical perfection; real-world efficient facilities target below 1.4, with the best-performing sites reaching 1.2 or lower through free cooling integration and optimised thermal management.
The Uptime Institute's 2024 Global Data Centre Survey recorded an average global PUE of approximately 1.56, meaning roughly a third of total energy consumed goes to infrastructure rather than compute.
A provider sharing their PUE is important. It reflects the quality of the cooling design, the maturity of energy management, and the operational discipline of the team running the facility day to day. A notably lower PUE relative to facility age and market is one of the more honest signals about how the infrastructure is actually managed.
Cooling infrastructure is one of the most consequential technical factors in a colocation evaluation, yet it rarely receives proportionate attention in RFP processes and vendor assessments. Two areas in particular tend to be underweighted: the relationship between power density and cooling capacity, and what a provider's SLA actually commits to in thermal terms.
Qu Data Centres operates nine carrier-neutral facilities across five Canadian markets, each built with N+1 or N+2 cooling redundancy and purpose-built to support enterprise and high-density workloads. TOR3 in the Greater Toronto Area carries over 1,400 tons of Liebert DSE cooling with EconoPhase free-air cooling technology. CGY3 in the Calgary region and EDM2 in Edmonton both run 1,450 tons of N+1 cooling capacity.
These data centre cooling systems reflect infrastructure designed for the thermal demands of modern enterprise and AI workloads, not the density planning standards of a decade prior. Qu's colocation services span shared space, dedicated cages, and private suites, with cooling infrastructure matched to the deployment type.
A facility's maximum supported rack density is not a power figure; it is a cooling figure. The reason a provider can support 10 kW per rack in one zone and 30 kW in another comes down entirely to what the cooling infrastructure in that area can remove.
When evaluations focus on available floor space and utility power without addressing per-rack cooling capacity at the zone level, the picture is incomplete in a way that creates real operational problems after contracts are signed.
This becomes particularly significant when workloads change. An enterprise deploying standard compute today but planning GPU-accelerated workloads in 18 months needs to know whether the facility's cooling design supports that transition without requiring a physical relocation.
Data centre cooling requirements are not static; they are planning parameters. Asking a prospective provider for their supported kW per rack range, by zone where the facility has varied density areas, gives you a meaningful signal about whether the infrastructure can grow with your needs.
Pairing that with an understanding of the provider's managed services and active monitoring capabilities tells you whether the team can respond to thermal load changes as your environment evolves.
Colocation agreements vary considerably in what they commit to on the cooling side. Standard SLAs address uptime percentages and sometimes specify supply air temperature ranges, but they don't always cover partial thermal degradation that falls short of a formal outage threshold.
Before finalising a provider, these questions help close the gap between a marketing spec sheet and an operational commitment:
Cooling is one of the most operationally invisible systems in a data centre right up until it becomes the most visible problem. Qu Data Centres builds and operates its nine facilities so that cooling is never the variable that limits your workloads or creates an unplanned conversation.
Across facilities in Calgary, Edmonton, Ottawa, Toronto, and London, Ontario, every site is built with N+1 or N+2 cooling redundancy, advanced climate monitoring, and infrastructure sized for high-density and AI workloads.
Four facilities within the Qu network hold Uptime Institute Tier III certification, independently verifying that cooling systems are concurrently maintainable and built to the reliability standards that regulated enterprises need before committing to a provider. Canada's climate gives Qu facilities a structural efficiency advantage, enabling free cooling integration that improves PUE and reduces mechanical chiller load for a significant portion of the year.
Our team works alongside customers before deployment to map workloads to the right facility zone, making sure cooling capacity, power density, and connectivity align with what you're running today and where you're headed. That planning conversation happens before a contract is signed, not after an installation reveals a gap.
Want to see whether our facilities match your specific density requirements? Book a tour and see the infrastructure firsthand.
ASHRAE recommends that air temperature at server inlets be maintained between 18°C and 27°C for optimal hardware performance. Most enterprise facilities target the lower portion of that range as a buffer under variable load conditions. Humidity is managed alongside temperature, with a target relative humidity of 40-60% to prevent static discharge and condensation damage to sensitive components throughout the facility.
Air cooling circulates chilled air through racks to absorb and remove heat. Liquid cooling uses coolant, either through cold plates on processors or by submerging hardware in dielectric fluid, to transfer heat more directly and efficiently. Liquid transfers heat significantly more effectively than air, making it better suited for high-density workloads where airflow alone cannot keep pace with thermal output above roughly 20 kW per rack.
N+1 cooling means a facility operates one more cooling unit than the minimum required to handle its full thermal load. If the required capacity is met by four units (N=4), an N+1 facility runs five. This allows any single unit to fail or be taken offline for maintenance while the remaining units sustain full cooling capacity, protecting live workloads from any thermal impact during both unplanned failures and scheduled servicing.
PUE measures how much of a facility's total energy reaches IT equipment versus supporting infrastructure like cooling and power distribution. A lower PUE means more energy reaches compute. Cooling is the primary driver of PUE in most facilities, so a lower number reliably signals better-designed, better-operated cooling infrastructure. It is one of the most useful single metrics for benchmarking facility efficiency during a provider evaluation process.
Canada's cold climate allows facilities to use free cooling, substituting ambient outdoor air for mechanical refrigeration during extended cold periods. Cities like Toronto, Calgary, Edmonton, and Ottawa experience long winters that make economisation viable for a significant portion of the year. This reduces mechanical chiller load, lowers energy consumption, and improves PUE, giving Canadian facilities a structural efficiency advantage over comparable facilities in warmer climates.