Data Centre Cooling: Keeping the Cloud from Overheating

A data centre looks like rows of computers, but from an engineering view it's a giant heat problem: servers convert nearly all the electricity they draw into heat, packed into dense racks that run 24/7. If the cooling stops for even minutes, temperatures spike and equipment fails — so data-centre HVAC is precision cooling that must be reliable, continuous and efficient at a scale few other buildings see. With India's data-centre sector booming, this is one of the fastest-growing HVAC specialisms in the country. This guide explains how data-centre cooling works and why PUE rules it.

It builds on the Special Buildings HVAC Guide, part of the HVAC Knowledge Hub.

Scope. This explains data-centre cooling concepts to inform understanding. Mission-critical cooling design is specialist engineering with strict redundancy and validation. Use our data centre cooling calculator for a planning estimate only.

The core problem: concentrated, continuous heat

What makes data-centre cooling unlike anything else:

• Nearly all IT power becomes heat. A rack drawing 5–20+ kW turns almost all of it into heat that must be removed continuously — a heat density far beyond any office.
• Failure is catastrophic — no comfort tolerance; a cooling outage means overheating, throttling and hardware damage within minutes, so reliability and redundancy are paramount.
• It runs 24/7/365 — so efficiency (the cooling energy per unit of IT) is a massive, permanent operating cost.

The whole design is about removing concentrated heat reliably and efficiently, without interruption.

Hot-aisle / cold-aisle containment

The foundational layout that makes data-centre cooling work:

• Arrange racks in alternating aisles — server fronts (air intakes) face each other across a cold aisle; server backs (hot exhaust) face each other across a hot aisle.
• Supply cold air to the cold aisle, where servers draw it in; collect hot exhaust from the hot aisle and return it to be re-cooled.
• Contain the aisles — physical barriers (doors, roofs) stop hot and cold air mixing, which is the single biggest efficiency win: the cooling only has to handle genuinely hot return air, not a lukewarm mix.
• The result — predictable airflow, no hot spots, and far less wasted cooling.

Containment turned data-centre cooling from brute-force over-cooling into an engineered airflow problem.

The cooling equipment

• CRAC / CRAH units — Computer Room Air Conditioners (direct-expansion) or Air Handlers (chilled-water); the workhorses that cool and circulate the air.
• Chilled-water plant — chillers, pumps and cooling towers for larger facilities.
• In-row / rear-door cooling — cooling brought close to the racks for high densities.
• Liquid cooling — direct-to-chip or immersion cooling for the highest-density (AI/HPC) racks, where air alone can't cope — an emerging frontier.
• Free cooling — using cool outdoor air or water to cool without running compressors when ambient conditions allow; a big efficiency gain (harder in much of hot India, but valuable in cooler seasons/locations).
• Redundancy — N+1 or greater, so any single unit can fail without losing cooling.

PUE — the number that rules data centres

The defining efficiency metric:

• PUE (Power Usage Effectiveness) = total facility power ÷ IT power. A PUE of 1.0 would mean all power goes to IT and none to overhead; real data centres are higher because cooling and losses add overhead.
• Lower is better — a PUE of 2.0 means as much power for cooling/overhead as for the IT itself; efficient modern facilities push toward 1.3–1.5 or lower.
• Cooling dominates the overhead, so cooling efficiency is most of PUE — which is why containment, free cooling, higher supply temperatures and efficient chillers matter so much (see efficiency).
• In India — warm ambient temperatures make low PUE harder, so efficient cooling design is both an environmental and a serious cost lever. See the cooling calculator.

The one-line answer

A data centre is really a giant, concentrated heat problem: servers turn nearly all their power into heat, packed into dense 24/7 racks where a cooling outage means overheating and hardware failure within minutes — so its HVAC is precision cooling that must be reliable, continuous and efficient. The foundational layout is hot-aisle/cold-aisle containment: racks are arranged so server intakes face a cold aisle (fed cold air) and exhausts face a contained hot aisle (returned to be re-cooled), with barriers stopping hot and cold air mixing — the biggest efficiency win. The equipment runs from CRAC/CRAH units and chilled-water plant to in-row, liquid and free cooling, always with N+1 redundancy so no single failure stops the cooling. And the number that rules it all is PUE (total facility power ÷ IT power — lower is better, efficient facilities near 1.3–1.5), of which cooling is most of the overhead — making cooling efficiency the central lever, and a harder, costlier one in warm India. All specialist, mission-critical engineering.

Where to go next

• Size a planning estimate: Data Centre Cooling Calculator.
• The overview: Special Buildings HVAC Guide.
• The plant behind it: Chiller Guide · Central Air Conditioning Guide.
• The efficiency lens: HVAC Energy Efficiency Guide.

References

• ASHRAE Technical Committee 9.9 (Thermal Guidelines for Data Processing Environments) & ASHRAE Standard 90.4 (Energy Standard for Data Centers).
• The Green Grid — PUE (Power Usage Effectiveness) metric; Uptime Institute — data-centre tier & redundancy standards.
• ISHRAE data-centre HVAC guidance; Bureau of Energy Efficiency (BEE) — data-centre energy efficiency: https://beeindia.gov.in/

This guide informs understanding; the calculator is a planning estimate only. Mission-critical cooling design is specialist engineering with strict redundancy and validation.