Data Center Air Conditioning: What to Know Before Requesting a Quote

Quoting air conditioning for a data center is not like quoting it for an office. The most common mistake is treating the problem as “buying a bigger unit,” when in reality it is about designing a redundant system with growth capacity, measurable energy efficiency, and failure response. Most of the cost overruns seen in Mexican critical infrastructure projects come precisely from decisions made at this stage, not at the server stage.

This article is a guide for the Mexican buyer — facility manager, CTO of an industrial SME, corporate IT director — to arrive at the first meeting with a vendor speaking the same language and know what to ask before receiving the first quote.

Why comfort air conditioning is not suitable for a data center

A 5-ton mini-split that perfectly cools an office fails in a data center for three technical reasons:

• It does not control humidity with the precision required by ASHRAE TC 9.9 (the American Society of Heating, Refrigerating and Air-Conditioning Engineers’ guide for data processing environments). The recommended range is 40-55% relative humidity. Below 40%, electrostatic discharges are triggered; above 60%, there is condensation on contacts and accelerated corrosion.
• It lacks redundancy. If the compressor fails, the rack’s intake temperature begins to rise within minutes. Modern servers start experiencing throttling (reducing their speed to lower generated heat) at 27 °C intake; some critical equipment shuts down at 32 °C.
• It does not handle variable loads. A data center has “heat islands” depending on the workload; comfort equipment is designed to maintain a uniform temperature in an open space, not to respond to localized gradients within a rack.

CRAC, CRAH, in-row, and liquid cooling: which one do you need

The cooling architecture decision depends on the project’s rack density. In 2026, the ranges are:

• Up to 8 kW per rack: direct expansion CRAC (Computer Room Air Conditioning), with N+1 redundancy. It is the dominant solution in Tier II sites and on the lower floor of Tier III. Brands: Vertiv Liebert, Stulz, Schneider Electric.
• From 8 to 25 kW per rack: CRAH (Computer Room Air Handler) with chilled water produced by external chillers. Better energy efficiency in temperate climates. Brands: Trane, Carrier, York (Johnson Controls).
• From 25 to 60 kW per rack: in-row cooling between racks. It reduces the path of hot air to the exchanger and eliminates hot spots. Brands: Vertiv, Stulz, Rittal.
• More than 60 kW per rack: direct liquid cooling (DLC) to the chip or immersion cooling. Here, it’s no longer about “choosing a brand” but redesigning the rack, piping, and fluid distribution network.

Four figures the vendor must provide before quoting

Any serious vendor provides four pieces of data before quoting a price. If they don’t have them, they are quoting by guesswork:

• Net sensible capacity (kW) at the site’s design conditions: outdoor dry-bulb and wet-bulb temperature of the worst month of the year, altitude above sea level, and air return temperature to the equipment. In Querétaro or Monterrey, the design dry-bulb temperature should be between 38 and 42 °C.
• EER (Energy Efficiency Ratio) and COP (Coefficient of Performance) of the equipment at partial load ((50% and 75%), not just at full load. Comfort equipment has an acceptable COP at full load that plummets at partial load. In a data center, the equipment operates 70% of the time between 40% and 70% load.
• Airflow (CFM, cubic feet per minute) and available static pressure: the equipment must deliver the necessary airflow against the resistance of the raised technical floor, containment grilles, and diffusers. If the vendor does not ask about the floor configuration, they do not understand the problem.
• Capacity vs. outdoor temperature curve: the net capacity of the equipment drops when the outdoor temperature rises. In warm Mexican climates, the difference between the equipment at 25 °C outdoor and 38 °C outdoor can be 25-30% less capacity. If the vendor does not provide the curve, it will fail you in July.

Trane, Carrier, York vs. Vertiv, Stulz, Schneider: two distinct families

In the Mexican market, there are two families of vendors that quote in similar ranges but with opposing approaches:

• Trane, Carrier, York (Johnson Controls): traditional commercial HVAC (Heating, Ventilation and Air Conditioning) manufacturers with specialized lines for data centers. Advantage: mechanical robustness, local spare parts, service network throughout Mexico. Disadvantage: control electronics are generic, and integration with DCIM (Data Center Infrastructure Management) requires extra licensing.
• Vertiv (Liebert), Stulz, Schneider Electric (Uniflair): native data center cooling manufacturers. Advantage: native integration with their own DCIM, granular control per rack, pre-configured operating modes for containment. Disadvantage: more expensive spare parts and reliance on the manufacturer for advanced maintenance.

For a site with less than 200 kW of IT load, commercial brands are usually more cost-efficient. For sites of 500 kW and above, native DC brands justify their premium due to management integration and lower annual electricity consumption.

Oversizing that increases cost by 30% without reason

The classic trap is sizing the system for the theoretical peak load of a full site. Three facts change that calculation:

• Servers rarely operate at 100% of their nominal capacity simultaneously. The average actual load factor in operational data centers is between 35% and 55%.
• Modern cooling equipment operates better at partial load than at full load. Oversizing by 30% means operating at 50% load most of the time, with worse energy efficiency and higher acquisition cost.
• Growth can be absorbed modularly. Buying two redundant 100 kW chillers (2N at 50% each) is usually more efficient than buying one oversized 250 kW chiller.

What the maintenance contract should include

The most expensive mistake is buying the equipment and forgetting about maintenance until it fails. The maintenance contract must specify:

• Monthly preventive visits with a documented checklist signed by the field technician.
• Written committed response time in case of failure: 4 hours during business hours, 8 hours during non-business hours. Without this clause, the vendor may prioritize another client.
• Supply of original spare parts with guaranteed delivery time. Expansion valves, compressors, and fan motors are not universal; using generics invalidates the warranty.
• Annual water quality analysis if the system uses chilled water. Hard water from Monterrey and the Guadalajara metropolitan area destroys chillers in 3 years if untreated.
• Quarterly energy efficiency report. Request the kWh consumed by the cooling system and compare it against the kWh delivered to the IT load. The trend should be downward, not upward.

An informed decision at this stage determines whether the data center will operate reliably and economically for the next 10 years. It is not the most visible decision of the project, but it has the greatest impact on the final TCO (total cost of ownership).

Sources

[1] ASHRAE — Technical Resources for Data Center Cooling: https://www.ashrae.org/technical-resources

[2] TIA — TIA-942 Telecommunications Infrastructure Standard for Data Centers: https://tiaonline.org/products/tia-942/

[3] Uptime Institute — Research and Reports on Data Center Operations: https://uptimeinstitute.com/resources/

[4] Vertiv — Thermal Management Catalog (Data Center Cooling): https://www.vertiv.com/en-us/products-catalog/thermal-management/

[5] Stulz — Data Center Precision Cooling Solutions: https://www.stulz.com/en-us

[6] Wikipedia — Data center cooling: https://en.wikipedia.org/wiki/Data_center

[7] Wikipedia — Free cooling: https://en.wikipedia.org/wiki/Free_cooling