The power chain, end to end
Trace a single watt from the grid to a server and you will understand half of data center engineering. Power arrives from the utility at medium or high voltage and is stepped down through transformers. It passes through medium-voltage switchgear into the building, where it is distributed to uninterruptible power supply (UPS) systems. The UPS conditions the power and bridges the gap — using batteries or flywheels — between a utility failure and the moment standby generators come online. Downstream of the UPS, power distribution units (PDUs) or busway deliver power to racks, where rack PDUs feed the individual servers.
Two ideas matter most at this stage. First, the chain is only as strong as its weakest link, which is why critical facilities duplicate components and pathways (redundancy, covered in the next lesson). Second, standby generators — usually diesel — are the true backstop: batteries are sized in minutes, generators in days, limited mainly by fuel logistics.
Every watt becomes heat
Servers do not consume electricity in any permanent sense — nearly 100% of the electrical energy delivered to IT equipment is converted to heat. A 10 MW data hall is therefore also a 10 MW heater, and the cooling plant must continuously remove that heat or temperatures in the room will climb within minutes. This is why cooling is not an amenity but a critical system with the same redundancy expectations as power.
Cooling failures are in some ways less forgiving than power failures: a UPS gives you ride-through by design, but thermal "ride-through" is only the thermal mass of the room and, in chilled-water plants, the water stored in pipes and tanks. High-density halls can exceed safe temperatures in tens of seconds without airflow — a fact that shapes both design and emergency procedures.
How air cooling works
The dominant model for the last two decades is air cooling with hot/cold aisle separation. Racks are arranged in rows so that all server intakes face a "cold aisle" supplied with conditioned air, and all exhausts face a "hot aisle." Containment — physical barriers, doors, and ceilings over an aisle — prevents hot exhaust air from mixing back into the cold supply, which is the single most impactful airflow improvement a facility can make.
The air itself is conditioned by CRAC units (computer room air conditioners, with their own refrigeration compressors) or CRAH units (computer room air handlers, cooled by chilled water from a central plant). Modern hyperscale designs often replace perimeter units with fan walls feeding the whole hall. The heat collected indoors is ultimately rejected outdoors by chillers, cooling towers, or dry coolers.
Free cooling and why climate matters
Mechanical refrigeration is the most energy-hungry part of the cooling plant, so designers avoid running it whenever outdoor conditions allow. "Economization" or free cooling uses cool outside air directly (air-side) or uses outdoor air to chill water without running compressors (water-side). In cool, dry climates a facility can spend most of the year on free cooling — one reason site selection weighs climate alongside power availability and fiber.
ASHRAE publishes recommended and allowable temperature/humidity envelopes for IT equipment; running data halls warmer than the historic 68°F habit — often 75–80°F supply air today — dramatically extends free-cooling hours with no reliability penalty for modern hardware.
Key takeaways
- The power chain runs utility → transformer → switchgear → UPS → distribution → rack; generators are the real backstop, batteries only bridge the gap.
- Effectively all IT power becomes heat, so cooling capacity must match IT capacity watt for watt.
- Hot/cold aisle separation with containment is the foundation of efficient air cooling.
- Free cooling (economization) avoids running compressors and makes climate a first-order site-selection factor.