Capacity planning for data center space is usually expressed in racks. The constraint that actually binds is almost always power and the cooling required to reject the resulting heat — and both are considerably harder to add later than rack space is.
How the problem arrives
Equipment power density has risen substantially while physical footprint has not. A rack provisioned for a certain load a decade ago will not support a modern equivalent population, which means a refresh that reduces physical space can simultaneously exceed the electrical and thermal design.
The failure mode is rarely dramatic. It presents as thermal throttling, unexplained performance variation, equipment failing earlier than expected, and hot spots that persist despite adequate total cooling capacity. Those symptoms get diagnosed as hardware problems for months before anyone measures inlet temperature at the top of the rack.
What to establish before purchasing
Actual power draw per rack, measured rather than derived from nameplate ratings, which overstate substantially. Rack-level metering pays for itself the first time it prevents a bad assumption.
Available capacity at the distribution board, not just at the rack. Adding load to a rack that has spare outlets but sits on a fully loaded circuit is a straightforward way to cause an outage.
Cooling capacity and, more importantly, airflow. Total cooling capacity being adequate says little if the air is not reaching the equipment. Containment — separating hot and cold air paths — usually delivers more improvement than additional cooling capacity, at lower cost.
UPS runtime at the new load. Battery autonomy is a function of load. Adding equipment shortens runtime, sometimes below the time needed for an orderly shutdown, and this is rarely recalculated during a refresh.
Practical measures
Blanking panels in unused rack units. Cheap, unglamorous, and one of the most effective interventions available — without them, hot air recirculates to the equipment intake.
Hot and cold aisle discipline, maintained rather than merely designed.
Inlet temperature monitoring at the top of each rack, which is where thermal problems appear first.
Recalculating power and cooling as part of every equipment purchase, not as a separate facilities exercise. The two decisions are the same decision, and separating them is how the constraint gets discovered after delivery.