Ask any data center operator how long their UPS holds the load and you will get a number, usually to the minute, usually backed by a load-bank test. It is one of the most carefully established figures in the building.
Ask the same operator how long the room holds if cooling stops and the answer is far less likely to be a number. It is more often a shape: a pause, then something about N+1 chillers and automatic failover. Those are good answers to a different question.
The two quantities are not related, and in a high-density hall they are not even the same order of magnitude.
Two different clocks
Electrical ride-through is a storage problem. A UPS holds a known quantity of energy, the load draws it at a known rate, and the runtime falls out of the arithmetic. It is testable, repeatable and boring, which is exactly what you want.
Thermal ride-through is not a storage problem. It is a balance problem. The IT equipment produces heat continuously, and in normal operation the cooling system removes it at the same rate. When removal stops, production does not. Everything that happens next depends on how much heat the room can absorb before something reaches a temperature it will not tolerate.
That absorption capacity is the air in the hall, the water in the loops, and the mass of the equipment and the structure itself. It is finite, it is usually modest, and almost nobody knows what theirs is.
Why density made this worse
A conventional hall at a few kilowatts per rack carries a large volume of air relative to the heat going into it. The temperature rise after a cooling failure is gradual, and an operator has a genuinely useful amount of time to diagnose, switch over, or begin an orderly shutdown.
Move the same compute into high-density racks and that comfort disappears from both directions at once. The heat per unit volume goes up sharply. The air available to absorb it goes down. Our guide to power density tiers from 5kW to 100kW covers what changes across that range; the thermal consequence of the top of that range is that the room stops being a buffer and starts being a thin film.
Containment sharpens it again. Hot and cold aisle containment exists to stop supply air mixing with exhaust, and it does that job well. The same seal that delivers the efficiency removes the reservoir of neutral air that used to slow things down. A well-contained high-density hall is, thermally, a much less forgiving room than a badly contained low-density one, which is an uncomfortable thing to be true.
Liquid cooling changes the shape again rather than simply improving it. A direct-to-chip loop has far better heat transfer and a far smaller fluid volume, so it removes heat beautifully and buffers it hardly at all. When circulation stops, the temperature at the chip climbs quickly. Our material on coolant distribution units covers the equipment; the operational point is that pump redundancy matters more in a liquid-cooled hall than chiller redundancy does in an air-cooled one, because there is less margin behind the pump.
The failure that is easy to miss
Most cooling-loss planning assumes the cooling plant failed. There is a more common path that is less often rehearsed.
Utility power is interrupted. The UPS picks up the IT load instantly, exactly as designed. The generator starts, and takes some seconds to come up and transfer. During those seconds, and sometimes for longer if the mechanical plant restarts on a staged sequence, the IT equipment is running at full load on battery and producing its full heat output into a room where the chillers, pumps and fans have stopped.
The electrical system is behaving perfectly. The thermal system is in free fall. A facility that has tested generator transfer without watching hall temperature during the transfer has tested half of the event.
The question worth asking of any facility, including when evaluating a colocation provider, is simple: is the mechanical plant on protected power, and if so, which parts. Chillers are often not. Pumps and controls sometimes are. The answer determines whether a brief utility event is a non-event or the start of a thermal one.
What sets the number
There is no useful formula to print here, because the inputs vary so widely between halls that any worked example would mislead more readers than it helped. What is worth knowing is which inputs matter, because they are the things you can change.
- Heat load in the affected space. Not contracted capacity. Actual measured draw, which on a busy AI cluster may be close to nameplate and on a mixed hall may be well below it.
- Air volume that can participate. Containment, hot aisle geometry and whether doors can be opened all change this, and some of it is available as a deliberate emergency action.
- Water volume in the loops. Chilled water systems carry real thermal mass. Thermal storage, where it exists, carries considerably more and is the main reason some sites ride through plant failures comfortably.
- Headroom to the limit. A hall run at a conservative supply temperature has further to climb than one run warm for efficiency. This is a genuine trade-off: running warm saves energy every day and costs you minutes on the worst day.
- What the limit actually is. Throttling, shutdown and damage are three different thresholds. Hitting the first is an availability event. Hitting the third is a capital event.
The relationship between supply temperature and resilience deserves emphasis, because efficiency programs push in one direction and almost nothing pushes back. Every degree of headroom given up to free cooling and adiabatic operation is a degree that is no longer available as time during a failure. That is not an argument against efficiency. It is an argument for knowing the price.
Climate makes this a regional question
In a temperate climate, outside air is often a usable emergency resource. Open the building and ambient air will, at least for part of the year, do something helpful.
In the Gulf that option largely disappears for much of the year. When outdoor dry bulb exceeds the temperature you are trying to hold indoors, outside air is not a mitigation; it is an additional load. The emergency playbook that works in a mild climate has to be rewritten for one where ambient is above the hall setpoint for months at a time, and any plan that quietly assumes fresh air is a fallback should be checked against the design conditions covered in our note on chiller plant design for hyperscale facilities.
Finding out, and what to do with the answer
Thermal ride-through is measurable, and measuring it is the only honest way to know it. The practical approach is to establish it under controlled conditions during commissioning or a planned maintenance window, with the load that is actually in the hall, monitoring intake temperatures at the worst-positioned racks rather than at a room sensor that will lag.
Commissioning is the natural place for this, and a facility that has run the test will have the data from it; our guide to commissioning and acceptance testing covers where it fits among the other scenarios worth proving.
The result is useful in three ways. It tells operations how long they actually have, which turns an alarm from an emergency into a procedure. It tells the business whether an orderly shutdown is even achievable in the time available, which for a large cluster is not a given. And it exposes the gap between what the design assumed and what the hall does, which is frequently where the surprise lives.
A facility that cannot tell you its thermal ride-through is not necessarily a bad facility. It is a facility where nobody has asked, which is most of them. The question is worth asking anyway, because the answer does not change when it goes unasked. It only arrives later and less conveniently.
If you are placing high-density or liquid-cooled equipment and want to know how a site behaves when cooling stops rather than when everything works, talk to our team. It is a reasonable question and the answer should contain numbers.
Frequently Asked Questions
What is thermal ride-through?
It is the time between cooling failing and equipment reaching a temperature at which it throttles or shuts down. It is the thermal equivalent of the electrical ride-through a UPS provides, but it is governed by entirely different factors: the heat being produced, the air and water volume available to absorb it, and how much temperature headroom sits between normal operation and the protective limits. Unlike UPS runtime, it is rarely stated anywhere in a facility's documentation.
Why is it shorter in a high-density hall?
Because the heat load rose faster than the air volume did. Consolidating the same compute into fewer, denser racks does not change the total heat much, but it concentrates it into a smaller space with less surrounding air to absorb it. Containment makes this sharper still: it is designed to stop hot and cold air mixing, which is exactly what you want in normal operation and exactly what removes your buffer when the cold supply stops.
Does a UPS help during a cooling failure?
Only indirectly, and it can make matters worse. A UPS keeps the IT load energized, which means it keeps producing heat. If cooling is on utility power and the IT load is on battery, an outage can produce the worst case: full heat production into a room with no heat removal. Whether cooling equipment is on protected power is a question worth asking of any facility.
How do we find out what our own ride-through actually is?
Measure it, under controlled conditions, as part of commissioning or a planned test. Thermal behavior depends on too many site-specific variables to be usefully estimated from a formula, and a facility that has never measured it does not know the answer either. The result is one of the more useful numbers an operations team can hold.
A Real Energisation Date, Not a Brochure
Rax operates where transmission investment has run ahead of load, and we are direct about staged capacity and real dates. Talk to our team about sizing a deployment against a timeline.
Contact Rax Data & Energy