Sizing Thermal Ride-Through for AI Data Centers
Key Takeaways
- Size thermal storage for the heat it must absorb during the cooling shortfall.
- Tank size alone does not determine how much cooling protection is available.
- Storage can cover a temporary cooling shortfall, but it cannot compensate for insufficient ongoing cooling capacity.
- Locate storage where it can deliver cooling to the protected equipment within the required response time.
- Commissioning should verify that the system maintains required temperatures for the specified duration under defined test conditions.
AI data centers can have enough total cooling capacity and still experience brief periods of instability. Dense racks may add heat to the technical cooling loop before pumps, valves, heat exchangers, and upstream cooling equipment have fully adjusted. The plant can eventually catch up, but the racks need protection with less lag than the plant can usually support.
Thermal storage can bridge that gap, but only when it is sized around the event it is expected to cover. A standard tank volume or generic number of minutes does not account for the size of the load change, the cooling system’s response time, or the amount of temperature rise the equipment can tolerate. Data center teams should define those conditions first, then determine how much usable cooling reserve is needed.
What Is Thermal Ride-Through in a Liquid-Cooled Data Center?
Thermal ride-through is the period during which the cooling system can keep IT equipment within its required temperature range after cooling performance degrades or a load changes faster than the upstream plant can respond. This gives the cooling system time to recover or adjust to the new load, but it cannot keep equipment within that temperature range indefinitely.
OCP liquid-distribution guidance identifies ride-through as a consideration for loss-of-flow events, operating procedures, and service-level expectations. Designers must determine how much excess heat the local loop must absorb before the rest of the cooling system stabilizes.
If rack heat rises before upstream cooling can respond, the local loop must absorb the excess heat during the delay. Thermal storage will absorb that heat as water from storage reaches the load at the required temperature and flow.
Why Is a Generic Ride-Through Duration a Weak Design Basis?
A generic ride-through duration is a weak design basis because time alone does not tell designers how much heat the storage must absorb. Specifying ten minutes of protection, for example, sets a duration but leaves out the heat load, how much cooling remains available, and how long it will take to recharge the tanks capacity. Designers need those conditions to calculate how much storage will provide the specified ten minutes of protection.
A universal duration can also hide the actual problem. If the disturbance is a sustained shortage of plant capacity, local storage only delays the inevitable. If the disturbance is a brief control or equipment transition, a smaller amount of well-placed storage may be sufficient. Designers need to establish what causes the cooling shortfall and how long it will last before choosing a tank size.
Which Inputs Determine Required Thermal Storage?
Required thermal storage is determined by the protected heat load, credible load step, upstream response time, allowable coolant-temperature rise, coolant density and specific heat, loop losses, and the portion of the stored water that remains cool enough to use. Flow and pressure constraints determine whether that stored energy can reach the load when needed.
For an initial calculation, engineers can evaluate the difference between heat entering the loop and cooling available over time and relate it to the mass of coolant and permissible temperature change. That calculation is a starting point. Tank mixing, piping transport, heat-exchanger response, and control behavior must then be incorporated into the project model.
How Should Designers Separate Temporary Cooling Shortfalls From A Capacity Deficit?
Designers can tell the difference by checking whether the cooling system can handle the new heat load once it has fully responded. If it can, the shortfall is temporary, and thermal storage can provide cooling while the system catches up. If the system still cannot remove heat as fast as the equipment produces it, the system is undersized for that load. Storage can delay the temperature rise, but it cannot make up for that ongoing shortage of cooling capacity.
This distinction protects projects from using storage to mask undersized infrastructure. A design should show what happens after the tank discharges: whether the plant has caught up, whether the storage can recharge, and whether repeated events can occur faster than recovery.
Where Should Ride-Through Capacity Be Located?
Ride-through capacity should be located where it can support the protected load within the required response time and without creating unacceptable space, service, or leak risk. Storage in the technology cooling loop can reduce the transport distance to high-density racks. Plant-level storage can support broader load management and may be easier to service.
The answer may be centralized, distributed, or hybrid. Hydraulic separation, redundancy boundaries, ownership, floor loading, containment, maintenance access, and available temperature difference all affect the decision. Proximity is valuable only when the surrounding system can use the stored cooling effectively.
How Should Uncertainty Be Included in the Calculation?
Uncertainty should be included through bounded scenarios rather than an arbitrary multiplier. Teams can test the expected rack mix, a planned refresh, simultaneous load changes, degraded plant response, and a high-density sensitivity. The range reveals which assumptions dominate tank size.
Google’s public roadmap for racks up to 1 MW illustrates the direction of infrastructure development, but it should not be assigned to every project. Scenario analysis preserves readiness without pretending that one vendor roadmap is the final installed condition.
How Can Commissioning Prove Thermal Ride-Through?
Commissioning can prove thermal ride-through by applying a known heat imbalance and measuring supply temperature, flow, pressure, tank discharge behavior, and the time until upstream equipment catches up. The observed result should be compared with the sizing model and acceptance limit.
A simple way to think about the sizing problem is to separate heat, time, and allowable temperature change. The greater the extra heat entering the loop, the more stored cooling is required. The longer the upstream system takes to respond, the longer that reserve must last. The wider the temperature range the racks can accept, the more energy a given amount of water can absorb.
This relationship explains why a single rule of thumb is unreliable. Two facilities can protect the same rack load but need different storage because one plant responds faster or one equipment design allows a different supply-water range. Flow also matters. Stored cooling has little value if the hydraulic path cannot deliver it to the affected racks. Engineers will still need project-specific calculations and modeling, but the plain-language decision remains accessible: define how much heat arrives, how long the system must bridge it, and what temperature limit cannot be crossed.
What Should Teams Define Before Sizing a Thermal Storage Tank?
Data center teams should define the transient before selecting the tank. Identify the load increase that needs to be covered, how long the upstream cooling system takes to respond, and the maximum temperature change the protected equipment can tolerate. Then determine how much stored cooling remains usable under the expected flow conditions.
That sequence prevents a tank-size discussion from replacing the underlying engineering question. Thermal storage is valuable when it bridges a defined timing gap. It is not a substitute for adequate sustained plant capacity. The decision to carry forward is therefore not simply how many gallons to add, but what event the storage must cover, where the reserve must be located, and how the finished system will prove that it delivers the expected protection.
About the Author
|
|
Greg Berry Director of Sales, CoolVault™ Thermal Energy Storage Solutions Bringing more than 25 years of experience in sales, business development, and technology solutions for industrial and enterprise customers, Greg Berry works with customers, engineering firms, and strategic partners to improve cooling efficiency, operational resiliency, and sustainability. ludes leadership roles at Watts, Carrier, and Philips, and he holds a Bachelor of Science in Electrical Engineering Technology from Purdue University. |