Thermal Storage and AI Rack Roadmap Risk
Key Takeaways
- AI hardware can change before a data center opens, changing both cooling demand and how cooling must be delivered.
- Thermal storage can cover short cooling-response delays, but ongoing capacity shortages and flow, temperature, or control problems require structural fixes.
- Reserve space and plan piping connections, controls, power, and equipment access during early design; select equipment as requirements become clearer.
- Size storage for the cooling needed during a structural demand response, and check that it can recharge before demand rises again.
- Test several likely rack configurations to determine whether the facility needs thermal storage, more cooling capacity, or both.
Data centers are built on a longer timeline than the AI hardware they support. A rack configuration selected during early design may no longer represent the equipment installed when the facility reaches full operation. Cooling requirements can shift with it, including flow, temperature, pressure, and how quickly loads change. That does not mean every project should build immediately for the highest projected rack power demand. Instead, design should allow for upgrades as rack requirements become clearer. Thermal storage can be one of those options when the likely challenge is a short-term gap between rack demand and cooling-system response.
Why Are Facility Designs and AI Rack Roadmaps Misaligned?
Facility designs and AI rack roadmaps are becoming misaligned because buildings, utility service, and mechanical systems require long development cycles while AI computing hardware continues to quickly evolve. A rack configuration assumed during concept design may no longer represent the first full deployment.
New AI hardware can change both how much cooling a facility needs and how that cooling must be delivered. It may require different coolant flow rates, temperatures, or pressures, concentrate more heat in one area, or cause cooling demand to change more quickly. A design based only on the original rack plan may not meet the requirements of the equipment ultimately installed. Building for the highest published projections, however, can leave the owner paying for unused capacity.
During design, projects can set aside space and plan piping connections for a future thermal storage system. Owners can then decide whether to install it once the rack configuration and cooling requirements are confirmed.
How Much Power Will Future AI Racks Need?
Some future AI racks could require more than 500 kW of power before 2030, according to Google’s projections for machine-learning racks. Google has also described power delivery infrastructure that can scale from 100 kW up to 1 MW per rack. These projections and capabilities point to higher rack power demands, but a facility’s requirements will depend on the equipment installed.
For cooling design, teams need to evaluate both the total heat that equipment will produce and how quickly cooling demand could change. Sustained demand determines the cooling capacity required. Short-term changes help determine whether thermal storage could bridge a delay in the cooling system’s response.
How Can Thermal Storage Add Flexibility Without Overbuilding?
Thermal storage allows for flexibility by providing extra cooling for a short time when rack demand changes faster than the cooling system can respond. This gives the system time to adjust without requiring every component to respond immediately. It can also provide a temporary cooling reserve in high-density areas while teams confirm how the installed racks perform.
Storage cannot make up for a cooling system that is too small to meet ongoing demand. The design should show how much extra cooling storage can provide, how long it can provide it, and what is needed to recharge it. It should also identify when a larger chiller, heat exchanger, pump, or piping upgrade is still needed.
How Can Projects Plan for Future Cooling Upgrades?
Projects can plan for future cooling upgrades by reserving space, checking that the fluid infrastructure can support added equipment, and planning piping routes and connections during design. The plans should also account for valves, controls, power, drainage, spill containment, and leak detection, along with access for maintenance and equipment removal. These provisions are difficult or expensive to add later.
OCP guidance for liquid-cooling distribution takes a similar approach: leave room for future piping, provide connection points, and plan for equipment changes over the facility’s life. Applying this approach to thermal storage allows owners to prepare for future upgrades without buying every component at the start.
How Should Teams Test Future Cooling Needs?
Teams should evaluate several possible rack configurations to check whether the cooling system can meet both ongoing demand and sudden increases. Compare the racks expected at opening, a planned equipment upgrade, and an area with higher-power racks. For each case, check cooling capacity, coolant flow, pressure, and temperature, including when cooling equipment is unavailable.
Next, check how quickly the cooling system responds when several racks increase demand at once. If thermal storage is needed to cover a delay, determine how much cooling it must provide, for how long, and whether it can recharge before demand rises again. These checks show whether each rack configuration requires thermal storage, more cooling capacity, or both.
When Is Thermal Storage the Wrong Tool?
Thermal storage is the wrong tool when the primary constraint is sustained heat rejection, undersized piping, chronic low flow, excessive pressure drop, unsuitable coolant temperature, or a control defect that should be corrected directly. Adding volume to those problems can delay diagnosis without solving them.
A disciplined study identifies the binding constraint first. Storage is justified when the value comes from time: bridging a transition, limiting a short excursion, or protecting against a known delay in upstream response.
When Should You Start Planning for Changing Cooling Needs?
Start planning during early facility design, before the final rack configuration is known. AI hardware may change before the data center is fully operational, so reserve space and plan piping connections, controls, power, and equipment access while these provisions are easier to include. Equipment can be selected later as cooling requirements become clearer.
Use tests of likely rack configurations to decide which upgrades are needed. Thermal storage can provide temporary cooling while the system responds to a rise in demand; demand that stays above the system’s capacity requires a capacity upgrade. Planning for both possibilities early helps owners adapt to new hardware without buying equipment before it is needed.
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. |