In early September 2023, Forbes reported that AI and liquid cooling are the future of data center development. While at nearly the same time Fortune mentioned that AI fueled a 34% increase in Microsoft’s water consumption at their AI data center in Iowa and now citizens are concerned about the effect on the residential supply of water. Clearly, finding solutions for more efficient and effective cooling for data centers is quickly growing from an “inside industry” issue to one that is making its way to Main Street.

Colocation data center providers have been supplying the market with the space for their data-driven needs using traditional methods, such as air-cooled chillers, to cool vertical rack systems down to 67 degrees F. In some circumstances even down to 50 degrees F to avoid hot spots from a high density of server racks with intense computing power. Given the heat that the equipment is generating, 130+ degrees F, this takes a great deal of energy to lower the spaces to optimal operating temperatures.

With immersion cooling, which places components side by side in a sealed cabinet and then submerges them in a liquid substance (generally mineral oil-based), cooling temperatures can be maintained at a much higher levels – over 100 degrees F in some cases. Heat is directly transferred into the cooling fluid as opposed to being transferred into air which is then transferred into a cooling fluid. So, if the temperature of the liquid when it passed over the equipment raises to 130 degrees when it goes to the chiller or fluid cooler, it only has 27-30 degrees to drop versus room air cooled methods that must reduce it 60 degrees or more. The cost savings are immense – it can reduce data center operating costs by more than 90% and capital expenses by more than 50%(1) . Additionally, since the fluid is much more effective at conducting heat than air, it requires less space for distribution within the white space, which allows for greater space to accommodate these high power servers.

Why isn’t everyone using it? While immersion cooling is not particularly new – most current applications trace back to start-ups from 2017(2), it is still considered to be in the early adopter phase. Most use has been from big tech companies, or hyper-scalers, who are building their own massive data centers like Amazon Web Services (AWS), Microsoft Azure and Google Cloud who have the funding to explore alternative technologies for cooling and space conservation.

Now, with the explosion of AI in niche markets and analysts predicting that “the technology will be mainstream within the next four years – growing from $251 million in 2021 to over $1.6 Billion by 2027”(3), colocation providers can see that the shift is coming. Many are already starting to see demand for alternative cooling methods, especially as load density increases (+50kW per cabinet). Forward thinking developers are looking at their spaces to understand if they can create a hybrid environment that allows them to provide immersion technologies alongside legacy technologies, without having to replace their entire operation. The short answer is they can, with the proper analysis of current capacities and infrastructure, and development of tailored designs specifically for dual cooling capabilities with provisions for conversion and scale up in the future to accommodate these new technologies.

Practical Considerations for Creating a Hybrid Data Center Space may include:

1. Hybrid Cooling Plants: Scalable plants that utilize a combination of fluid coolers and chillers.

1. Fluid coolers and chillers are piped together. The main cooling loop would reject heat through the fluid coolers utilizing chillers to inject cooler water as needed to maintain required supply distribution temperatures. Under this arrangement, thermal efficiency is sacrificed to minimize initial costs and to maximize flexibility.
2. Fluid coolers and chillers operate in separate cooling distribution loops to allow each to operate at the desired fluid temperatures. This arrangement maximizes thermal efficiency, minimizing operational costs, but requires greater upfront costs to install.
3. Fluid coolers may be selected as dry coolers since they will operate at higher temperatures and will experience additional free cooling hours. However, if non-potable water is available and inexpensive, then the fluid coolers may be selected as a hybrid type that allows for the implementation of evaporative cooling when ambient temperatures are at peak levels.

2. Flexible Distribution Loops:

1. Multiple, dedicated piping loops are combined at the plant level to distribute cooling water at two different temperatures – one set for lower temperatures associated with traditional legacy equipment and one for equipment that will operate at higher temperatures.
2. Segregated piping systems with one for equipment served by lower water temperatures and one for higher temperatures.
3. If a segregated loop approach is used, strategically placed isolated cross ties could be employed to allow for future operation as a single system when spaces are developed or converted for immersion cooling applications.
4. Localized heat exchangers and pumps are utilized to transfer heat between the plant cooling water and the immersion cooling fluid.
5. To minimize costs it is recommended that specific areas be initially identified as high density, immersion cooling spaces in an effort to minimize the expanse of the distribution loop(s) required on Day 1.

3. Flexible White Space:

1. Eliminate raised floors. Develop cooling, power and communication systems designs based on a non-raised floor to accommodate the heavier, dense structural loads associated with immersion
   cooling technologies. Eliminating the raised floor avoids the necessity of strengthening it for future immersion cooling equipment weights that may or may not be deployed in a given white space, while also reducing the cost to install a raised floor.
2. Plan dedicated space for localized heat exchangers and pumps that are required for immersion cooling systems. These may be located within dedicated CRAH unit galleries adjacent to white spaces or within the white space itself. If placed within CRAH galleries, the design should accommodate the phasing from legacy cooling to immersion cooling as it may be necessary to remove and redeploy
   CRAH units to other white spaces to accommodate the localized heat exchangers and pumps.

Profit margins for Colocation developers are currently being heavily impacted by large cloud providers who “do it all” and by providing the option of immersion cooling to tenants they will be able to charge a premium for a portion of their available space, without displacing current tenants who do not have that need … yet. As the demand for immersion cooling options grow, they can expand that space footprint to accommodate new installations and eventually convert fully if the need dictates. Meanwhile, they will also have a marketing advantage with prospective tenants to show they are staying on the cutting edge of data center development. When considering this option, it is important to engage engineers who are fluent in the considerations for a hybrid data center to achieve a successful transition of the facility.

As the demand for more efficient cooling systems continues to grow as a result of higher load densities to support AI, immersion cooling technologies will become more prevalent and cost effective resulting in more deployments by a wider range of colocation developers and enterprise data centers throughout the industry.

Scott Davis, PE is VP & Director of Mechanical Engineering at Bala Consulting Engineers, Inc. He can be reached at smd@bala.com.

Read Other Articles