Liquid immersion cooling has quickly become an “it” topic within data center design circles. The technique, which cools equipment using specially manufactured dielectric liquid instead of relying on forced convection or a liquid-cooled cold plate, is touted as the next step for high computation density applications and efficiency in mainstream data centers. Still, widespread adoption may not be on the cards anytime soon. Although liquid immersion cooling offers an innovative and effective approach to the pressing concern of climate change (ACM Tch Brief, PDF), the significant barriers to implementation may hinder the industry at large from adopting it as the next step in efficient design.
The idea of using a liquid to cool data center hardware isn’t new, as direct liquid cooling has been a staple of the field for quite some time. Liquid immersion cooling relies upon the same principle but alters the design to fully submerge the servers or other equipment into a bath filled with dielectric fluid like mineral or virgin coconut oil.
The result is more even and consistent cooling of hardware with less waste. Eliminating extraneous cooling architecture like fans and A/C compressors reduces energy use to cut costs, while improved cooling functionality allows operators to get more computing power into the same footprint to meet growing demand. It’s a compelling pitch, but operators considering bringing liquid immersion cooling into their data centers may face a tough road.
Challenges to adopting liquid immersion cooling
At present, the barriers to adoption that operators across the globe will need to overcome are significant. With liquid immersion cooling being a relatively new addition to the data center cooling landscape, those who want to adopt it will need to rethink their operations from the ground up. While some data centers can accommodate liquid immersion cooling equipment within their current designs, just as many may not be able to. Those operators will need to reorganize and reimagine their data center floor plans to accommodate available solutions—or start from scratch.
Then, there’s the cost of procuring the system which will likely prevent many data center operators from transitioning to liquid cooling anytime soon. Although the energy savings and efficiency boost are likely to yield financial benefits in the long run, the price tag on the new equipment may be prohibitive to many at this point. Going all-in on liquid immersion cooling will not be a matter of replacing fans or racks with newer models. Facilities that want to use liquid immersion cooling will need new equipment: servers that are sealed, racks or baths that are compatible with those sealed servers, and any auxiliary equipment needed to install and run liquid immersion cooling setups.
Now, let’s say that a company does invest the resources to overhaul its facility and equipment. It will still need cooling liquid to facilitate the immersion process—and that can be hard to get, especially if the operator adopts a two-phase approach that uses even more specialized (and apparently harmful) materials. Very few plants manufacture two-phase immersion cooling liquid today, making widespread adoption tough to facilitate at this time. An interruption at one of them could (and did) send the market into a shortage, leaving data centers using liquid immersion cooling without alternative options.
Should a company obtain a steady supply of dielectric liquid, it will still need specialists to service and maintain the immersion cooling equipment. Those specialists can be hard to come by due to ongoing labor shortages and the specialized training these technicians require. The data center techs that operators currently rely upon to keep their operations in running order may not be equipped to address issues that can arise within an immersion cooling system. They will likely need to be retrained, supplemented, or replaced—and that means even more time and money being poured into the transition.
Furthermore, the above situation imagines the process for a well-funded, enterprise facility operated by the company utilizing the center for its own operations. In this example, that company has complete control over all the equipment and the facility’s up- and downtime. They can choose to undertake a complete overhaul of an existing facility or build a new one for their own use. However, at that scale, the energy savings the investment yields would likely be small compared to the complexity of the undertaking. The same can be said for existing colocation data centers (colos). The constraints of existing contracts would make retrofitting an existing colo for liquid immersion cooling difficult if not impossible, making new builds the most likely candidates for immersion cooling architecture.
Hyperscale facilities owned and managed by large corporations may be the best positioned for immersion cooling adoption at this time, as their massive scale would actually allow operators to see the energy savings associated with more efficient cooling. Still, shifting these facilities might also be limited to new builds for reasons similar to those described above. Any retrofit would likely require a total renovation of the facility and significant downtime. Companies that want to use immersion cooling at that level would be better off building a new facility that integrates the technology into the initial plans.
Start-ups specializing in liquid immersion cooling solutions are appearing left and right, each offering products and services that promise to save data center operators time, money, and energy. Liquid immersion cooling does work, and it does offer benefits to adopters. That said, its status as the next big leap in data center cooling has yet to be cemented in part due to its relative immaturity on the market.
If the transition were a simple plug-and-play, adopting liquid immersion cooling across the globe would be a no-brainer. However, the time, money, and manpower that must be spent on adoption make the decision a significantly more difficult one. Will the technology be able to overcome the significant barriers to adoption that exist in today’s landscape? Will another, more accessible option come along to take its place? Only time will tell.