Executive Summary
- Modular cooling is a promising solution to the thermal management issues of the high-density hardware demands that data centres are facing with the growth of AI.
- In the AI race, speed to market is a common competitive issue and pre-fabricated modular cooling units enable faster installation and deployment time, albeit with higher capital costs and complex maintenance routines.
- Operators must be strategic with their thermal management, now more than ever, if they want to be able to compete in the AI race, and modular cooling is a promising strategy for this.
The growing demand for AI applications has far exceeded anyone’s wildest fantasies, but it’s no secret that it’s presented several cooling solution issues; with environmental debates around water usage, liquid cooling becoming the new baseline, exploring hybrid cooling, it’s safe to say that AI has placed significant strain on infrastructure and cooling.
Data centres built seven years ago aren’t equipped to handle the high-density hardware demands of today, let alone tomorrow.
Modular units now enter the room, as a promising solution to the flexibility and scalability issues caused by AI’s power demand and having the infrastructure that can cope, keep up and evolve as needed.
What is modular cooling?
Modular cooling is the use of pre-fabricated, standardised cooling units as a thermal management system that is flexible and scalable; more units can be added to handle extra demand, or units can be removed or switched off during times of low demand.
A typical modular system comprises of indepedent modules that each contain:
- Compressor
- Condensor
- Evaporator
- Control system
The system is usually managed by a central integrated controller for unified operation and shares water pumps, cooling towers and piping.
Benefits of modular cooling
Speed to market is a pressing issue in the AI race; it takes time to build data centres, but this is where modular cooling comes in. It enables faster deployment times in a market that’s moving at the pace of a wild forest fire. The plug-and-play capability of modular cooling enables data centres to significantly reduce installation time and complete in a matter of weeks, rather than years.
Another unique benefit is their scalability, which enables operators to start on a smaller scale and get to market faster, and increase incrementally as and when it’s needed. This reduces overheads and allows operators to expand their capacity in a timely manner to avoid overinvestment. The flexibility of adding and removing modular units aligns business growth and infrastructure spending well.
Modular units are equipped to handle the high-density AI workloads and intense power demands on racks and can complement a hybrid cooling strategy well. For example, for your traditional servers that are less power hungry, you can deploy air cooling solutions, whereas in the next room, with your 100kW+ racks handling complex AI workloads, you can deploy the necessary amount of units in a pod to handle heat management.
Modular cooling does have its drawbacks
Modular cooling systems are a product, not a project, so they’re ready-made for you on the shelf, which is a higher up-front cost, especially for liquid cooling units, with reports indicating it can be up to three or four times higher than air cooling capital costs.
Their physical footprint must be kept in mind as the space they take up can determine whether or not it’s a viable strategy to employ to manage thermal demands, particularly if the facility has been retrofitted, as that places strain on design.
With several units as part of a larger cooling system strategy comes complex maintenance. Regular servicing becomes a necessity to ensure all of the components, piping and heat pumps are operational, because just one failure in one unit in the system could be critical to the entire cooling system.
Lastly, choosing a manufacturer for your cooling solutions gives a “locked-in” effect and makes switching to another company in the future a complex process due to proprietary connections.
Cooling needs to be more strategic
Every day presents new challenges, more demand and more growth opportunities, and to have a space in this AI race, operators must be strategic with their cooling. Hybrid cooling strategies to thermal management that can scale and evolve with the peaks and troughs of demand 24/7 will triumph – and modular cooling units will have a huge part to play in this.
By productising the cooling loop, operators gain speed to deploy AI-ready infrastructure within months rather than years, the scalability to meet demand where it’s at in real-time and the flexibility to operate liquid and air cooling without rebuilding their footprint from scratch.
The cost per unit is high, but the cost of not being able to operate smoothly and cope with AI’s appetite is much higher.
