In the new digital economy, artificial intelligence has taken on an increasing role in business operations and captured the attention of both the public and corporations. No doubt, data center owners and operators are closely watching industry chatter, worried about the evolution of their facilities to stay competitive in the 'arms race' for AI readiness.

At a recent Uptime Institute Leadership Conference in Perth, Sanjay Motwani, Legrand's Vice President for Data Center - White Space, APAC, stressed the importance of maintaining a clear head when trying to sort facts from hype.

Too Early to Tell?
It will take time for AI to become more widespread and require more data center resources and support, according to Mr Motwani. Data center managers reported that 87% of their facility rack densities were below 30kW in Uptime Institute's 2023 Global Survey of IT and Data Center Managers.

The Uptime Institute Global Data Center Survey 2024, published in July 2024 by Uptime Institute, confirms this finding by showing that server rack densities average below 8 kW, even as they are increasing, and most facilities do not have racks above 30 kW.

Based on the rack density trends above, the degree of change heat loads will see in the near future are unlikely to grow as quickly as the chatter around AI will lead most to believe.

Another study by Omdia projected that by 2027, data center racks with density between 10kW and 30kW would increase to 90%, with priority placed on power-efficient designs. These densities will be most common in three to five years.

Here are some numbers to help put things in perspective. AI applications fall into three main ranges based on their purpose.
·         Skilling and training: Around 7,200 kW
·         AI as a supportive component of business infrastructure: Around 100kW
·         Applications for end users: 40 to 60 kW

However, the industry also must not forget about non-AI demands on existing DC capacities.

Digital economic consumption – consumer usage of e-commerce, digital financial services, streaming services, cloud services, edtech and more – is also expected to continue growing. The e-Conomy SEA 2023 report by Google, Temasek and Bain & Company forecasts that the internet economy in Southeast Asia alone may reach US $1 trillion by 2030.

Traditional digital infrastructure, including data centers, may not require as much power as AI processes, but they will need to scale up to accommodate the increasing spending, entertainment, work, and learning habits of a digital population.

When considering their next move, operators and designers will need to consider: where do their facilities sit in the local and regional ecosystem? What demands on capacity, capability and management do they need to anticipate?

The Rising Need to Keep Cool
In the process of expanding or upgrading hardware, data centers must consider heat loads.

Cooling technology has changed a lot since the 1960s, when CRACs and chilled water systems were the norm. The 2000s saw an increase in server density, which led cooling systems to focus on temperature control at the rack level, such as in-row cooling systems, liquid cooling, and rear door heat exchangers.

As legislative and regulatory pressure grew in the 2010s, data centers had to balance environmental concerns, sustainability and energy efficiency. A reaction to the rise of high-performance computing (HPC) and early AI explorations, free cooling (air or water), direct-to-chip liquid cooling, and immersion cooling emerged.

High performance computing workloads account for 51% of data center workloads, making efficient and cost-effective cooling a critical consideration. While the industry mulls over its next steps for AI readiness, rear door cooling is likely to be the best strategy for now.

Designed with a small footprint and requiring no other supporting infrastructure, rear door cooling units can be easily deployed on as few or as many racks as needed, making them a flexible solution with plenty of room for future scaling when needed. This makes them ideal for both older facilities looking to upgrade legacy cooling systems, and newer facilities needing the option to scale up as future retrofits take on higher density racks and heat loads.

By cooling air at the point of generation using chilled water systems, these units prevent overcooling and reduce the workload on fans and air distribution systems. This reduces operational costs and power consumption.

For example, the Coldlogik CL23 Rear Door Heat Exchanger from USystems, a brand of Legrand, can support up to a 200kW HPC rack with 14/20°C water supply/return from mechanically cooled external plant. Designed to operate on closed loop water circuits, this rear door system removes heat at the source, allowing facilities to do away with the need for room-based temperature control.

Furthermore, rear door heat exchangers are compatible with sustainability: closed-loop water circuits minimize water waste, and they can be incorporated into sustainable energy sources or waste heat recovery systems to reuse heat.

Conclusion

As the AI revolution unfolds, data centers must separate truth from hype, and adapt to meet the huge power and resource requirements AI processes will require. Also, operators must prepare for the increasing demands on data centers for non-AI processes in the next three to five years, and the need for more efficient, effective, and sustainable cooling systems.