Getting data centers ready for the AI boom
With the rapid improvements in AI capabilities, are data centers doing enough to keep pace?
The buzz around artificial intelligence (AI) that erupted in the past year turned commercial real estate’s attention on data centers: namely whether they could support the pace of AI’s quickening advance, from building design and rack densities to its environmental impact and compliance with evolving government regulations.
“Data centers only rose to prominence for AI in 2024, though they have long been the backbone supporting AI's progress,” says Dr. Glen Duncan, Head of Data Center Research, Asia Pacific (APAC), JLL.
“The availability of AI to the masses drove the rapid and profound growth of data centers, raising concerns about their ability to handle the immense computational needs,” says Duncan.
The primary focus lies in the rack densities, or the amount of power the equipment within a server rack uses.
Companies that operate data centers have ramped up rack densities from 3 kilowatts (kW) a decade ago to the current average deployment of 10kW per rack. Still, that’s barely sufficient for AI and high-performance computing, which require rack densities of up to 100kW per rack.
Some operators are already leading the charge. Take data center giant Digital Realty, which recently launched a high-density colocation service for workloads of up to 70kW per rack.
“Data centers will have to undergo design and structural updates to accommodate AI workloads,” says Celina Chua, Data Center Client Solutions Director, Capital Markets, APAC, JLL. “These modifications must cater to the increased power consumption, floor loading, and heat generation associated with higher rack densities.”
The evolution of design
Effectively managing the degree of rack density involves tackling the critical issue of cooling, which directly influences the design of a data center.
Most high-density racks can only function with the help of liquid cooling technologies because conventional air-cooled methods in existing data centers often fall short in efficiently handling the heat generated by high-performance AI chips.
However, a majority of data centers lack the liquid delivery systems or mechanisms necessary for liquid cooling, says Duncan. “There are also lingering concerns around the storage and transportation of the cooling liquids used.”
Still relatively nascent, liquid cooling is widely considered a more energy-efficient alternative to air cooling. For instance, a new liquid immersion cooling system, developed by Mitsubishi Heavy Industries, reduced the energy consumption for cooling by 92% in a data center operated by Japanese telecommunications giant NTT.
“Retrofitting existing data centers to integrate liquid cooling also may not be as straightforward, or as financially viable, compared to constructing purpose-built AI data centers from the ground up,” Chua says.
Another major barrier hindering adoption is the issue of maintenance. “In most cases, liquid cooling poses a challenge for the concurrent maintainability of the server racks as they must be powered off and removed during the process,” says Chua.
As rack densities grow further, however, the adoption of liquid cooling as the main solution over air cooling has become “inevitable”, Chua says. “It’s still the most energy-efficient and viable alternative to cool the servers effectively.”
Balancing operations and regulations
The emphasis on efficient energy usage comes amid mounting pressure on data centers to adhere to more stringent environmental, social and governance (ESG) regulations.
In Australia, for instance, data center providers pursuing government contracts must have a five-star National Australian Built Environment Rating System (NABERS) rating, and a target power usage effectiveness (PUE) below 1.4.
The German government recently approved legislation requiring data centers opening from July 2026 to achieve a PUE of 1.2, and to implement measures to reuse at least 20% of their waste heat by 2028.
“Regulations around data centers will continue to build out in every country, be it on ESG or other aspects like data privacy and data sovereignty,” says Duncan. “Countries that are lagging behind on legislation will have to catch up eventually.”
Demand on the rise
Despite the tightening regulations, investor appetite for data centers has been strong.
In the first nine months of 2023, investors spent $2.2 billion in data center deals in APAC, just $100 million short of the full-year volumes recorded in 2022, JLL data shows.
“The region hasn’t yet experienced the explosive growth in AI adoption and demand seen in the U.S. and EMEA, which has led to significant investments,” says Chua. “But there’s been signs of land-grabbing in some APAC markets due to the scarcity of land for development amid rising demand.”
Capacity is increasing as operators press forward to satiate the appetite for more data.
The presence of a business-friendly environment and robust government policies that support the data center industry strengthen the appeal of new markets to investors and operators.
Markets like Indonesia, India, and Malaysia have been among the fastest growing in data center capacity, according to estimates by independent research and consulting firm Structure Research.
In 2023, APAC witnessed a series of noteworthy developments, including plans by Singapore’s ST Telemedia Global Data Centres to establish a 120MW data center campus in Malaysia’s Johor state.
“The development of data centers has progressed from building big with hyperscale data centers to building small with edge data centers,” says Chua. “Today, the focus has shifted to building bigger data centers at speed, particularly to meet the demands of AI.”