At the core of the AI revolution is the rapid advancement in semi-conductor technology. Over the past two years, GPUs have become substantially more powerful, leading to higher rack densities ranging from 40 kW to 130 kW per rack, with future chips projected to reach an astounding 250 kW per rack.

GPU innovation presents a significant hurdle: managing the heat generated by densely packed, energy-intensive GPUs. The necessity to keep this tech cooled and load variability stable, combined with new power usage effectiveness (PUE) regulations, will shift thermal management strategies toward liquid cooling as the standard for new data center developments. In the future, immersion cooling will become a common solution as GPUs surge past 150 kW.

Most new data centers are being designed to house a combination of both AI and traditional workloads. Though a significant driver, even optimistic adoption scenarios suggest that AI will represent less than 50% of data center demand in 2030, with traditional, lower-intensity workloads like data storage and cloud-based applications comprising most of the demand.

“While not every data center is or will be a specialized AI facility, all data centers – new and existing – can benefit from more energy efficient operations and improved technology integration,” said Andrew Green, JLL Regional Data Center Practice Lead, Asia Pacific. “Data center operators must contend with the demand for massive power needs while satisfying the need for more energy efficient facilities. AI is transforming data center management through predictive maintenance applications, which optimize energy usage, lead to longer lifespans for equipment and result in less down time.”