Analysts expect the global capacity of hyperscale data centers will grow almost threefold in the next six years to meet the needs of generative artificial intelligence (genAI). According to data from Synergy Research Group (SRG), more than 900 hyperscale data centers are in operation today and an additional few hundred are in the works.

The firm analyzed the data center footprints of 19 hyperscale cloud service providers across software-as-a-service (SaaS), platform-as-a-service (PaaS), infrastructure-as-a-service (IaaS), social networking, ecommerce and gaming. SRG Chief Analyst and Research Director John Dinsdale told SDxCentral those providers account for 926 existing data centers, with another 427 in the pipeline.

Major hyperscale cloud providers like Amazon Web Services (AWS), Microsoft, Google and Meta have all maintained a strong pipeline of future data centers in various stages of planning, construction, equipping and commissioning, which means they'll be well-positioned to keep up with the impact of genAI technology and the subsequent need for more powerful data centers. "They were all planning for continued aggressive growth in their data center footprint," Dinsdale said.

Along with new construction, the firm predicts there will be a certain degree of retrofitting existing data centers to increase capacity. Especially for data center operators with large global networks and long-term growth plans, like Equinix or Digital Realty, "retrofitting will certainly be one of the ways in which they meet increased demand," he said. The process of retrofitting a legacy data center "runs the gamut from easy to impossible depending on location, the nature of the existing facility and the availability of power sources."

Overall, new builds and retrofitted facilities will make up a three-times increase in global data center capacity in the next six years, according to SRG.

Although the hyperscale providers were ready for the infrastructure impacts of genAI, "there may have been some level of underestimating the power density required in AI-oriented facilities," Dinsdale said. In addition to capacity, high-performance computing (HPC) capabilities required for AI training and inferencing needs more energy to run (and cool) infrastructure.

To prepare for increased energy consumption, hyperscalers are "doing a great job of making their data center operations as efficient as possible, and they've made big strides toward using renewable energy wherever possible," he said. While he anticipates locals may resist new data center builds, "this is mostly manageable, and usually there are alternative locations that can be used if necessary. In some specific locations there will be short-term constraints in the availability of large amounts of additional power, but usually these can be worked around or different locations can be targeted."

"Basically, money shouts, and there is a lot of money being spent by hyperscale operators on data center construction. That tends to mean that solutions will always be found," Dinsdale said.