Dive Brief:
- On-premise data centers account for 40% of global compute capacity, only slightly more than the 37% owned by hyperscaler cloud providers, according to a Synergy Research Group report published Wednesday. The remaining 23% of capacity rests offsite in colocation facilities and leased data centers, the report said.
- The total capacity of all data centers is expected to rise steadily over the next five years, driven primarily by a doubling in hyperscaler infrastructure, according to the firm's analysis of 19 major cloud and internet service firms, including the largest operators in SaaS, IaaS, PaaS, search, social networking, e-commerce and gaming.
- While on-premise capacity is expected to hold steady in the near term, its share of overall compute will decline, as new workloads migrate steadily to cloud. “On-premise data centers will not disappear any time soon, but their scale is being increasingly dwarfed by hyperscale and colocation companies,” the report said.
Dive Insight:
Global computing is moving toward more servers with more capacity run by giant CSPs and smaller third-party vendors, with enterprises continuing to increase investments in cloud infrastructure and other related services.
Annual global spending on cloud reached $227 billion last year, growing at an average rate of 42% per year, the report said. That represents a complete reversal from just a decade ago, when on-prem IT hardware and software consumed over $80 billion annually and cloud spend stood at $10 billion.
While investment in on-prem data center equipment and software hasn’t ground to a halt, it’s grown at a paltry rate of just 2% per year, as an expanding proportion of infrastructure spending has been pushed off-site to colocation facilities, the report found.
Generative AI technologies are poised to further skew the balance toward cloud by inflating the cost of maintaining on-prem compute.
As large training models come online, hyperscaler such as AWS, Microsoft and Google Cloud are retooling their data centers, upgrading servers with GPUs and other high-capacity chips designed to handle massive workloads more efficiently.
“The chips are getting more powerful and more expensive, which then percolates up through the chain [to] more expensive servers, more expensive populated equipment racks and more expensive data centers,” John Dinsdale, chief analyst at Synergy Research Group, said in an email.
“The more powerful chips also consume more energy, so energy costs are rising,” said Dinsdale.
The writing is on the data center walls. Five years ago, nearly 60% of compute capacity lived on-prem, the report found. By 2028, the research firm expects more than half of computing will take place in the cloud as on-prem’s share of capacity drops below 30%.