Hyperscalers preparing for AI tasks are increasingly purchasing massive amounts of cloud infrastructure equipment.
In the fourth quarter of 2023, spending on compute and storage infrastructure for cloud deployments increased by 18.5% year on year, according to the most recent version of the IDC Worldwide Quarterly Enterprise Infrastructure Tracker, which was published last week. However, the total number of units sold dropped, demonstrating hyperscalers’ preference for high-capacity, GPU-intensive servers with higher average selling prices.
Lidice Fernandez, a group vice president at IDC and one of the report’s co-authors, attributes the sharp increase in cloud infrastructure spending to AI. Specifically, it’s the quick expansion of AI capability by some of the technology sector’s biggest companies, including major hyperscalers like Google and Microsoft, along with the fact that managing AI-centric workloads requires expensive GPU hardware.
She said, “These large, mega-datacenters are the ones that swing volumes. So this year, we began to see these large hyperscalers and cloud service providers prioritize GPU-rich configurations, which tend to have a higher processing capacity cost.”
At the same time, more expensive, GPU-intense machines are able to process more substantial workloads, indicating that fewer units are needed to reach a certain capacity, according to Fernandez.
“To achieve the same work output, fewer machines are necessary, which is driving the trends we’re observing among the biggest infrastructure purchasers. It’s likely to continue at an exponential rate for just a few more years,” Fernandez said.
In the raw statistics department, cloud infrastructure product sales increased 18.5% year-over-year in the last quarter of 2023, reaching a total of $31.8 billion. According to IDC data, this is higher than the spending on non-cloud computing and storage infrastructure, which increased by 16.4% year-over-year, totaling $18.9 billion. The total individual shipments of cloud equipment fell by 22.8% over the same period.
The huge demand from hyperscalers and major service providers shifted the data more towards public cloud, as indicated by IDC data. The “shared” cloud infrastructure, which mainly represents public cloud deployments, saw a growth of 27% compared to the previous year, reaching $22.8 billion. This figure also accounts for nearly 45% of total global infrastructure expenditure, whether cloud-based or not. On the other hand, “dedicated” cloud, which primarily refers to private cloud infrastructure, witnessed a minor growth of 1.4% during the same period, amounting to $9 billion.
“Most of the spending continues to be concentrated on the public cloud area,” Fernandez said.
Nonetheless, as Fernandez pointed out, private cloud will keep its relevance for a conceivable future, particularly since ordinary cloud users—excluding hyperscalers and service providers—have a significant and potentially increasing demand for it.
“Everything that stays either on-site or in a private cloud is there for a valid reason,” she explained. “In numerous cases, the reasons include privacy, compliance, and security.”
Many mission-critical workloads—especially those subject to industry-specific regulations or new general regulations in regions like the EU—have not only been retained in private or on-premise deployments, but many that were shifted to the public cloud are being “repatriated,” as per Fernandez.
“If you see the numbers overall, they remain relatively stable in terms of spending,” she said. “That infrastructure needs to continue being sustained and deployed and refreshed at a certain pace.”
