Efficiencies and better hardware prompt Azure to extend server-refresh cycles from four years to six.
Cloud service providers have been on a hardware spending spree for years, deploying hundreds of thousands of servers as they build out data centers the size of football stadiums as fast as they can.
But the party may be ending. On its recent earnings call with financial analysts, Microsoft announced plans to extend the lifespan of its cloud servers from four years to six years.
CFO Amy Hood said the reason for the longer deployments is due to “increased efficiencies in how we operate our server and network equipment as well as advances in technology have resulted in lives extending beyond historical accounting useful lives.”
The dollar savings can’t hurt, either. Hood said the slower purchasing pace could save the company $3.7 billion in fiscal year 2023 alone.
And Microsoft isn’t alone. Both Amazon Web Services and Google said on respective earnings calls this February that they had extended their hardware lifespans by a year.
Microsoft declined to comment further on Hood’s statement, but it wouldn’t be surprising if, at some point, it publicly shared some of the efficiencies she cited.
Many enterprises already have longer refresh cycles than cloud-providers. A 2021 IDC survey found 10% of enterprises surveyed kept their servers for primary workloads for five years or longer, and 19% kept servers running secondary/less critical workloads for five years or longer.
Some of Microsoft’s lessons-learned might help enterprise data-center pros, despite the very significant differences between enterprise data centers and those of cloud providers.
Enterprises have warranties for their hardware while cloud service providers usually do not, says Ashish Nadkarni, group vice president in IDC’s worldwide infrastructure practice. CSPs use commodity hardware at such scale that if one fails, it’s repaired or replaced by extensive in-house staff. Enterprises generally rely on service contracts with vendors. Hyperscale data centers are also models of redundancy, while enterprises may have financial pressures that dictate less resiliency.
Over time, changing acquisition models and better technology have helped enterprises extend the useful lives of servers. Many enterprise data centers are moving from buying servers outright (capex model) to leasing (opex model) through programs like HPE GreenLake and Dell Apex that take responsibility for the gear.
Today’s servers are more durable than those of a decade ago, Nadkami says. “I think the mean time between failures is much higher now, especially component failures that would render the server basically inoperative, like hard drives failing, memory failing. So, the overall lifespan of a server, the overall uptime that you get from a server has gone up significantly,” he said. The shift from mechanical hard drives and all their moving parts to SSDs has definitely contributed to longer lifespans because hard drives were the most likely to break than any other component.
Today’s servers have more efficient heat management, too. “Fan design has improved, and system design has improved, especially now with different kinds of cooling techniques, including liquid cooling. So, silicon doesn’t fail that quickly,” he said.