Data centers aren’t ready for AI – yet

  • Dell'Oro Group raised its full year growth forecast for data center physical infrastructure

  • Instead of shrinking as expected, backlogs have actually grown

  • Dell'Oro's Lucas Beran said data centers aren't yet equipped to handle massive AI demand and enterprises which jump the gun by ordering too much IT infrastructure could be in for a big surprise

To hear analysts tell it, 2023 was supposed to be a year of digestion and backlog drawdowns. While that may be happening in some sectors, that’s certainly not the case in the data center physical infrastructure (DCPI) market, Dell’Oro Group Research Director Lucas Beran told Silverlinings.

On the contrary, backlogs in this space have actually been increasing and the analyst firm has consequently raised its full year forecast for DCPI revenue growth from 10% to more than 13%.

The time spent chewing on pandemic-era backlogs means data center and colocation providers haven’t yet been able to beef up their facilities to meet the needs of artificial intelligence (AI) kit. And that could be a big problem for enterprises buying in to the latest and greatest IT infrastructure.

DCPI specifically refers to equipment within data centers that is used for power and cooling, as well as the IT infrastructure itself. Think uninterruptible power supplies (UPS), thermal management solutions, cabinet power distribution and busway, rack power distribution, IT racks and containment, and software and services. Schneider Electric, Eaton and Vertiv are the top three players in this space.

Given all the hype around artificial intelligence (AI), it might be easy to point to the technology as a driving force behind 2023’s DCPI growth. But Beran said that’s not what’s happening. Instead, the growth is coming from the realization of price increases vendors have passed along to their customers as well as those aforementioned growing backlogs.

As far as the DCPI market goes, AI is really more of a late 2024, 2025 story, he added.

2Q23 DCPI backlog delloro.
This chart is based on publicly available data from Vertiv, Eaton, ABB, Munters and JCI. Schneider Electric was notably omitted from this analysis. (Dell'Oro Group)

Beran noted data centers are nowhere near prepared to meet the wave of AI demand that’s swelled up with today’s facilities. That’s because of both supply chain delays and those pesky backlogs.

During the pandemic “lead times got push out to up to a year for certain products and so we’re really just continuing to work through backlog that is related to growth from pandemic induced digitalization and we gotta get through all of that or most of that before we start to plan and build for AI workloads,” Beran said.

He warned folks that get too far ahead of the curve when it comes to AI-focused IT infrastructure could end up temporarily stranded.

Beran said already he’s heard of a specific case in which a lack of adequate power and thermal management capabilities has led to tens of millions of dollars worth of AI IT infrastructure sitting around idle.

“You need higher rack densities, with higher power distribution to the rack, high power distribution within the rack and then higher thermal management capabilities to capture heat from the rack. And there’s significant work to do there,” Beran said.

But there are some signs things are headed in the right direction. For instance, he flagged thermal management as the fastest growing segment within DCPI. Cabinet power distribution and busway is right on its heels — see the chart below.

2Q23 DCPI segments Chart

So, what’s on the horizon? Well, Beran said 2024 is set to be a something of a transition year and tipped DCPI growth to slow due to tough year on year comparisons.

“Price will no longer be a driver in 2024 and we’re not really going to see demand from AI workloads start to materialize in data center physical infrastructure really at a meaningful scale until 2025,” he concluded. “So, growth is likely to dip and I’m forecasting it to dip to mid-single digits in 2024, but I kind of consider that time period the calm before the storm, before these AI workloads really dominate headlines in 2025 and beyond.”

Want to discuss AI workloads, automation and data center physical infrastructure challenges with us? Meet us in Sonoma, Calif., from Dec. 6-7 for our Cloud Executive Summit. You won't be sorry.