Hungry for resources, AI redefines the data center calculus

The AI revolution is driving demand for massive computing power and creating a data center shortage, with data center operators planning to build more facilities. But it’s time for data centers and other organizations with large compute needs to consider hardware replacement as another option, some experts say.

Data centers are running at near capacity in many areas of the world, according to the Global Data Center Trends 2024 report from CBRE, a real estate services firm. Singapore’s data centers have a 1% vacancy rate, and data center availability in the huge northern Virgina data center hub stands at 0.9%, despite an 18% increase in capacity between early 2023 and early 2024, according to the report.

The 2023 CBRE report found that 83% of the data center capacity under construction at the time was presold.

While there’s more than 10% of capacity available in Europe, Latin America, and the greater Asia Pacific region in 2024, the growth of AI, cloud computing providers, and other power-hungry applications requires a new approach to data center operations, the report says. “High-performance computing will require rapid innovation in data center design and technology to manage rising power density needs,” it adds.

A current data center building boom isn’t likely to stop soon, with credit rating firm Moody’s, in a July 15 report, projecting that global capacity will double in the next five years. But with finding space for new data centers becoming increasingly difficult, some experts say it’s a good time for data center operators, cloud computing providers, and companies running AI and other major workloads in house to think about replacing old hardware instead.

Efficiency is money

Modern CPUs not only offer more computing power, but they also are more power efficient than older models and generally take up much less data center space, some advocates say. Using new CPUs, data centers can consolidate servers running tens of thousands of cores into less than 50 cores, says Robert Hormuth, corporate vice president of architecture and strategy in the Data Center Solutions Group at AMD.

An estimated 100 million 5-year-old servers are still in operation, partly because the COVID-19 pandemic stalled some planned upgrades, Hormuth says. About 21 million new servers could replace those old machines, leaving data centers and in-house server rooms more space to add computing power.

With power efficiency gains and other savings, a large-scale hardware replacement could give data center operators and other hardware-dependent organizations a return on investment in as little as two months, Hormuth claims. Meanwhile, building a new data center can cost hundreds of millions of dollars.

Power efficiency gains of new hardware can also give data centers and other organizations a power surplus to run AI workloads, Hormuth argues.

“That’s the race that seems to be going on in enterprises: ‘How do I go make room and power to do AI?’” he adds. “That pressure is just really driving the enterprise customers, whether it be in a co-lo or create their own, to get those capabilities.”

Big upgrades ahead

While AMD has a dog in this hardware-replacement fight, several other IT experts also say it’s a good time to replace servers and other hardware. Many data center operators appear to be thinking the same way, with Gartner projecting a 24.1% data center spending increase, covering servers, external storage, and network equipment, in 2024. Data center spending, which doesn’t include new buildings in Gartner’s calculations, saw just 4% growth in 2023, Gartner says.

Jim Warman, vice president of infrastructure architects and engineers at Myriad360, a data center and cybersecurity consulting firm, sees the same trend. Many hardware users are prioritizing replacement.

“Businesses are recognizing the advantages of modernizing their infrastructure to support new applications and services, reduce costs, and maintain competitiveness,” he adds. “The emphasis is on utilizing the latest technology to drive business growth and operational efficiency.”

With data centers near capacity in the US, there’s a critical need for organizations to consider hardware upgrades, he adds. The shortage is exacerbated because AI and machine learning workloads will require modern hardware.

“Modern hardware provides enhanced performance, reliability, and security features, crucial for maintaining a competitive edge and ensuring data integrity,” Warman says. “High-performance hardware can support more workloads in less space, addressing the capacity constraints faced by many data centers.”

The demands of AI make for a compelling reason to consider hardware upgrades, adds Rob Clark, president and CTO at AI tool provider Seekr. Organizations considering new hardware should pull the trigger based on factors beyond space considerations, such as price and performance, new features, and the age of existing hardware, he says.

Older GPUs are a prime target for replacement in the AI era, as memory per card and performance per chip increases, Clark adds. “It is more efficient to have fewer, larger cards processing AI workloads,” he says.

While AI is driving the demand for data center expansion and hardware upgrades, it can also be part of the solution, says Timothy Bates, a professor in the University of Michigan College of Innovation and Technology. Data center operators can use AI to monitor efficiency, he says.

“By using AI tools to predict and manage hardware degradation — such as PCIe cards, SSDs, and memory components — businesses can replace individual components rather than entire systems, optimizing costs and extending the lifespan of existing infrastructure,” Bates says. “This approach, combined with strategic hardware upgrades, can maximize the efficiency and performance of data centers.”