AI Accelerated Servers Fuel Growth in Data Center Spending

Data centers increased capital expenditures (capex) by nearly 50% in the second quarter of 2024, driven by the appetite for AI-ready hardware, according to a new report from Dell’Oro Group.

Insight from the California-based research firm indicates that data center capex grew by 46% in the second quarter of 2024, compared to the same period a year earlier.

The spending spree is expected to continue for the remainder of the year, albeit at a slightly slower pace. Dell’Oro forecasts a robust 35% growth in worldwide data center capex overall for 2024.

The organization said the growth is largely fueled by the increasing demand for GPU-accelerated servers designed for AI applications.

“Accelerated servers are servers with GPUs and other custom accelerators, such as the Google TPU,” Baron Fung, senior research director at Dell’Oro Group, told Data Center Knowledge.

The Dell’Oro numbers are the latest in a series of double-digit data center expenditure growth predictions. Gartner forecasts that overall spending on data center systems will grow by 24% in 2024. Like Dell’Oro, growth is in part attributed to the need to power generative AI systems and services.

Omdia forecast at the end of 2023 that the global server market would be worth $196 billion by 2027, while, a report from Fundamental Business Insights & Consulting released earlier this month estimated the market for AI servers to be worth $183 billion by 2032.

Related:How Utilities, Hyperscalers Are Working to Tackle ‘Extreme’ Data Center Power Demands

AI Acceleration: The Engine of Growth

While growth in accelerated servers is no surprise, the market is growing faster than expected, according to Dell’Oro researchers.

“Accelerated server revenue did exceed our prior forecast or expectations,” Fung said. “Momentum for accelerated server adoption is strong, especially for the hyperscale market.”

GPU-accelerated servers are outpacing normal server growth in the data center in terms of both revenue and shipments.

Fung said that over five years from 2023 to 2028, the compound annual growth rate for general-purpose servers is forecast to grow by only 3%, versus 31% for accelerated servers.

AI workloads are forecast to grow at a pace of 25-35% through 2027, according to Bain & Company’s latest Global Technology Report.

“Generative AI is the prime mover of the current wave of change, but it is complicated by post-globalization shifts and the need to adapt business processes to deliver value,” said David Crawford, chairman of Bain’s global technology practice.

Accelerated Computing’s Impact on Data Center Capex

The impact of accelerated computing on data center capex is being felt around the world, but it won’t last forever, analysts said.

Related:How LLMs on the Edge Could Help Solve the AI Data Center Problem

"It’s both increasing as a share of the total and probably also driving the total market size," Omdia analyst Alexander Harrowell, told Data Center Knowledge.

“That said, it’s increasing substantially faster than the total market, so at some point, the classic S-curve dynamic will kick in, probably in the next two to three years, just because the pool of potential adopters will be exhausted," Harrowell said. “That will leave us with a world of much more heterogeneous computing.”

In terms of market dynamics, Harrowell said Nvidia’s data center business is likely to be around 10 times the size of Intel’s data center and AI division this year. The market is further buoyed by hyperscalers’ custom ASIC chips, along with offerings from chipmakers including AMD and Huawei.

AI Server Market Leaders and Revenue Distribution

In the competitive landscape of data center server providers, Dell emerged as the leader by revenue for Q2 2024, followed by HPE and Supermicro, according to Dell’Oro analysis.

Notably, accelerated servers accounted for an estimated 41% of OEM server revenues during the quarter, highlighting their significant impact on the market.

Related:Mapping the Best Data Center Locations in 2024

The increased demand for accelerator services will also likely drive more spending in related areas as well, such as networking.

Fung noted that networking is also benefiting from an increase in accelerated computing adoption.

“Accelerated servers usually require more network bandwidth than general-purpose servers,” he said.

“If the accelerated servers are interconnected in a large cluster that is used for the training of large language models or generative AI models, then a dedicated network, called a back-end network, would be needed.

“This back-end network is even faster than the front-end network, and there have been a lot of recent innovations behind this new market.”