Gartner Inc.’s latest worldwide information technology spending forecast predicts a 24.1% jump in spending on data center systems, more than double the 10% growth expected just three months ago and six times the actual growth of 4% last year — thanks to generative artificial intelligence.

That’s according to John-David Lovelock (pictured), a distinguished vice president analyst at Gartner who is the principal author of the report. He said spending by managed service providers on servers tuned for AI training is driving most of the surge and that the impact is likely to last for years.

A veteran of a decade of forecasting, Lovelock said even he hasn’t seen a change this dramatic since COVID-19. “Servers growing at this rate, I’ve never seen it, and I never would have expected it,” he said.

Supply chains easing

Two factors are behind the sudden jump, he said. Nvidia Corp., which dominates the market for the graphic processing units that power AI servers, has solved most of its supply chain problems. “The world now has the ability to make more AI-optimized servers,” Lovelock said. “But we still see backlogs on orders for these new generative AI-type servers in the six months to one-year range.”

Large hyperscalers and many managed service providers “are buying as many of these things as they can to build out and prepare for the wave of gen AI inferencing that they see coming,” he added.

Lovelock said the AI training rush will shift base server spending up for at least the next several years as large language models grow larger and consume more server resources. The next generation of models could cost as much as $1 billion each to generate.

As the action shifts to inferencing – or putting trained models to work making predictions or decisions based on new data – demand will remain high, Lovelock said. Inferencing is less computer intensive than training, but “while you build a model once, you inference it all year,” he said.

No end in sight

Gartner estimates that large hyperscalers will spend $70 billion on cloud infrastructure this year, growing to $140 billion in 2025 and $210 billion in 2028. “We still don’t have an end state for these tools,” he said. “We’re not anywhere near commodity status for a set of LLMs or even the basic agreed-on functionality that an LLM should have.”

Though enterprise spending isn’t a major factor in server spending growth, corporate data centers aren’t entirely out of the picture. “Cloud providers are in the 40% to 60% growth range for servers while businesses are in the 20% range,” Lovelock said. “Most CIOs are still at the starting line asking what they can do with these tools.”

Other sectors of the IT economy are expected to roughly maintain past growth patterns. Spending on devices is expected to rise 5.4% this year after falling 6.5% last year. Software will maintain healthy 12.6% growth, and the services sector will grow 7.1% off a base that’s larger than the software and data center sectors combined.

Gartner noted that although software companies are rushing to add generative AI capabilities to their products, most aren’t yet in a position to profit from the new features. The research firm said generative AI today more closely resembles a tax than a value-added feature.

“To sell product and maintain market position, software vendors are going to need to have a generative AI feature by the end of 2025,” Lovelock said. “That doesn’t give anybody time to develop an AI model so most are going to go to somebody who already has a model.”

In effect, any additional revenue from adding AI features will simply flow back to the companies that operate the models. Generative AI “may be a loss leader or a breakeven proposition until we know where gen AI adds value,” he said. “Once they learn more, they’ll be able to price according to the value delivered.”

Worldwide IT Spending Forecast (millions of U.S. Dollars) 

Source: Gartner

Photo: SiliconANGLE