The Data Center Construction Surge May Be Nearing A Plateau

Tech giants keep ramping up their artificial intelligence spending, but the blistering growth of the AI data center construction boom may not last much longer. 

The pace of data center construction has accelerated to unprecedented levels as Big Tech companies increase investments in the infrastructure to support their AI ambitions. The combined capital expenditures of the world’s largest tech firms jumped 73% year-over-year last quarter, and hyperscalers Amazon, Google, and Microsoft project it will grow even faster in the months ahead.

Overall data center construction spending doubled between 2024 and 2025 and is expected to nearly double again in 2026. But the data center sector’s stunning skyward growth trajectory may have an imminent expiration date. 

Despite the barrage of headlines about massive data center megacampuses and escalating Wall Street concerns about the trillions of dollars hyperscalers are slated to spend on AI infrastructure, there is growing sentiment within the data center industry that its growth curve is set to plateau as soon as next year. Projections of a sudden moderation in data center construction spending over the next 18 months have headlined recent studies and gained further credence through comments from leaders of major tech firms. 

U.S. spending on data centers is projected to surge to $86B by 2026, according to an October report from Moca Systems. This represents a staggering 782% increase from the $11B spent in 2022, which was only $2B more than in 2021.

But this period of dramatic growth isn't expected to last, as Moca’s data projects 2026 will be the market's peak.

Following that peak, Moca foresees a gradual decline in data center spending. While levels still will be well above those before the AI boom — the $63B in estimated 2030 spending is five times more than in 2022 — the report indicates the era of rapid acceleration in data center construction spending soon will be over.

These projections aren't an indication of tech giants slowing overall spending on AI infrastructure, but rather of the path that investment is taking: a front-loaded surge on building new data centers that will taper off and shift toward spending on chips and other computing hardware. 

Most of the oft-cited data center spending projections from firms like Goldman Sachs, JLL, McKinsey and Boston Consulting Group — forecasts that range from 10% to 30% annual growth through 2030 — focus on the sector’s compound annual growth rate, painting a linear growth picture.

But the far bumpier reality of the data center construction sector’s growth trajectory could have repercussions for players across the market.

Developers, contractors and firms in their supply chains, from equipment vendors to design firms, must align their own capital deployment strategies with the demand landscape that lies ahead, Moca Systems principal economist Brandon Michalski said. 

“Planned spending signals point to a market trajectory where short-term acceleration outpaces smoothed long-term models, making reliance on CAGR alone a strategic risk for firms planning capital, staffing, and infrastructure commitments,” Michalski wrote in the October report.  

Moca isn’t the only firm predicting this imminent market plateau. Market analytics firm GlobalData projects that data center construction will follow a nearly identical growth trajectory. This thesis is also supported by construction spending data from Dodge Data and Analytics, which tracks construction starts and permitting.

Beyond these economic analyses, leaders of the tech firms driving this spending have also fueled the notion that the peak of the data center building boom may be coming soon. Meta CEO Mark Zuckerberg last month described the firm’s AI infrastructure strategy as a bid to "aggressively front-load building capacity.” 

At Microsoft, Chief Financial Officer Amy Hood hinted at a coming plateau in data center construction as the firm looks to shift its AI capex away from new data centers and toward buying the GPUs and other processing equipment housed inside them. 

“We're pivoting toward, increasingly ... short-lived assets, both GPUs and CPUs,” Hood told investors last month. “We've spent the past few years not actually being short GPUs and CPUs per se — we were short the space or the power to put them in, so we spent a lot of time building out that infrastructure.”

Hood’s comments highlight a fundamental shift underway in Big Tech’s AI infrastructure push that is the primary force poised to temper data center growth, according to Moca’s Michalski. 

Since the release of ChatGPT kicked off an AI gold rush in 2022, a scarcity of land with access to the power needed to build gigawatt-scale data center campuses has been the primary constraint limiting Big Tech’s AI ambitions. The ongoing tsunami of development stems from what Michalski calls the “training surge,” as hyperscalers front-load land and power acquisition and campus construction in new locations to develop and train AI models.

Tech giants still don’t have the data center capacity to meet today's demand. But while they’re still playing catchup, the massive upfront capital expenditures on land, power and infrastructure being made today are meant to create scalable campuses that will lower the capital required to keep up with growing AI demand in the years ahead. 

Today, hyperscalers are spending billions on acquiring land with significant predevelopment risk and pursuing costly energy infrastructure projects — even building their own power plants — by themselves or in partnerships with energy firms and utilities. Eventually, they will be able to expand far more efficiently within these existing campuses, building on land that has already cleared most predevelopment hurdles and utilizing preestablished utility power agreements or their own self-developed energy infrastructure.

“I think a lot of people are saying the same thing: You do need a certain number of physical structures to handle the hardware, but going out into the future, how many 1,500-acre sites are really economically feasible?” Michalski said. 

Hyperscalers are eager to shift their spending from the physical infrastructure layer to GPUs and other high-performance computing hardware, which are what will actually drive the returns on investment that Wall Street is so eager to see, industry insiders say. 

Some analysts, such as Bain & Co.'s Padraic Brick, say they initially expected a slowdown in data center spending to begin earlier this year. But even though construction spending continues to shoot skyward, he wrote in an October research note that there is evidence to suggest hyperscalers are no longer scrambling to find powered land at any cost anywhere they can, but are being more calculating as to how they can get the most bang for their buck. 

“[W]e are seeing more deliberate investments by hyperscalers as they scale capacity focusing more on capital efficiency and getting more selective on locations for new deployments, particularly for AI,” Brick said.