How Realistic Is the Datacenter Buildout Forecast?

The generative AI boom can expand only as quickly as datacenters are constructed, connected to power, and equipped to support new systems. Political, social, or economic pressure cannot eliminate the time required to install power-generation and distribution equipment at datacenters and utilities.

The industry did not anticipate the speed of the current AI expansion. Supply chains covering chip-manufacturing equipment, semiconductors, electrical infrastructure, and datacenter power-distribution systems were not prepared to scale at the pace now being demanded.

People are still required to build datacenters, homes, offices, and public facilities. High mortgage interest rates may be helping datacenter developers attract construction workers who might otherwise be working on residential or commercial projects.

Jason Furman, an economics professor at Harvard University, published a chart on X showing growth rates for US construction projects through August 2026:

This data provides another indication of how datacenter and utility construction are supporting the US economy. The figures come from the US Census Bureau and the US Bureau of Labor Statistics. They include construction materials and project labor, as well as power-distribution and backup-energy equipment installed in datacenters. They do not include compute, storage, or networking equipment inside those facilities.

Datacenter construction, along with power-utility construction to a lesser extent, is offsetting declines in US commercial and residential construction.

US Datacenter Construction Spending

The growth rates alone do not show the overall spending levels, so a separate monthly series from the US Census Bureau covers the period from January 2014 through August 2026. The Our World in Data project at Oxford University assembled the series.

The figures use inflation-adjusted spending measured in 2021 US dollars.

The impact of the generative AI boom becomes especially visible in November 2022, when US datacenter construction spending began a sustained upward acceleration.

The Census Bureau data excludes datacenter spending by federal, state, and local governments. That spending is small compared with the investments made by hyperscalers, cloud providers, and AI model developers, so including it would not materially change the overall trend.

A Forecast Through 2030

Analysts at Jeffries developed a detailed datacenter-spending forecast with assistance from Aterio, a market research firm that specializes in datacenter forecasts among other sectors. Aterio supplied historical context, while Jonathan Petersen, managing director at Jeffries, prepared a base forecast along with bullish and bearish scenarios covering 2026 through 2030.

The forecast focuses on installed IT capacity in gigawatts across North America. It does not estimate the cost of constructing datacenters, installing the equipment inside them, or operating their power and cooling systems. It also adjusts project announcements according to the likelihood that projects will be completed and the extent to which their stated plans are likely to materialize.

In a report released on October 9, Petersen wrote:

“The AI industry has a datacenter problem. Demand for inference and frontier model training is growing faster than developers can bring capacity online. Our dataset tracks ~65 IT GW of DCs under construction (~140 percent of installed capacity), and another ~361 GW of future announced projects (~760 percent). However, it will take time to bring that capacity online, and many projects will be delayed or canceled. We expect 12.9 GWs of new capacity in 2026 and 18 GWs in 2027.”

The forecast takes a cautious view of the buildout because of constraints affecting equipment, labor, power, and project approvals.

“The ability to bring new datacenter capacity online is often reduced to a discussion measuring new grid interconnects or behind-the-meter power generation as primary gating factors,” Petersen wrote. “Or tracking hyperscaler capex and assuming that spending more capex will create new data centers. Or, measuring chip sales projections from the semiconductor industry. But any of those discussions in isolation misses the bigger picture: it takes all of those inputs, plus available labor, manufacturing of electrical and cooling equipment, and a cooperative political and regulatory environment to move a data center project from the pipeline to online capacity.”

Planned Capacity Versus Expected Capacity

The bearish scenario reflects labor shortages, increased political pressure for datacenters to pay an appropriate share of electricity-generation costs, and growing skepticism about the financial benefits datacenters provide to host communities.

The bullish scenario assumes that datacenter construction supply-chain problems will ease, although not enough to complete every planned project. Even this more optimistic case includes substantial delays and some cancellations.

The difference between announced plans and expected construction is visible among hyperscalers, despite their substantial financial resources and strong balance sheets. Planned and estimated North American hyperscaler capacity is as follows:

• 2026: Hyperscalers planned 6.21 gigawatts, broadly matching the 6.21 gigawatts Jeffries expects to be built.
• 2027: Planned capacity is 8.37 gigawatts, compared with an estimated 7.53 gigawatts.
• 2028: Planned capacity rises to 10.13 gigawatts, while the estimate falls to 6.54 gigawatts.
• 2029: Plans approach 18 gigawatts, but the estimate is only 5.86 gigawatts.
• 2030: Plans decline to 13.82 gigawatts across hyperscaler facilities in the United States and Canada, while the expected buildout is 4.46 gigawatts.

These gaps are substantial. Jeffries identifies labor as the largest constraint on datacenter spending in 2026. Cooling equipment is expected to be the primary limitation from 2027 through 2029. By 2030, available power is projected to become the main constraint.

Those limitations are present throughout the forecast period, even as their relative importance changes from year to year.