Aerial view of a data center being constructed.

getty

In the last year and a half there has been a huge build out in data centers to support the explosion in AI training and inference to support the widespread use of AI for industrial and consumer applications. This demand has caused the prices of critical components for computing memory and storage to surge and major memory and storage companies report that they have long term contracts with hyperscale data centers extending into 2028.

However, besides the shortage of components and rising prices there are limitations on the infrastructure to support these data center buildouts as well as growing resistance in many rural and suburban communities where data centers are slated to be built. These issues could slow the pace of AI data center construction and consequently demand for the components that go into these data centers.

Two important issues are availability of power and water to support running and cooling these data centers. The current plan for data center projects in the US would require 780GW of additional power by 2030. This is more than the country’s current peak load of 759GW. That level of expansion would be difficult to achieve.

Likewise, water consumption to cool these data centers is reported to grow significantly as well. U.S. data center direct water consumption is on track to double or quadruple 2023 levels annually. Additionally, about two-thirionds of US data centers built since 2022 are in high water-stress areas. Building more data centers in arid areas would put even greater stress on water resources for other urban uses as well as agriculture.

An additional factor in building new data centers is a growing concern by nearby residents about having data centers as a neighbor. About 70% of Americans polled oppose having data centers built in their communities, largely due to concerns over surging utility bills, excessive water and energy consumption as well as noise and air pollution.

However, the actual number of data center projects that may come online in that time may be only a fraction of the projections. The number of data centers that companies say they need on paper doesn’t necessarily account for the power generation, grid infrastructure, and capital required to build them. Brett Forster, VP of Renewables at McCarthy Building Companies, speaking at Latitude Media’s Transition-AI 2026 Conference in April said that only one in four data centers projects is considered real and likely to move forward.

In addition to the impact of infrastructure issues, local opposition is also having a toll on data center construction. At least $156B across 48 projects with publicly disclosed values was blocked or stalled amid coordinated local opposition in 2025, and this opposition is hardening into additional legislative hurdles. One example of this growth in opposition is in Virginia, an early leader in data center construction. In Virginia, opposition has grown to end the state’s tax break for data center construction, which was worth at least $1.6B in 2025.

If planned data center build-outs do not take place, demand for critical AI components such as GPUs, memory and digital storage could decline. At the same time many suppliers of memory and solid-state storage started production expansion projects to meet prior expected demand that will start to come on line by 2027.

The result of reduced demand and increased supply would lead to decreased prices and excess supply, perhaps leading to another bust time in boom-bust cycle that is typical for this industry.

Infrastructure and political issues slowing AI data center growth and expanded production of solid-state memory and storage could result in a glut by 2027, reversing recent trends.