AI data center boom hits power limits

The buildout of artificial intelligence data centers in the United States is increasingly limited by power infrastructure rather than available capital. In the largest U.S. markets, grid interconnection queues now stretch for years, delaying the connection of new facilities to the electrical system. Substations and broader utility capacity additions also take years to build, further limiting the pace at which new data centers can become operational. These structural delays constrain how quickly AI-related demand can translate into completed, revenue-generating projects.

Even with strong interest in AI computing capacity, the physical realities of the grid slow project timelines. Developers cannot accelerate substation construction or interconnection approvals with software or financing alone. As a result, the timing of power availability has become a central factor in determining when and where new AI data centers can be brought online.

In this constrained environment, energized brownfield sites have become especially important. These are facilities that already exist, are drawing electrical load, and have substation infrastructure in place. By reusing existing grid connections and power systems, such sites can significantly shorten construction timelines for AI data centers. They also reduce interconnection and ramp-up risk, since key elements of the power infrastructure are already established.

The advantage of these brownfield locations lies in not having to start from the ground up. Once a site with ready power is secured, the remaining milestones are largely physical: completing construction, commissioning the facility, and then delivering it to a tenant. This approach can bring forward the recognition of contracted rent compared with greenfield projects that must wait through lengthy grid queues and new substation builds.

A data-center project in Oklahoma illustrates both the potential and the constraints of this model. Construction and commissioning work at the facility are underway, supported by existing power infrastructure at the site. The project is scheduled for delivery to its tenant in the first half of 2027, at which point the first dollar of contracted rent is expected to be recognised. This timeline underscores how energized brownfield sites can move more quickly toward revenue generation than many greenfield alternatives.

However, the broader scalability of this ready-power strategy remains uncertain. While one such data center can clearly produce revenue, there is no evidence yet that developers can systematically find and secure enough energized brownfield sites across multiple markets to support a large, repeatable platform. Without a proven process for replicating this model, AI-driven demand for data centers may be insufficient to offset wider softness in U.S. construction activity. The constraints imposed by grid queues and slow utility buildouts therefore remain central to the outlook for the sector.