The AI Infrastructure Industry Has Entered Its Coordination Crisis
A data center can now be built in 18 months. The transformer that powers it can take as long as four years to arrive. And the transformer is only one of the clocks that has to line up.
Power, cooling, electrical equipment, permitting, construction and utility upgrades all operate on different timelines. Those timelines were never designed around one another. Any one component can be available, and the project can still fail to come together.
The next bottleneck in AI infrastructure may not be any single resource. It may be coordination.
Suhail Tayyeb. PHOTO: Courtesy NYU
The building still matters, but some of the hardest problems now sit between the systems required to make it work. Power has to arrive when the customer needs it. Cooling has to match rapidly changing rack densities. Electrical equipment may have to be procured years before deployment. Contractors, utilities and equipment manufacturers are often working on very different clocks.
PJM offers a useful example of how quickly the bottleneck can move. Its reformed interconnection process can now process new generation projects in one to two years. Yet roughly 57 gigawatts of projects have completed PJM’s study process and have either signed or been offered interconnection agreements. Many are still being slowed or stopped by factors outside PJM’s process.
PJM’s analysis of its recent capacity shortfall makes the mismatch even clearer. Gigawatt-scale load can materialize in 12 to 18 months. The generation required to support it takes considerably longer. Permitting accounted for 29 percent of the generation-project milestone changes PJM tracked since 2023. Generator step-up transformers and gas turbines can take three to four years to procure.
Consider power. A developer can control land beside transmission infrastructure and still have no project if the utility cannot deliver capacity on time. Behind-the-meter generation provides another path. But that brings its own requirements for generation, fuel supply, interconnection and operations. Those pieces now have to be coordinated with the data center itself.
Solve the power problem, and another seam can appear. AI racks can require very different cooling systems from traditional computing. Liquid cooling adds pumps, heat exchangers, piping, controls and water chemistry. Each component may work on its own. The challenge is making the entire thermal system work with the compute.
Construction faces the same problem. A hyperscale campus can involve thousands of workers and multiple contractors. It can require enormous quantities of electrical and mechanical equipment. Meanwhile, the technology going into the facility may evolve faster than the building itself. Traditional construction management divides complexity into packages and coordinates them through a schedule. AI infrastructure increasingly requires something closer to systems orchestration.
A transformer manufacturer can deliver a transformer. A cooling company can deliver a cooling system. A utility can deliver electricity. None of them delivers a functioning AI infrastructure campus. The value increasingly lies in making those systems work together at the right capacity and at the right time.
That is also why AI infrastructure is becoming difficult to describe as simply another real estate asset class. Operators increasingly confront utility coordination, thermal management, controls integration, commissioning and generation strategy. These projects combine elements of energy infrastructure, industrial systems engineering, software operations and network architecture. Real estate remains essential, but the building is only one part of the system.
The coordination problem also extends beyond the property line. A utility may plan transmission on one timeline while a hyperscaler plans compute deployment on another. Permitting agencies, water authorities and economic development organizations may operate on still others. A viable project has to navigate all of them.
This helps explain why projects that look viable on paper can become difficult to execute. A site is not a project simply because it has land. Power is not deliverable capacity simply because it appears on a utility plan. Cooling equipment is not a thermal solution until it works with the compute architecture. The project works only when enough of these pieces become available at the same time.
That changes where competitive advantage may emerge. The first phase of the AI infrastructure boom rewarded companies that controlled scarce components: land, power, graphics processing units and capital. Those advantages remain important. But another capability is becoming scarce as projects grow larger and more complex. It is the ability to orchestrate them.
That may favor firms that can coordinate utilities, energy developers, equipment manufacturers, contractors and technology companies around one executable schedule. The next generation of AI infrastructure companies may resemble industrial integrators as much as real estate developers. Their advantage will come from making the entire project work as one.
The AI infrastructure industry has spent the last several years racing to secure land, power, equipment and capital. Securing those components individually is no longer enough. The next phase will be defined by whether companies can turn them into functioning infrastructure on the timeline customers actually need.
Suhail Tayyeb is a clinical assistant professor at New York University’s Schack Institute of Real Estate and director of the Center for the Sustainable Built Environment.
Source: commercialobserver.com — article syndicated from the publisher’s feed; all rights remain with the original publisher.