The aging US power grid is the hidden bottleneck holding back the AI boom

The wires, transformers, and substations carrying electricity across the United States were mostly built for a grid that grew slowly and predictably, adding a new power plant or transmission line every few years to keep pace with modest increases in household and factory demand. That grid is now being asked to absorb data center campuses that each want hundreds of megawatts delivered on a two- or three-year construction timeline, and the mismatch between how fast AI developers want power and how fast the grid can physically deliver it has become the real ceiling on how quickly new computing capacity can come online.

600,000 Miles of Wire Built for a Slower-Growing Grid

The scale of the existing system helps explain why upgrading it is so slow. The Department of Energy’s Office of Electricity describes the U.S. grid as more than 9,200 electric generating units connected by over 600,000 miles of transmission lines, delivering more than 1 million megawatts of generating capacity through a system it calls, in its own words, an “engineering marvel.” But the same agency is blunt about the system’s age, noting that electric infrastructure is aging and is being pushed to do more than it was originally designed to do. Lines and transformers built decades ago for steady, predictable load growth are now expected to absorb sudden, concentrated demand spikes from single industrial customers the size of small cities.

A Queue Holding Nearly Double the Grid’s Existing Capacity

Layered on top of an aging physical network is a planning backlog that has grown for years largely independent of AI. New power plants and battery storage projects seeking to connect to the grid must go through a generator interconnection process, and according to research from RMI, more than 2.2 terawatts of generation and storage capacity is currently sitting in interconnection queues nationwide, nearly double the total generating capacity already installed on the grid today. That is, in theory, more than enough new supply to meet the load growth data centers are expected to create. In practice, very little of it is moving.

The bottleneck is time, not ambition. RMI’s analysis found that the average span between a developer’s initial interconnection request and commercial operation had stretched to nearly five years by 2024, up from under two years in 2008, as grid operators work through legacy, time-intensive study processes designed for a much smaller queue. Of the projects that formally requested interconnection between 2000 and 2019, only 19 percent had actually reached commercial operation by the end of 2024. Costs have grown alongside the delays, since developers are often required to pay for network upgrades that can exceed 10 percent of a project’s entire construction budget, a threshold past which many projects simply withdraw rather than proceed.

Data Centers Need a Different Kind of Connection

Connecting a data center is technically a different process than connecting a power plant, called load interconnection rather than generator interconnection, and it has historically had no consistent federal standard at all. That gap is precisely why the Department of Energy directed the Federal Energy Regulatory Commission to open a rulemaking specifically aimed at speeding up how large loads like data centers get connected to the grid, a notable federal intervention in a process that utilities and regional grid operators had previously managed on their own with wide variation from state to state.

FERC’s First Reforms, and Their Limits

Reformers argue the two problems have to be solved together, not separately. Speeding up how quickly a data center can plug into existing wires does nothing to fix the underlying shortage of new generation and transmission capacity needed to actually power that connection once it is made, and RMI’s analysis notes that accelerating load interconnection alone will only increase the urgency of bringing new generation online. Regulators have started responding: the Federal Energy Regulatory Commission’s Order 2023, issued in 2023, required grid operators nationwide to shift toward batch “cluster” studies instead of processing projects one at a time, to penalize speculative requests that clog the queue without ever being built, and to set firm deadlines with financial penalties for missing them. Those reforms raised the floor on how interconnection studies are run everywhere, but even their authors describe them as a first step rather than a fix, leaving faster software-driven study tools, grid-enhancing technologies on existing lines, and co-located generation-and-load projects as the next wave of changes utilities are being pushed to adopt before the AI-driven demand curve outruns the physical grid’s ability to keep up.

Software, Sensors, and Recycled Interconnection Rights

Some of the proposed fixes already have real-world results behind them. Software from Pearl Street Technologies deployed inside the Midcontinent Independent System Operator has cut a full interconnection study phase from months down to days while maintaining accuracy, and a separate look at grid-enhancing technologies in the PJM Interconnection found that installing devices like dynamic line-rating sensors and power-flow controllers on wires that already exist, rather than building brand-new transmission from scratch, could unlock about 6.6 gigawatts of additional generation capacity for roughly $100 million in installation costs against an estimated $1 billion a year in production-cost savings. Other proposed shortcuts include surplus interconnection service, which lets a new generator plug in using interconnection rights an existing plant at the same site is not fully using, cutting a review that can otherwise take years down to as little as six months, and generator replacement rules that let a retiring plant’s grid connection pass directly to a newer unit built on the same site without triggering a fresh multi-year study.

None of those reforms change the underlying math overnight. Grid operators are still working through a queue built up over more than a decade, and the transmission lines and transformers that make up the bulk of the physical grid take years to permit and build no matter how quickly the paperwork moves. That combination is why utility planners increasingly describe electricity, rather than land, chips, or capital, as the true limiting factor on how quickly the current wave of AI data centers can actually be switched on.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

More from Morning Overview