When a data center developer announces a new campus, the headlines focus on the investment figure and the job count. The less visible story — the one that determines whether the facility actually gets built on schedule — is interconnection: the process of physically and contractually connecting a massive new electrical load to the grid. For large data centers, interconnection has become one of the industry’s binding constraints, with queues stretching years and upgrade costs running into the hundreds of millions.
What interconnection actually means
Every electricity customer connects to the grid, from a single home to a steel mill. For ordinary customers, the process is routine: the utility checks local capacity, installs a meter, and turns on service. For a customer requesting 100 megawatts or more, nothing is routine. The utility must determine whether the existing wires, substations, and upstream transmission can carry the new load safely — and if not, what must be built to make it possible.
The process has several stages. First comes the application: the developer submits a formal request specifying the location, the requested capacity (in megawatts), the desired energization date, and technical details about the facility’s electrical characteristics. Then come the studies — engineering analyses that model how the new load affects the grid. Finally comes the agreement: a contract specifying what gets built, who pays for it, and when the load can energize.
The studies are the heart of the process. They typically proceed in phases of increasing detail and cost. An initial feasibility or screening study gives a rough answer: can the grid serve this load, and roughly what upgrades would be needed? A system impact study models the load’s effect on power flows, voltage, and system stability under normal and contingency conditions — what happens if a transmission line or generator trips offline while the data center is drawing full power? A facilities study then produces the detailed engineering design and cost estimate for the required upgrades.
Each stage takes months. A full interconnection study sequence for a large load can take 18 months to three years, depending on the utility, the region, and the complexity of the upgrades. And the studies sometimes deliver unwelcome news: the upgrades needed to serve the load may cost far more, or take far longer, than the developer expected.
Why large loads are hard to serve
The difficulty scales nonlinearly with size. A 5-megawatt load can usually be served from the existing distribution system with modest upgrades — a new feeder, a transformer replacement. A 50-megawatt load typically requires a dedicated substation. A 200-megawatt load may require new high-voltage transmission lines, expanded substation capacity, and upgrades deep into the transmission network.
The reason is physics. Power flows from generators to loads along the paths of least resistance, and the transmission network was designed around historical flow patterns. Injecting a massive new load at one point changes flows across the whole region. A line that was comfortably loaded may become overloaded. Voltage at nearby buses may sag. Under contingency conditions — the grid’s standard design test, which asks what happens when any single element fails — the new load may push the system past its limits.
Fixing these problems means building: new transmission lines, new transformers, new substations, sometimes new generation or reactive power support. Transmission lines are the long pole in the tent. A new high-voltage line can take five to ten years from planning to energization, with permitting, land acquisition, and construction each contributing years. The data center wants power in two to three. That gap — between the customer’s timeline and the infrastructure timeline — is the central tension of large-load interconnection.
The queue problem
Utilities and grid operators process interconnection requests in queues, and those queues have exploded. In several regions, the total large-load interconnection requests in the queue now exceed the region’s entire existing peak demand. Not all of these requests represent real projects — developers routinely submit multiple requests for alternative sites, and some requests are purely speculative, filed to hold a place in line.
This creates a cascade of problems. Studies take longer when the queue is long. Each new request must be studied not only against the existing grid but against the other queued loads ahead of it, because those loads, if built, change the grid conditions. When a speculative project drops out of the queue, the studies for projects behind it may need to be redone. The result is a slow, expensive, churn-prone process that frustrates serious developers and utilities alike.
Utilities are responding with reforms. Common measures include higher application fees and study deposits to discourage speculation, milestone requirements that force developers to demonstrate site control and financing before advancing, and “first-ready, first-served” rules that prioritize projects with real timelines over placeholders. Some utilities now require large-load customers to post significant financial security or sign take-or-pay commitments before the utility begins building upgrades on their behalf.
Who pays for upgrades
Cost allocation — who pays for the infrastructure needed to serve the new load — is the most contentious part of interconnection. The traditional principle is “cost causer pays”: the customer whose load triggers an upgrade pays for that upgrade. In practice, the boundary is blurry.
Direct connection facilities — the substation, the short line extension to the site, the metering — are clearly the customer’s responsibility. But deeper network upgrades, like a new transmission line that also benefits other customers or relieves existing congestion, raise harder questions. If the upgrade would have been needed eventually anyway, should the data center pay the full cost? If the upgrade enables future growth beyond the data center, how should costs be shared?
Regulators are grappling with these questions in rate cases across the country. The emerging toolkit includes: requiring large-load customers to pay for all identified upgrades; minimum bills and long contract terms (10 to 20 years) that ensure the customer can’t leave ratepayers holding the cost of stranded infrastructure; and new large-load tariff classes with demand charges designed to recover the fixed costs of serving them. The details vary by jurisdiction, but the direction is consistent: protect existing customers from bearing the costs of speculative or footloose large loads.
The timeline in practice
For illustration, here is how the timeline of a hypothetical 200-megawatt data center interconnection might unfold. Months 0–3: site selection and initial utility engagement, including a pre-application meeting to discuss feasibility. Months 3–18: formal studies — feasibility, system impact, facilities — with the developer posting deposits at each stage. Months 12–24 (overlapping): negotiation of the interconnection agreement, including cost allocation and construction milestones.
If the studies identify only local upgrades — a new substation and short line extension — construction might take 18 to 30 months, putting energization roughly three to four years after the initial request. If the studies identify major transmission upgrades, add several more years. This is why experienced developers begin utility engagement before they even close on land: the interconnection timeline, not construction of the building, usually determines when the facility can operate.
Some developers try to shortcut the process with phased energization: bringing a portion of the load online early on existing capacity while upgrades are built for the remainder. Others explore temporary solutions like on-site generation to bridge the gap. But there is no way around the fundamental physics: the grid can only serve what its wires can carry.
Flexibility as an interconnection strategy
A newer idea is gaining traction: what if the data center doesn’t need its full requested capacity all the time? Traditional interconnection assumes the customer might draw its maximum contracted demand at any moment, and the grid must be built for that worst case. But many data center workloads are at least partially flexible — batch processing, model training checkpoints, and non-urgent computation can be paused or shifted.
Flexible interconnection — sometimes called curtailable or non-firm load — would let a data center connect faster and cheaper in exchange for agreeing to reduce consumption during grid stress events. The utility avoids building peaking capacity for a load that rarely coincides with the system peak; the customer gets energized years earlier. For illustration, a 200-megawatt facility that agrees to curtail to 120 megawatts during the top 100 hours of the year might interconnect on existing infrastructure that could never have served its full 200-megawatt peak.
The concept is proven in other industries — interruptible industrial tariffs have existed for decades — but applying it to data centers raises new questions. How much of an AI training workload is truly interruptible without destroying its economics? Who decides when a curtailment event is called, and with how much notice? How are the customer’s lost compute time and the utility’s avoided infrastructure costs valued against each other? Pilot programs are beginning to work through these questions, and flexible interconnection may become a standard option alongside traditional firm service.
Reform efforts and what comes next
The interconnection bottleneck is now widely recognized as a drag on economic development, and reform efforts are underway on several fronts. Grid operators are streamlining study processes — moving from serial studies to cluster studies that evaluate groups of requests together, which is faster and more accurate. Utilities are publishing hosting-capacity maps that show developers where the grid can accommodate large loads without major upgrades, steering development toward ready sites.
Longer-term, some planners argue for a more proactive approach: building transmission capacity in anticipation of large-load growth in designated zones, rather than waiting for individual requests. This “build it and they will come” strategy is controversial — it risks stranding investment if the load doesn’t materialize — but in regions with overwhelming demand signals, the risk calculus is shifting.
Standardization would help too. Today every utility runs its own interconnection process with its own forms, study methods, and timelines, forcing developers active in multiple territories to navigate a patchwork. Industry groups are working toward common technical standards for large-load interconnection — standard study scopes, standard data requirements, standard modeling assumptions — which would cut administrative friction even where the underlying engineering remains complex. The grid will never interconnect a 200-megawatt load as casually as a new subdivision, but it can stop making developers learn a new process in every service territory.
The interconnection process will never be fast or simple for loads of this size; the engineering is genuinely complex and the stakes — grid reliability for everyone — are genuinely high. But the current system was designed for an era of slow, predictable load growth. Adapting it to the era of the 200-megawatt customer is one of the power sector’s defining challenges.