Large-Load Interconnection Agreements and Curtailment Obligations

The interconnection agreement is the contract that converts the studies and the tariff into enforceable obligations, and in the large-load era its center of gravity has shifted toward one idea: conditional, flexible service. For decades the default a customer wanted was firm, network-quality service, the utility's promise to deliver full power on demand at all times, backed by whatever network the grid must build to keep that promise. Firm service is expensive precisely because it obligates the grid to plan for the customer's peak under the worst credible contingency, even if that peak occurs for only a few hours in a year. For a several-hundred-megawatt load, building the network to guarantee that rare peak can be the single largest and slowest item in the whole project.

Curtailable service is the alternative, and it has moved from a niche option to a central bargaining chip. Under a curtailment provision the customer agrees that, during defined conditions, a system emergency, a transmission constraint, a period of peak stress, the utility may order it to reduce or interrupt its draw. In exchange the customer gets a faster connection, a lower upgrade cost, or a better rate, because the grid no longer has to build for the customer's full peak. The logic is clean: if the load will step back during the handful of hours that drive the worst-case planning, the grid can serve it with the network it already has, or with far smaller additions. The customer trades a small amount of availability for a large amount of speed and money.

What makes this workable for data centers specifically is that their flexibility is real, not theoretical. A large AI campus can shift non-urgent training workloads in time, can throttle to a lower power state, and increasingly can lean on substantial on-site batteries to ride through a curtailment event without dropping its critical load. Utilities and grid operators have begun to recognize that a data center is a far more controllable large load than, say, a steel mill mid-heat, and the newest large-load tariffs and agreements are being written to reward that controllability explicitly. The precise terms matter enormously: how many hours per year the load can be curtailed, how much notice it receives, how deep the curtailment can go, and what compensation or rate benefit flows in return. Those numbers determine whether a curtailment clause is a manageable operating constraint or an unacceptable risk to a tenant's service-level commitments.

Alongside curtailment, these agreements carry the provisions that assign the hard risks. There is the cost-responsibility schedule that says exactly which upgrades the customer funds and on what timeline. There is the ramp schedule that governs how quickly the load may grow, protecting the grid from a step change it cannot absorb. There are the milestone and security provisions, the deposits and letters of credit the customer posts so the utility is not exposed if the project stalls after upgrades are ordered. And there are the terms governing what happens if the customer fails to meet its minimum take or walks away, echoing the tariff's exit provisions in enforceable contract language. Each clause is a place where risk is being allocated between the customer, the utility, and, indirectly, every other ratepayer.

When I read one of these agreements, I read it as a risk-allocation map. The curtailment terms tell me how much of the peak-capacity risk the customer has agreed to absorb in exchange for speed. The cost-responsibility schedule tells me how much of the upgrade cost stays with the customer rather than migrating to the public. The security and milestone terms tell me who is exposed if the schedule slips. A well-structured large-load agreement is one where the customer has taken on flexibility and cost responsibility proportionate to the burden it places on the system, because that is not only the fair arrangement, it is the durable one. The agreement that shoves risk onto other ratepayers may close faster, but it is the one most likely to be reopened when the politics catch up, which is the subject I turn to next.