A utility that signs a 400-megawatt transmission service agreement without a purpose-built large-load tariff is not just leaving money on the table: it is creating a rate-case precedent that will constrain every future data-center negotiation for years. The tariff is the product. Getting it right is the strategic work.

The AI Data Center Compact

The phrase "AI-data-center compact" describes a relationship that is maturing into one of the defining commercial arrangements of the decade. Hyperscale operators need power, and they need a lot of it, fast, at a location close to fiber networks, cooled by a reliable grid. Utilities need the revenue, the load diversity that large customers provide, and, increasingly, the operational flexibility that data centers can offer as a dispatchable asset. Neither party thrives without the other.

But the compact only works if its terms are built on a durable tariff foundation. When a utility offers large-load service under general tariff provisions not designed for loads of this size, both parties pay a price. The data-center operator receives ambiguous service terms that create business uncertainty. The utility receives revenue that may not fully recover the costs triggered by the load, creating a rate-base shortfall that residential customers ultimately fill. And the commission, when it reviews the arrangement in a rate case, is likely to find problems with whatever ad-hoc framework was used.

A well-designed large-load tariff makes the compact explicit and enforceable. It specifies what the data center receives (transmission service, power quality, reliability standards), what it owes (interconnection costs, demand charges, potentially flexibility obligations), and what happens when the terms are not met. This is not just legal protection: it is the commercial document that gives both parties confidence to invest in a long-term relationship.

Tariff Architecture for Large Loads

A large-load tariff is not a single document section but a layered structure. Understanding the layers helps you see both the design choices and the points of regulatory exposure.

Service Definition

The service definition layer specifies what the large-load customer is buying. For a transmission-level large load, the key elements are: the contracted demand level in megawatts; the service class (firm, non-firm, or a defined interruptible variant); the point of delivery (the specific substation and voltage level); power quality standards, including voltage regulation and harmonic limits; and the commercial operation date and initial service term. Each of these has tariff implications. The contracted demand level determines the capacity charge. The service class determines the reliability obligation and, for interruptible or flexible service, the curtailment terms. The point of delivery defines the study scope. The power quality standards create performance obligations for the utility's transmission system. Getting these definitions right in the tariff prevents the disputes that arise when a data-center operator discovers that the service they are receiving does not match what they thought they had contracted for.

Demand and Energy Charges

The rate design for a large-load customer starts with cost causation: who caused which costs, and how do you allocate them? The primary cost categories are: transmission capacity costs (the capital cost of the transmission infrastructure needed to serve the load), demand charge structure (how those costs are recovered monthly), energy charges (variable costs that track with consumption), and supplemental charges for any dedicated infrastructure. For loads that will operate at high utilization factors, a flat demand charge based on contracted capacity may be the appropriate structure. For loads with significant variability, a two-part structure combining a base demand charge with a coincident peak demand charge may better align revenue recovery with actual system cost. The rate design needs to be defensible in a rate case: the commission will ask whether the large customer is paying its proportional share of system costs or whether the structure shifts costs to other customer classes.

A practical test: run the large-load rate design through a fully allocated cost-of-service study. If the study shows the large-load customer is paying below its allocated share, the deficit will eventually appear as a rate pressure on other customer classes, and the commission will flag it. If the study shows the customer is paying above its allocated share, you may have a competitive disadvantage when the customer compares your service costs against other locations. The rate design should aim for cost-proportional recovery with a credible methodology behind it.

Interconnection Cost Provisions

The interconnection cost provisions govern who pays for what infrastructure. At minimum, these provisions need to address: the customer's responsibility for interconnection facilities (the dedicated infrastructure connecting the specific load to the transmission system); the allocation of network upgrade costs (the broader transmission system reinforcements triggered by the load); the payment structure (lump-sum, installment, or rate adder); any true-up provisions if upgrade costs exceed or fall short of estimates; and the disposition of infrastructure if the customer terminates service. For network upgrades in particular, the allocation methodology must be consistent with FERC's large-load rulemaking framework: FERC's rules govern what is permissible, and a state commission cannot override FERC on interstate transmission cost allocation.

The network upgrade cost-allocation question is where the most significant rate-case risk lives. The challenge for utilities is that two legitimate positions create tension: charging the data-center customer for the full cost of network upgrades triggered by their load is economically clean and politically defensible to residential ratepayers, but it may make the cost of interconnecting in your territory prohibitively expensive compared to locations where network upgrades are socialized. A utility in a state that needs data-center economic development may accept more socialization than one in a state where commission intervenors are aggressive on cost shifting. There is no universal answer, only a defensible methodology applied consistently.

Flexibility Provisions

The flexibility provisions are the most strategically important part of the large-load tariff because they convert a passive load into a grid management resource. A well-designed flexibility tariff provision specifies: the curtailment obligation type (emergency reliability curtailment, economic dispatch curtailment, or both); the notice period for each curtailment type; the maximum number of curtailment events per year; the minimum duration of each event; the minimum load reduction required to count as a curtailment event; the telemetry and metering requirements for verification; the compensation structure (a rate reduction, capacity payment, or both); the performance penalty for failing to meet the curtailment obligation; and the testing requirements to verify that the capability exists.

The design of flexibility provisions requires honest conversation with operations. The operations team needs to define what curtailment notice period is operationally meaningful: a 24-hour notice is useful for day-ahead economic dispatch but not for a real-time reliability event. A 4-hour notice is marginally useful for an emergency. A 15-minute notice is genuinely valuable for grid stress management but requires a data center with the technical capability to reduce load that quickly. Most hyperscale operators can shift workloads with a few hours of planning, but immediate load shedding to near-zero requires deliberate system design and is not universally available. Do not put curtailment provisions in the tariff that a majority of data-center customers will not be able to meet.

Defending the Tariff at the Commission

Every large-load tariff provision will eventually be scrutinized, either in a general rate case where the utility's cost recovery is reviewed, or in a specific proceeding triggered by a customer complaint or an intervenor challenge. Knowing where the challenges will come from helps you build the defense before you need it.

The Cost-Shift Challenge

The most common challenge from commission consumer advocates and residential customer intervenors is that the large-load customer is not paying its proportional share of system costs. The defense has three parts. First, the cost-of-service study: present a professionally executed fully allocated cost study showing the large load's proportional contribution relative to its share of system cost drivers. Second, the revenue adequacy argument: show that the large load's revenue, at the proposed rates, covers its allocated costs with the approved methodology. Third, the flexibility offset argument: if the tariff includes flexibility provisions, show the reliability value of having a dispatchable load asset and quantify how that value benefits other customers, for example through deferred capacity investment or reduced ancillary services costs. None of these arguments is self-executing: they require witness testimony, supporting analysis, and cross-examination preparation.

The Competitive Distortion Challenge

Competitive energy users and economic development advocates may challenge the tariff from the opposite direction: arguing that the rate structure is so onerous that it discourages data centers from locating in your service territory, costing the state jobs and tax revenue. This challenge matters particularly when neighboring states or utilities have more favorable large-load tariff structures. The defense here is demonstrating that the tariff structure is competitive, meaning that after all charges including interconnection costs, the total cost of power delivered is within a defensible range compared to comparable locations. This requires market intelligence on competing tariff structures, which your regulatory affairs team should be gathering as part of the competitive landscape assessment in the large-load readiness program.

The Precedent Management Problem

One of the subtler regulatory risks in large-load tariff design is precedent management. Every service agreement signed under the tariff creates a factual record that subsequent customers will cite. If the first data center in your territory negotiated favorable network upgrade cost allocation because the utility was eager to sign a precedent-setting customer, the second data center's lawyers will arrive with that agreement and argue for the same treatment. If the tariff provisions are clear and consistently applied, this is manageable. If the first agreement was an ad-hoc negotiation outside the tariff framework, you have created a precedent you cannot easily escape.

The discipline is: every large-load service agreement must be authorized under the tariff as filed. If a customer asks for something not in the tariff, the answer is to seek a tariff amendment through the appropriate process, not to grant it by individual negotiation. This is more constraining in the short term but far better for long-term rate design integrity.

The Long-Term Service Agreement

Data centers are not transactional customers: they make multi-hundred-million-dollar investments in facilities that expect to operate for 15 to 20 years. They want rate certainty. A well-designed large-load tariff enables a long-term service agreement (LTSA) that provides the customer with a known rate trajectory while giving the utility revenue certainty that supports capital planning. The LTSA is the commercial embodiment of the compact.

A typical LTSA for a hyperscale customer will include: a committed demand level with provisions for incremental increases; a rate structure with defined escalation terms or a formula for adjusting rates as underlying cost drivers change; an initial term of 10 to 20 years with renewal options; provisions for early termination with make-whole payments that recover the utility's stranded investment in dedicated infrastructure; and, increasingly, flexibility provisions that define the customer's curtailment obligations and the associated rate benefits.

The LTSA rate design challenge is rate certainty versus cost recovery adequacy. The customer wants a fixed or lightly escalating rate so it can model its energy cost over the full investment horizon. The utility needs confidence that the rate will cover its costs as fuel, transmission, and capital costs evolve. The solution most commissions will accept is a rate structure with a defined base demand charge (relatively stable), an energy charge that tracks actual variable costs, and a formula mechanism for periodic true-up of the base charge based on updated cost-of-service studies. This gives the customer reasonable rate predictability while protecting the utility's revenue requirement.

AI-Assisted Tariff and Agreement Drafting

Large-load tariff drafting and LTSA preparation are document-intensive processes where AI tools can meaningfully reduce the time and cost of producing complete, accurate drafts. The highest-value applications are: generating first-draft tariff sections from detailed prompts that incorporate jurisdiction-specific requirements, approved cost-allocation methodology, and standard language from prior filings; comparing a draft tariff section against the utility's existing tariff to identify conflicts or inconsistencies; drafting the boilerplate portions of LTSA exhibits (rate schedule tables, metering specifications, telemetry requirements); and summarizing commission orders and prior rate case decisions to identify relevant precedents for tariff language decisions.

The discipline required with AI-assisted tariff drafting is more stringent than in most contexts because tariff language has legal consequence. A single word choice in a cost-allocation provision can shift hundreds of millions of dollars of responsibility. Every AI-generated tariff section must be reviewed by a qualified regulatory attorney and, for technical sections, by the interconnection engineering team before it is used in a filing. The review is not perfunctory: the reviewer must be able to explain why the language is correct and defend it under cross-examination by commission staff or intervenors. The verification standard for tariff language is, in practice, the same as for a legal brief.

Key Takeaways

• The AI-data-center compact only works on a durable tariff foundation: a large-load service agreement signed under ad-hoc provisions creates rate-case exposure, cost-shift risk, and precedent problems that constrain every future data-center negotiation.
• A large-load tariff has four primary layers: service definition, demand and energy charges, interconnection cost provisions, and flexibility provisions; each layer has distinct design choices and distinct regulatory exposure points.
• The network upgrade cost-allocation methodology must be defensible simultaneously to data-center customers (who ask why they bear costs generators do not), to residential ratepayers (who ask why they subsidize technology companies), and to your commission (who asks whether the methodology is consistent with FERC's framework).
• Flexibility provisions are the most strategically valuable part of the large-load tariff because they convert a passive load into a dispatchable reliability asset, but the curtailment parameters must reflect what data-center operators can actually deliver operationally.
• Precedent management is a long-term rate design discipline: every deviation from the tariff negotiated with the first large-load customer becomes a baseline that the second customer's lawyers will cite, so consistency in tariff application is essential from the first agreement.
• Long-term service agreements require a rate structure that balances the customer's need for rate certainty with the utility's need for cost recovery adequacy over a 10 to 20 year horizon; a defined base demand charge plus a formula escalation mechanism typically satisfies both.
• AI-assisted tariff and LTSA drafting compresses document preparation time but requires qualified regulatory attorney review of every section before filing, because tariff language has legal consequence that the advisory AI standard alone cannot satisfy.