On May 7, the Oregon Public Utility Commission approved Portland General Electric's Schedule 96, a large-load tariff aimed squarely at data centers over 20 MW. A few weeks earlier, Pennsylvania published what the Environmental Defense Fund called a first-of-its-kind model tariff for loads above 50 MW. By the latest count, 77 large-load tariffs are pending or in place across 36 states. Twenty-nine of them were approved in 2025 alone, up from 14 in the six years before that. The question of who pays for the data center boom is being answered in tariff language, one commission at a time.
For two years the sector has treated this as a design problem. What threshold defines a large load? How much of the network upgrade should the customer fund? How long should the contract run? Those questions are close to settled, and the answers are converging across states.
PGE's Schedule 96 makes large-load customers cover 100% of the distribution upgrades needed to serve them, sets minimum generation and transmission demand charges at 90% of contracted capacity regardless of actual usage, and runs contracts from 10 years up to 30 for loads of 220 MW or more. Pennsylvania's model requires large loads to pay upfront through Contributions in Aid of Construction and to fund any upgrade that "would not have been needed 'but for' the interconnection," whether or not other customers eventually benefit. Both add collateral requirements, load ramp schedules, and exit penalties. The mechanics are becoming standard.
Every one of those protections is a promise the utility now has to keep, and prove it kept, for the life of the contract.
Read Schedule 96 again, this time as a list of things a utility will have to demonstrate to a regulator, an intervenor, or a judge years from now.
The customer funds 100% of the distribution upgrades caused by their load. To bill that and defend it, you have to prove which upgrades were caused by that load and not by organic growth, reliability work, or a project that was already on the books. The minimum demand charge is 90% of contracted capacity regardless of usage. To collect it through a slow ramp or a stalled project, you have to prove the contracted capacity, track actual usage against it, and apply the floor correctly every billing cycle. The contract carries an early termination penalty tied to remaining demand obligations and unspent distribution investment. To enforce it, you have to know, at any moment, exactly how much of the committed investment has been spent and how much demand obligation remains.
Pennsylvania's "but for" standard is the sharpest version of this, essentially a causation test written into the tariff. A utility that cannot trace a given upgrade back to the specific interconnection that triggered it has no basis to assign the cost to that customer, and no defense when someone asks why. Causation is easy to assert in a filing and hard to prove in a hearing.
None of this is billing as usual. It is a multi-decade evidentiary burden, created the moment the tariff is approved and owed continuously from then on.
The reason this is hard has nothing to do with the tariff and everything to do with where the evidence lives. The facts that prove cost causation are scattered across systems that were never built to talk to each other: the interconnection queue, the engineering studies, the asset and work management systems, the outage and load data, the billing platform, and the general ledger. The trigger for a cost sits in one system. The cost itself sits in another. The load that justifies it sits in a third.
So the proof gets assembled the way most regulatory evidence still gets assembled: by hand, after the fact, when a rate case forces the issue. Someone pulls the work orders, someone else exports the load history, a third person reconciles it against the contract, and the causation story is reconstructed months or years after the decisions were made. That reconstruction is slow, expensive, and fragile, and it is exactly the point where allocations get challenged.
A Harvard Law analysis last year laid out how utilities can end up subsidizing data center growth by shifting costs onto everyone else's bill, often without meaning to. The mechanism is simple: if you cannot prove a cost belongs to the large load, it defaults to the shared revenue requirement, which means residential and small commercial customers absorb it. The tariff was written to prevent that transfer. Weak documentation quietly reinstates it. And when a commission cannot follow the causation trail, the safest ruling for them is to disallow the cost or reallocate it, which lands on the utility's recovery.
This is happening while the industry's capital plans are climbing faster than at any point in living memory. Investor-owned utilities raised their five-year capex plans 21% in the last year, to $1.4 trillion through 2030. NERC projects summer peak demand will grow 224 GW over the next decade, a 24% jump, with data centers driving most of it. Someone has to pay for that build-out, and residential customers are already feeling it: retail prices rose 33% from 2019 to 2025, utilities filed for $31 billion in rate increases last year, more than double the year before, and 73% of Americans say they worry about their bills going up.
Stephen Smith of the Southern Alliance for Clean Energy called the moment a gold rush, with utilities proposing investment against load forecasts that are at times "pure speculation." He has a point worth taking seriously even if you build for a living: forecasts move, and they move fast. AEP Ohio cut its own large-load forecast by more than half, from 30 GW to 13 GW, after its tariff was approved. A utility that signed contracts, sized upgrades, and allocated costs against the higher number needs a clean, current, auditable record of what was actually committed and actually built, because the ramp it planned for is not the ramp it got. Intervenors and consumer advocates know all of this, and they are reading these cases more closely than they have read anything in years.
Anyone hoping for one clean national rule to document against did not get it. FERC's data center interconnection order handed the job to the regional grid operators, directing six RTOs and ISOs to write their own rules. As one attorney put it, "the era of one national standard for data center interconnection is over before it began." FERC also directed grid operators to establish cost recovery agreements ensuring large loads bear their proportional share of network upgrade costs, and to develop new services for flexible loads that can curtail. Each of those is another allocation that has to be measured, tracked, and defended, and each varies by region. The transmission layer now carries its own proof burden, sitting on top of the retail one the states are writing. There is more to document, in more places.
This should sound familiar: it is the same lesson we learned the hard way on wildfire mitigation and system resiliency plans. When the filing becomes an enforceable standard, the plan is no longer the deliverable. The proof of execution is. And proof that gets reconstructed after the fact is weaker, slower, and more expensive than proof that the work generates on its own.
The fix is the same here as it was there: Stop treating cost allocation as a report you assemble at rate-case time and start treating it as a byproduct of the interconnection workflow. When a large load enters the queue, the causation trail should begin building itself: the study that identifies the upgrade linked to the load that triggered it, the work order carrying that linkage into the field, the completion evidence tied back to the contract, the usage metered against the 90% floor, the committed investment tracked against the exit penalty. Done this way, the rate case is as simple as a query.
That is an orchestration problem, and it is the work we do with investor-owned utilities: a workflow layer that sits on top of the interconnection, asset, billing, and financial systems already in place and connects them, so that every large-load decision produces an auditable, per-cause cost record as it happens. The model stays AI-assisted and human-approved, so an engineer can adjust an allocation with local knowledge and the reasoning is still captured and defensible.
Pressure-test one recent large-load interconnection. Ask your team to produce, today, the causation trail that ties a specific network upgrade to the specific load that triggered it. If it takes more than an afternoon, you have a proof gap.
Map every promise in your approved tariff to the system of record that will have to prove it: the minimum-take floor, the ramp schedule, the collateral draw, the exit-penalty math. Any promise without a clear owner and an auditable trail is a future disallowance.
Wire causation capture into the interconnection workflow instead of the rate-case scramble. The cheapest time to document why a cost belongs to a customer is the moment the decision is made, not the month the case is filed.
Prioritize the reconciliation across interconnection, asset, billing, and finance before you prioritize the next tariff revision. The allocation rules are converging. The ability to prove them is what will separate the utilities that recover their costs from the ones that eat them.
The tariffs are getting written well. The next competitive line is who can prove, on demand and without a fire drill, that the language was followed.
Want to continue the discussion? Schedule a chat with Tom.