Who Pays for New Transmission Is the Live Question

When a gigawatt-scale load asks to connect, the grid it connects to usually has to be enlarged. The unresolved question is not whether those upgrades happen, but whose bill they land on.

A large load does not plug into spare capacity that is simply waiting. At the scale of the data-center requests now moving through Wisconsin, connecting a new customer often means building new generation and reinforcing the transmission that carries it. Someone has to pay for that steel, those transformers, and those miles of line. The debate over how that cost is divided, between the customer that triggered the buildout and the ratepayers who were already on the system, is the most consequential regulatory question in the state’s large-load pipeline. It is also the one least settled by existing rules.

We treat this as a cost-allocation problem rather than a question of whether any single project is good or bad. The mechanisms below apply regardless of who the applicant is.

The number that frames the debate

Before looking at any new proposal, it helps to hold one figure in view. Wisconsin ratepayers currently owe roughly $1 billion in remaining cost on generating plants that have already been retired: “stranded cost” for assets that are no longer producing power. That number is not an argument against building. It is a reminder that decisions made under one set of assumptions can outlive both the asset and the customer that justified it, and that the remaining balance does not disappear; it is collected from whoever is still on the system.

Every proposed allocation for new large-load infrastructure is, implicitly, being measured against that memory. The question underneath the technical filings is straightforward: if a commitment made today does not hold, who carries what is left?

Two default rules for a shared grid

The grid is a shared machine, and the wholesale market that operates much of it (the Midcontinent Independent System Operator, or MISO) already has conventions for splitting the cost of upgrades. Those conventions draw a line that matters enormously for large loads.

Broadly, a baseline reliability upgrade (one the system needed for its own health, independent of any single new customer) is typically socialized, spread across the network because the whole network benefits. An upgrade that exists only because one specific load appeared can instead be directly assigned to that load. The physical equipment can look identical in both cases. What differs is the reason it is being built, and that reason determines the bill.

MISO adds a further constraint that shapes the stakes: load growth is expected to be matched with accredited capacity, and the operator may require transmission upgrades to accommodate it. So the arrival of a very large customer does not only raise the question of who pays for a plant. It raises the question of who pays for the network reinforcement that lets the plant’s output actually reach the load, and whether that reinforcement is classified as a shared reliability need or a customer-specific cost.

There is no single correct place to draw that line. Socializing too much shifts the burden of one customer’s growth onto everyone else. Assigning too much can make the economics of a project impossible and push it to another state, since MISO spans fifteen of them and rules that differ at the edges influence where projects choose to sit. The classification is where much of the real money moves.

A tariff that makes the choice explicit

Wisconsin now has a concrete example of a utility trying to codify this split rather than litigate it customer by customer. A recently proposed data-center tariff offers an applicant two structured options.

Under the first, a full-cost option, the data center funds a dedicated new resource in its entirety and receives that resource’s output. The load pays for what the load requires, and the risk sits with the party that created it.

Under the second, a “capacity-only” option, the data center funds roughly 75 percent of a new plant’s cost, and the plant is counted toward system reliability for everyone. Because the plant now serves the broader system as well, the remaining share (about 25 percent of the capital, plus fuel) is recovered from other customers. Applied to a gas plant on the order of 700 MW and roughly $500 million in capital, that unfunded portion sits near $125 million. The presence of a nine-figure cost not directly borne by the customer that prompted the plant is a central reason the structure drew scrutiny.

Presented neutrally, the two options express the same underlying trade-off in different proportions. The full-cost path concentrates both cost and risk on the load and keeps existing customers insulated. The capacity-only path shares the cost, but it also shares a genuine benefit: a new, accredited resource that improves reliability for the whole system, not only for the data center. Whether the shared benefit is worth the shared cost is precisely the judgment a regulator is being asked to make.

The percentage is not the real question

It is tempting to reduce this to a single number: is 75 percent enough, should it be higher? But the more durable question is about method rather than any one figure.

A percentage is only as good as the assumptions behind it. If a large load underperforms its projected consumption, changes its operating profile, or leaves before the assets it justified are paid off, the shared portion does not vanish. It reverts to the same broad base of customers who inherited the earlier billion dollars in retired-plant cost. An allocation framework that looks balanced at full utilization can look very different under stress.

So the questions that decide whether an allocation is sound are less about the split at signing and more about its behavior over time. Does the framework specify what happens to the shared cost if the underlying load does not materialize as planned? Does it separate genuinely system-serving upgrades from customer-specific ones in a way that would survive an audit? Does it treat the reliability benefit to other customers as a real, quantifiable offset, or as a justification applied after the fact? These are the tests that determine whether today’s tariff becomes tomorrow’s stranded cost.

What planners should take from this

For anyone modeling a large project or setting policy around one, the practical lesson is that cost allocation is not a closing detail. It is a primary variable, and it is currently unsettled. The same megawatts can carry very different price tags depending on how an upgrade is classified, which option a tariff offers, and how a framework handles the case where reality diverges from the forecast. Treating that uncertainty as a modeling input, rather than a surprise discovered late, is the difference between a plan that survives regulatory review and one that stalls in it.

Wisconsin’s current cases will set reference points that other utilities and other states are likely to follow. That makes the design of these frameworks a long-horizon decision, not a one-off rate matter.

We will keep returning to a single institutional question: as large loads reshape the system, what allocation principle keeps costs with the customers who cause them while still recognizing the reliability value that new capacity delivers to everyone else?

Source: the Wisconsin AI Infrastructure Readiness Brief, regulatory and policy — cost allocation for large loads: the 100% full-cost and 75% capacity-only tariff options, the ~25% plus fuel recovered from other customers, ~$125 million on a ~700 MW / ~$500 million plant, the ~$1 billion in existing stranded-plant cost, MISO’s socialized baseline versus directly assigned load-specific upgrades, and load growth matched with accredited capacity.