This week, cost allocation stopped being an abstract principle and became a contract. Duke Energy’s two North Carolina utilities — Duke Energy Carolinas and Duke Energy Progress — filed a settlement with the state Utilities Commission that rewrites the rules for the largest customers on the system: data centers and other large users. The agreement, announced October 7 and negotiated with the North Carolina Public Staff, sets hard thresholds for who counts as a large load, requires upfront payment for grid infrastructure, and imposes minimum bills and exit fees designed to keep the cost of serving massive new demand off the bills of ordinary households.
The signatories are what make this unusual. Alongside the Public Staff — the independent agency that represents utility customers — the settlement was joined by Amazon, Google, Meta, and Microsoft, plus the Carolina Industrial Group for Fair Utility Rates and the U.S. Department of Defense. That is, the largest data center operators on the planet agreed to rules that make data centers pay more. For readers of our earlier piece on how utilities plan for large loads, this settlement is that analysis in action: tariff design and cost allocation, negotiated with the cost causers themselves, aimed at the fairness question that piece ended on.
The thresholds: targeting scale, not sector
The rules are deliberately written around size rather than industry. They apply to customers whose demand exceeds 50 megawatts and who operate consistently near peak capacity — reported as an 80 percent load-factor threshold — or to any customer whose total demand tops 150 megawatts. Customers meeting the thresholds would be moved onto a High Load Factor rate schedule, a separate rate class for very large loads.
The design is elegant in its honesty. A 50-megawatt threshold catches data centers, which run flat at 70 to 90 percent of contracted capacity around the clock, while the “near peak” condition lets variable industrial users with genuine off-peak relief slip through. The 150-megawatt catch-all exists for the rare customer whose sheer scale dwarfs everything else regardless of shape. As one local account of the settlement put it, the thresholds would likely limit the rules to the largest data centers, since most industrial users fall short of 150 megawatts and cut their power use at night and on weekends.
The High Load Factor schedule itself is the tariff-design move our earlier planning piece anticipated: a distinct class for customers whose load profile drives system costs in a specific way, with pricing built around contracted demand rather than metered consumption. Rather than stuffing very large, very flat loads into a general industrial schedule and hoping the average works out, Duke is pricing the load shape it actually receives.
Paying upfront: dedicated versus shared infrastructure
The settlement’s core mechanism is the upfront payment, and it distinguishes carefully between two kinds of grid investment. For facilities built exclusively to serve the connecting customer — the classic example is a substation — the customer must make a nonrefundable, upfront payment covering the full cost. There is no ambiguity here: the connection assets are yours, you pay for them before they exist, and you do not get the money back.
For grid upgrades that serve all customers, such as transmission lines, the requirement takes a softer but still binding form: upfront deposits and security guarantees. The customer posts collateral against its share of shared infrastructure rather than paying the full cost outright. Duke’s framing is direct. In the company’s words, “data centers will pay upfront for all costs to connect to the grid,” and the arrangement is meant to shield other customers from those costs “in a way that protects reliability.”
The new terms would apply to contracts signed after June 1, 2026, and Duke notes that the settlement memorializes and extends protections the company says it began implementing in 2024 — an early-mover posture that the company is now trying to lock into a commission-approved form. But there is a meaningful gap in the architecture that the settlement leaves for later: the Utilities Commission still has to approve the methodology for deciding what share of network upgrades gets assigned directly to a project. Transmission upgrades that serve a data center and also relieve regional congestion and improve reliability for nearby towns present exactly the shared-benefit problem our earlier piece described. The principle — large loads pay their own way — is settled. The apportionment math is not.
The minimum bill and the exit fee
The second layer of protection addresses the forecasting risk that haunts every utility planning for lumpy load. New large-load customers would pay a monthly minimum bill set at 75 percent of their projected demand — meaning a data center that forecasts big usage and then underdelivers still pays for most of what it said it would need. This is the minimum-bill provision from the planning playbook, applied at full strength: it converts the customer’s own projection into a binding financial commitment, which in turn disciplines speculative interconnection requests.
The commitment deepens with the exit fee. Early termination of a new contract could trigger damages ranging from $25 million to more than $587 million, depending on the customer’s demand and the contract’s length. These figures are the utility’s answer to the stranded-cost risk: if a customer reserves hundreds of megawatts, induces the utility to build, and then walks away, the remaining ratepayers should not absorb the loss. The scale of the numbers tells you how seriously both sides take the stranding scenario — and it explains why the hyperscalers, who rarely sign short-dated power contracts anyway, could accept terms that would terrify a smaller user.
Taken together — upfront payments, minimum bills, exit fees — the settlement amounts to a staged-commitment regime of the kind that is becoming standard nationally. The utility’s investment is released against the customer’s demonstrated commitment, and the commitment is priced high enough that nobody makes it lightly.
Why the hyperscalers signed
It is worth pausing on the most surprising fact in the filing: Amazon, Google, Meta, and Microsoft endorsed it. These are the customers the rules constrain, and they signed anyway. The reasons are not hard to infer for anyone who watches rate cases.
First, certainty is worth money. A negotiated settlement with the Public Staff and the commission’s blessing produces predictable, stable rules — far preferable to years of contested dockets with uncertain outcomes, or to the alternative where an affordability backlash produces genuinely punitive tariffs. For companies making 15-to-20-year capital decisions, knowing the cost structure is nearly as valuable as keeping it low.
Second, the industry is managing a political problem, not just an economic one. Data centers are visibly driving rate cases across the country, and the fastest route to a hostile regulatory environment is for household bills to rise while the public believes big tech got a free ride. Endorsing cost-allocation protections buys social license — and keeps the economic development competition on friendlier terms, since regulators in neighboring states are watching each other’s rules for large loads.
Third, the Carolina Industrial Group and the Department of Defense at the table remind us this settlement is also about protecting existing large customers from being saddled with costs caused by the newcomers. Everyone at the negotiating table except the utility had an interest in making sure somebody else’s load did not show up on their bill.
Why clean energy groups didn’t
No clean energy group signed on, and the reason is the settlement’s conspicuous silence on where the power comes from. The rules say who pays for the grid; they say nothing about requiring any of the electricity to be clean. Matt Abele of the North Carolina Sustainable Energy Association criticized the proposal as falling short, arguing that these facilities should be required to pay their own way while “increasingly supply more of their own demand with clean, low-cost, and reliable resources.”
The criticism points to the settlement’s real limitation as climate policy, as opposed to rate policy. Cost allocation is necessary — households should not subsidize hyperscale load — but it is not sufficient for advocates who want new large loads matched with new clean generation. The settlement treats load as a cost-allocation problem and leaves the resource question entirely to other proceedings. In a state where Duke’s long-range resource plan remains contested, that separation is doing a lot of work.
Governor Josh Stein’s role sharpens the point. Stein had pushed Duke Energy to create a large-load tariff like this one, and the settlement delivers the rate-design half of that ask. The clean half remains undealt with — delegated, implicitly, to the integrated resource planning process and to whatever procurement the market produces on its own.
The rate case in the background
The settlement’s timing matters because Duke is separately asking the Commission for more money from the same households it promises to protect. Duke Energy Carolinas filed a rate case in November 2025 seeking roughly $1 billion in new revenue — about a 15 percent increase over two years — and in June revised the request downward to approximately $622 million, or about 9.3 percent, after public hearings on affordability. On the Progress side, the Attorney General’s office calculated that residential customers would face about 9.3 percent more over two years — a higher figure than the company’s testimony initially suggested — and Attorney General Jeff Jackson declined to sign the rate-case settlement on the grounds that it was still too high for families. New rates are expected to take effect January 1, 2027, pending the Commission’s decisions.
There is no contradiction here, exactly: the rate cases cover past and planned investments in reliability, storm hardening, and system growth, while the large-load settlement governs future contracts for new mega-customers. But the sequencing creates an unavoidable optics problem. A household told its bills are rising 9 percent while being assured that data centers will pay their own way is entitled to ask why the assurance arrives simultaneously with the increase — and whether the protection applies to the infrastructure spending driving the rate case, or only to load contracted after June 1, 2026. The answer is the latter. That distinction is legally coherent and politically uncomfortable.
The open question: will it actually protect households?
Settlements are promises, and this one is promising about the parts of the problem that are easiest to measure. The genuine uncertainties sit in three places.
The first is the unapproved allocation methodology for shared network upgrades. Until the Commission approves how to apportion transmission costs between the interconnecting customer and everyone else, the household protection exists as a principle rather than a formula — and the history of cost allocation is the history of formulas that drift toward whoever argues hardest.
The second is threshold gaming at the margins. A 49-megawatt project that consistently runs near peak gets different treatment than a 51-megawatt one, and data center developers are adept at designing around bright lines — splitting campuses into phases, timing contracts around the June 2026 cutoff. The 150-megawatt catch-all helps, but thresholds always create edges, and edges invite planning.
The third is the oldest question in load forecasting: what happens if the load does not materialize, or materializes differently than projected. The 75-percent minimum bill and the exit fees handle the customer’s departure; they do not handle the subtler case of a facility that uses far less than contracted while staying within its contract, or of a market where AI investment cools and the region has overbuilt for a wave that crested. Committed payments protect ratepayers from the downside of the customer’s behavior, not from the downside of the industry’s hype cycle.
None of this means the settlement is hollow. The commitment architecture — upfront money, binding minimums, serious exit costs — is the strongest version of large-load cost allocation any southeastern utility has put forward, and the fact that the hyperscalers signed makes it the emerging industry template rather than a one-off imposition. But the settlement answers the question “who pays for the grid connection” more convincingly than it answers “who pays for the grid.” The latter is being asked right now, in the rate cases, by households being told their bills are going up.
What to watch: the Commission’s decision on the settlement, expected around mid-November; the follow-on proceeding that will set the shared-upgrade allocation methodology, which is where the household protection will actually be priced; and whether regulators in other load-growth states — Virginia, Texas, Georgia — treat the Duke template as a floor to match or a ceiling to exceed. The 50-megawatt line has been drawn. The fight now moves to everything that line does not cover.