Two regulators just moved the risk of a forecast back onto whoever made it
Serguey Asael Shinder3 min readTwo decisions this week, two thousand miles apart, answer the same question: when a utility builds for demand that has been announced but not signed, who holds the asset if the demand does not arrive?
North Carolina's answer was to refuse. The Utilities Commission denied Duke Energy a 255 MW gas turbine, calling the $584 million price "staggering" and finding that because much of Duke's projected growth "appears to be based upon anticipated data center customer additions", that growth is "insufficiently reliable for the Commission to act at this point."
California's answer was to reassign. Newsom signed seven bills; the operative ones direct the CPUC to make data centres pay for the transmission and distribution upgrades their load requires, and make them carry a larger share of wildfire mitigation and liability.
Refusal and cost-assignment look like different remedies. They are the same move: stop letting a forecast made by one party become an obligation carried by another.
Why this is structurally hard, and not merely a matter of regulators being tougher:
An enquiry pipeline is real information that cannot be added up. A hyperscaler evaluating sites talks to utilities in several states. Each utility hears a serious, well-resourced, genuinely intended project. Each puts it in a forecast. One of them is right. The sum of the forecasts is not a forecast of anything, and no participant is lying.
The asymmetry of who is bound. A data centre developer can stop. A 255 MW turbine, once built and in rate base, cannot — the ratepayer is committed for thirty years to a decision justified by a customer who was free to leave. That asymmetry is the whole of the North Carolina order, and it would hold even if every forecast were honest.
"Fair share" is easy to legislate and hard to compute. California's wildfire liability provision is the sharp end. Assigning a portion of a system-wide risk cost to one customer class requires a causal story about marginal contribution that does not currently exist in a defensible form. That will be litigated, and the number that survives litigation is the one that matters.
What I would watch, in order.
Does the load get contracted? The cleanest resolution is that hyperscalers sign firm, take-or-pay commitments and the forecast becomes a receivable. Some of this is already happening; whether it happens at the scale of the capital plans is the open question.
Does refusal spread? One commission declining one turbine is a data point. Duke's $103 billion capital plan is the largest on file at any regulated US utility, and it was built for this boom. If the reasoning in this order is adopted elsewhere, a lot of filed plans need a different justification.
And does the cost-assignment survive contact with siting? A state that makes data centres pay full freight is competing with states that do not. That is not an argument against doing it — it is the reason it will be tested politically within a year, and the reason the disclosure provisions may end up mattering more than the cost ones. Information about who is asking for what, published locally, is harder to arbitrage than a price.