New York Just Froze $10 Billion in Data Centers. The Blueprint Is What Actually Travels.

On Tuesday, Governor Kathy Hochul signed Executive Order 62 and made New York the first US state to freeze new hyperscale data center permits. The threshold is 50 megawatts of power draw, the pause runs up to twelve months, and the Department of Environmental Conservation will stop issuing any discretionary permit that had not already been deemed complete before the order landed. Bisnow puts the frozen pipeline at roughly $10 billion of early-stage projects in Upstate New York and the Hudson Valley. The wires ran it as a state regulator saying no to AI. The interesting piece is what she signed alongside it, and the fact that other governors were reading the draft in advance.

Headline: the pause is the small part. The payment gate is the big part, and it is designed to travel.

What Executive Order 62 Actually Does

Three moving parts, all separate, all consequential.

One, a moratorium on any new data center consuming 50 megawatts or more, in force until the Department of Public Service completes a Generic Environmental Impact Statement covering energy demand, water use, air quality, community impact, and noise. The GEIS timeline is up to a year. The 50 MW line is worth pausing on: the state legislature had passed a moratorium in June at a 20 MW threshold, which would have swept in almost every enterprise-scale colocation build in the state. The Governor moved the line up, which reads as a political concession to the trades and to Micron, and still ends up covering every hyperscaler and frontier lab project on the map.

Two, the Energize NY Development proceeding, which the Public Service Commission was already staffing after the January State of the State speech, gets a formal directive. The PSC is now instructed to write a new interconnection framework in which any large energy consumer (data centers explicitly named, industrial reshoring implied) either pays a higher, cost-of-service tariff that reflects the grid upgrades their load requires, or supplies its own power. The pay-or-supply framing is the piece the industry has to price.

Three, a proposed New York Grid Acceleration Fund that would require large data center loads to invest directly in state grid infrastructure as a condition of interconnect, and a community investment framework due within 60 days that codifies what a local data center deal has to deliver beyond the tax abatement (workforce contributions, childcare, direct payments). This is the piece governors in Virginia, Georgia, and Ohio have been asking their own staff to draft.

Why the Payment Gate Matters More Than the Pause

A one-year pause on new permits sounds severe if you are Meta or Google, and less severe if you have already spent two years on interconnect studies for a site somewhere else. Anyone who watches transmission queues at hyperscale scale knew the state was going to be an eighteen-to-twenty-four-month PPA process anyway. The pause adds some months, and mostly it puts unfinished paperwork into a different pile.

The Energize NY piece is different. Pay-or-supply says the state will no longer socialize the grid upgrades that hyperscale draw triggers. Either the data center covers the cost-of-service tariff that reflects the substation, transmission, and generation additions its interconnect requires, or the data center brings its own megawatts through onsite generation or a firm PPA. That is a re-underwrite of the entire economic case for a 50 MW to 500 MW site in New York. The number that used to sit inside the utility rate base moves onto the operator's pro forma. In our xAI gas turbines piece we put a $2.8 billion number on what supply-your-own-power actually costs at Colossus scale in Memphis. New York just told every operator in the state that Colossus-style math is the default.

The Grid Acceleration Fund idea, if it survives to a rule, is more aggressive still. It converts data center interconnect into a form of impact fee that funds statewide transmission, not just the hookup on your specific parcel. That is closer to how European member states treat industrial load, and it is a design choice a Republican governor could copy without changing the framing much.

Why the Blueprint Travels

Every governor whose grid is being loaded by AI has the same three incoming calls: the utility saying the queue is full, the local industrial association saying capacity is being eaten by out-of-state buyers, and the consumer advocate saying residential rates just went up. Virginia has been the loudest example. Its Data Center Alley in Loudoun County is now roughly 4 GW of live load, another 6 GW under study, and Dominion has publicly said it will need to double generation capacity to meet the interconnect queue. Georgia Power posted a 6,600 MW load-forecast increase for 2031 driven almost entirely by data centers. Ohio and Texas have each held press conferences on data center rate impacts in the last quarter.

None of those states have run a moratorium yet, because a moratorium is politically expensive and vulnerable to a preemption challenge on federal energy law. But every one of them has a rate case where a version of Energize NY could get bolted on. The New York framework separates the two politically dangerous pieces (a pause) from the two politically portable pieces (a payment gate and a community framework). Statehouses will keep the portable pieces.

Axios reported on July 15 that Hochul's team briefed at least five other governors on the draft before Tuesday. That is a coordination pattern the AI industry has not really seen at the state level before. Federal AI policy has been fragmented since the Trump administration pulled the review order in the spring, which we tracked in the review-order piece, and state legislatures have been trying to fill the gap ever since. New York just handed them a template.

What This Does to the 2027 Buildout Math

The frontier labs and hyperscalers have signed compute commitments well past the point where existing grid can deliver. Anthropic's five-year, $200 billion TPU commitment with Google, which we ran the numbers on in the $200B math piece, is anchored on gigawatt-scale delivery starting in 2027. OpenAI's Vera Rubin commitment with Nvidia lands on the same window. Meta's Louisiana Hyperion buildout was just extended to 5 GW and $50 billion, per the July 13 disclosure. Every one of these commitments implicitly assumes the grid interconnect is a solved problem and the marginal cost of a megawatt sits with the utility.

Two things change once payment gates start showing up state by state. First, the marginal cost of a megawatt moves onto the buyer's balance sheet, which raises the price of any inference call that runs on that megawatt. This is why the pieces of the industry that have been quietly buying their own generation, xAI with gas turbines, Amazon with the Susquehanna nuclear draw (still parked at the FERC bypass watch), Microsoft with the Three Mile Island restart we wrote up in the nuclear restart thesis, were early rather than eccentric. Onsite and firm-PPA power stops being a nice-to-have and becomes the only path that survives contact with a pay-or-supply rule.

Second, siting starts routing around jurisdictions that have adopted the framework. The states that inherit the growth are the ones with generation surplus and a political appetite for large industrial load: Wyoming, the Dakotas, parts of Louisiana and Mississippi, the Ohio counties still under existing rate structures. The states that copy New York keep their rate base intact and lose the tax abatement fights. This is the choice most governors are actively making right now, and it is why the Hochul team was sending the draft around.

The Micron Carveout

The 50 MW threshold, versus the legislature's 20 MW, is not an accident. Micron's $100 billion Clay campus outside Syracuse is the largest single industrial project in the state, funded in part by CHIPS Act money, and it would have been swept into any 20 MW pause. Micron is a fab, not a data center, but the definition of the load matters at the permitting margin, and the 50 MW line conveniently draws itself around the trades and semiconductor reshoring while landing squarely on the AI workload category. That is a political read the Governor was willing to make on the record, and it lines up with the federal industrial policy that everyone in this fight still agrees on.

The read for AI operators: the definition of what counts as a data center will get argued at every state boundary from here. Colocation providers are already lobbying Albany for exemptions on any facility not primarily serving external customers. Expect similar carveouts in the copycat states, and expect the frontier labs to press for their training clusters to be classified as research infrastructure rather than commercial data centers. Whether those arguments hold is a permit-by-permit fight, and it will get expensive.

Our Take

The New York moratorium is the piece of the story that ran on the wires, and it is the piece that will lapse in about a year. The Energize NY pay-or-supply framework is the piece that will outlive it, and the piece other governors are copying. Read this as the moment state utility policy caught up with the AI capex cycle: the same year the industry publicly committed something on the order of $700 billion to 2026 infrastructure buildout, one of the biggest energy markets in the country decided the load has to pay for its own grid. That is not an anti-AI signal. It is a repricing signal, and it lines up with the industrial policy the federal government has already committed to on domestic semiconductors and clean energy.

Practical implication for anyone underwriting AI compute commitments. Three things to watch over the next ninety days. First, which state files a version of Energize NY next; Virginia and Georgia are the highest-probability first movers, and if either one goes, the rest go inside two quarters. Second, whether the PSC's draft interconnection tariff prices grid upgrades in a way the operators can live with, or in a way that pushes them onto onsite generation by default. And third, whether the existing $10 billion of paused New York projects file for the "already deemed complete" exemption or start shopping themselves to jurisdictions further west. The answer to that third question tells you whether the 2027 gigawatt buildout is still coming online in New York, or whether the map just quietly redrew itself.

We are tracking state-level AI policy on the California AI bill sprint and the corresponding grid-buildout data on our Anthropic and OpenAI provider pages. The Hochul order is dated. The framework is not.