State-level data-center siting moratorium risk

One-line summary: New York's Executive Order 62 (2026-07-14) is the first US statewide pause on new hyperscale (≥50 MW) data-center permits — with no calendar end date, a 2:1 favorable poll, and a parallel legislative instrument — establishing that the siting and permitting step, which nearly every ai-infrastructure chain in this book treats as a given, is now a live political variable.

Why this concept exists

The live signal feed is 47% ai-infrastructure by distinct chain (2026-08-06 breadth report), and essentially all of it depends on one unstated assumption: that announced large load gets sited, permitted and interconnected somewhere in the United States. The wiki already tracks interconnection queues, FERC large-load policy, transformer lead times and grid capex. It did not have a page for the environmental-permit gate, which is a state power and, as of July 2026, a demonstrated one.

This page is a risk register for the dominant cluster, not a thesis. It names no tradeable of its own — the exposure is embedded in chains already in the book.

The forcing function

New York Governor Kathy Hochul signed Executive Order No. 62 on 2026-07-14, making New York the first US state to impose a statewide moratorium on new hyperscale data centers (From 2026-08-06-autoresearch-ny-eo-62-hyperscale-datacenter-moratorium).

Terms, from the order itself:

• Threshold: 50 MW or more. The Department of Environmental Conservation must hold pending permit applications "in abeyance" for data-center construction or expansion not deemed complete before 2026-07-14.
• No fixed sunset. The moratorium runs "until DPS submits its report of the final Generic Environmental Impact Statement and associated findings statement." The widely-reported "one year" is an expectation, not a termination date in the order. For a developer, an unbounded pause is materially worse than a dated one.
• Grandfathering: does not apply to permits already deemed complete before 2026-07-14.
• Exemptions: facilities "primarily used for manufacturing, research, education, or medical care," including accredited academic research and Empire AI consortium facilities.
• Agency work during the pause: DPS runs a public GEIS process covering energy demand, water use, air quality and noise, forms a Data Center Interconnection Working Group within 60 days, and develops grid-acceleration fund mechanisms; DEC assesses water-withdrawal regulations (report within 12 months); Empire State Development creates a Community Investment Framework within 60 days.

Why it is durable, and therefore replicable

• A June Siena Research Institute poll found 46% thought a one-year moratorium on new large-data-center permits would be good for the state, against 21% who said it would be bad — better than 2:1 on a technical siting question. (From 2026-08-06-autoresearch-ny-eo-62-hyperscale-datacenter-moratorium)
• The New York State Legislature separately passed a first-in-the-nation data-center moratorium, so the policy is not purely executive and survives a change of governor. (Same source)

Politically cheap + popular + already legislated is the profile of a policy that gets copied. That, not New York's own megawatts, is the exposure.

What it bears on

• ferc-large-load-to-dc-gridscale-construction — a federal large-load pathway does not help if a state holds the environmental permit in abeyance. Two independent gates; this is the newly-active one.
• phantom-data-center-load — corroborates. A moratorium mechanically converts announced-but-unpermitted pipeline into exactly the phantom load that concept describes. Note the double edge: this cuts for phantom-load and against the capex chains that read the same pipeline as real.
• ai-capex-to-power-and-materials-cascade, datacenter-construction-electrical-picks-shovels, pwr-transformer-moat-to-eps-doubling, aging-grid-replacement-to-td-capex-supercycle — the equipment/EPC demand leg is a function of projects breaking ground.
• nee-dominion-virginia-scc-block-risk — the sibling hypothesis in the state that does hold the marginal megawatt. EO 62 raises the prior that a Virginia version happens.
• pjm-capacity-prices-to-nuclear-premium, nuclear-baseload-for-ai-data-centers — ambiguous sign. Suppressed new load is bearish capacity prices; but a moratorium that pushes load toward behind-the-meter and existing-nuclear-adjacent siting is bullish the co-location premium.
• electricity-price-drivers-decomposition — retail-rate politics is the stated grievance driving these orders.

Held honestly

EO 62 does not falsify any existing chain today. Grandfathering protects the in-flight pipeline, the exemptions are broad, and New York is a small share of national siting. The correct disposition is a contradiction-to-watch at the cluster level — one state, one data point — not a downgrade of individual mechanisms. Extrapolating one state to a national trend is precisely the error vault/_meta/CALIBRATION.md keeps recording.

Update (2026-08-07) — the duration question is answered as far as public precedent allows, and the "one year" is confirmed absent from the order

From 2026-08-07-autoresearch-ny-geis-duration-base-rate-eo-62-tail. This was flagged on 2026-08-06 as the highest-value follow-up in the book. Three findings:

1. The unbounded read is confirmed from the primary text — a second, independent confirmation. Executive Order 62, verbatim: "Until DPS submits its report of the final Generic Environmental Impact Statement and associated findings statement, the Department of Environmental Conservation (DEC)… is directed to hold in abeyance all applications for any discretionary permit…" The end of the pause is an event, not a date.

2. Practitioners are split, and one side is wrong. Davis Wright Tremaine reads it correctly — "The Executive Order sets no deadline for DPS to complete the GEIS." Carter Ledyard & Milburn asserts an outer boundary — "Up to one year (by July 14, 2027)… the outer boundary for project timelines under the current pause." That date is not in the order. Anyone modelling a hard July 2027 resumption is modelling the press framing. Recorded as a live disagreement, not silently resolved.

3. A NY GEIS duration base rate exists, and it is very wide — sorted by contestedness, not complexity:

The narrow, uncontested technical GEIS closed in nine months; the statewide, environmentally charged one took seven years. EO 62's GEIS has the fracking process's structural features, not the code update's — it mandates "a formal public process, including public comment and a public hearing," and its scope spans energy demand, water, air, noise and disproportionate impacts on disadvantaged communities. Combined with the durability signals already recorded here (Siena 46/21, a parallel legislative moratorium), the honest statement is: central case over a year; the under-12-month left tail requires this to behave like a routine code update; the right tail is genuinely multi-year with precedent behind it. This constrains the shape of the distribution, not its mean — three data points across 37 years is a thin base rate and analogous rather than comparable.

What actually makes this monitorable — the interim milestones are dated even though the terminus is not:

• 2026-10-12 (90 days): DPS must "report to the Commission" on transmission owners' practices for studying data-center system impacts (per the order).
• 2026-12-31: technical conference (per DWT).
• 2027-02-12: white-paper publication (per DWT).

A slip in the October or December milestone is the earliest available evidence that the GEIS is heading for the long tail, and it arrives roughly nine months before any "one year" expectation would be tested. That is the conversion of an open-ended risk into a monitorable one that this follow-up was aimed at.

Narrowing the blast radius: per the primary text plus DWT, the pause bites only where a discretionary DEC approval is required. Projects already holding all required state permits, and those proceeding solely through local permitting, route around EO 62 entirely — alongside the manufacturing/research/education/medical carve-out already recorded.

Open questions

• How much permitted-but-unbuilt NY capacity was grandfathered? Unmeasured. Without it the near-term revenue effect on EPC/equipment names is unknown.
• How long does a New York GEIS actually take? Answered 2026-08-07 — see the update above. Base rate ~9 months (uncontested) to 7 years (contested); central case over a year; monitorable via the 2026-10-12 / 2026-12-31 / 2027-02-12 milestones.
• Does DPS running the GEIS change the pace? Every precedent in the base rate above was run by DEC or DOS. DPS is a utility regulator with its own procedural calendar. Unexamined and potentially material.
• What happens to applications filed but incomplete on 2026-07-14? "Held in abeyance" is not "denied," but queue position and re-review burden on resumption are unaddressed in the sources.
• Do the executive order and the legislative moratorium have the same thresholds, exemptions and duration? Unresolved which binds after the EO lapses.
• Which state follows? No second state has acted. Until one does this is a single data point.
• Does a NY pause destroy load or merely relocate it to PJM/ERCOT/MISO? If demand relocates, the cluster-level effect is near zero and the trade is geographic, not directional.

Sources

• 2026-08-06-autoresearch-ny-eo-62-hyperscale-datacenter-moratorium
• 2026-08-07-autoresearch-ny-geis-duration-base-rate-eo-62-tail — primary EO 62 text + the NY GEIS duration base rate
• 2026-08-24-autoresearch-shapiro-pa-eo-2026-05-pjm-siting-gate — PA EO 2026-05 primary + FERC ER26-3380 calendar hold
• 2026-08-21-podcast-all-in-podcast-dario-defends-himself-datacenter-panic-ai-doomer — Chamath/Jason/Sacks political-replication color

Update (2026-08-10) — ⚠ Texas gates too, twenty days later. This is no longer a one-state data point.

From 2026-08-10-autoresearch-datacenter-siting-gate-load-relocation. The "one state, one data point" disposition recorded above on 2026-08-06 is now obsolete, and it was obsolete within four days of being written.

On 2026-08-03, Texas Governor Greg Abbott directed the PUCT and ERCOT to run a "comprehensive verification and audit" of every data center in the ERCOT interconnection queue, halting approvals until it completes. ERCOT told Fox News the directive "effectively pauses all data center projects."

• Scale: 1,800+ projects, 474 GW of interconnection requests, ~90% of it data centers — "more than five times Texas's record peak electricity demand for ERCOT." Texas Tribune counts 335 operating data centers and 248 planned projects in the state.
• Abbott, verbatim: "Any data center project that fails to comply with the verification and audit process to protect the reliability and resilience of the Texas electric grid must be denied."
• Audit scope: tax incentives received, power use and self-generation, water consumption and cooling, community-impact mitigation, and facility ownership.
• The end-condition is unbounded here too. No completion date is stated in the directive or the reporting — structurally the same defect this page already documented for EO 62, in the state that holds the largest queue in the country.
• Already mechanically binding: ERCOT paused its "batch zero" transmission planning study and will request a good-cause exemption to regulatory timelines at its 2026-08-20 meeting. That is a transmission-planning deferral — it hits the engineering-and-construction leg directly, not the generation leg.

Why this changes the concept's weight, not just its evidence

The 08-06 framing was that EO 62's importance was replicability, not New York's megawatts. Replication has now occurred, in twenty days, in the largest market, under a governor of the opposite party and on a different legal instrument (grid-reliability/interconnection authority rather than state environmental permitting). Two independent legal pathways reaching the same operational result is a much stronger durability signal than a single popular executive order. Conviction medium → medium-high.

The escape valve is written into both gates — and it is off-grid, not out-of-state

This is the load-bearing detail. Purely behind-the-meter on-site generation is exempt from both. Texas: projects "utilizing purely 'behind the meter' on-site generation and not interconnecting through ERCOT may not be subject to this directive"; additionally ~50 Texas counties sit outside the ERCOT region (El Paso, parts of the Panhandle, Far East Texas). New York: EO 62 reaches discretionary state environmental permits, not local approvals. Reporting confirms developers are already using it — "some data centers are building on-site power generation that allows them to bypass traditional connection to the grid."

So the rotation these gates force is grid-connected → off-grid, not state A → state B. See btm-onsite-generation-to-bloom-fuelcell-gev-turbine, which now has a policy forcing function alongside its original physics one.

The destroy-vs-relocate question is answered — and the answer is "neither"

This was flagged as the highest-value open question in the book on 2026-08-07. The relocation leg fails on two independent grounds: (a) the largest destination market gated itself, and (b) the remaining destinations are already capacity-short — PJM faces a shortfall that could reach 15 GW by 2030 with a 6.6 GW deficit for 2027–2028, ERCOT's large-load queue quadrupled from 63 GW (end-2024) to ~226 GW (~77% data centers targeting 2030), and MISO is designing an expedited large-load process outside its regular queue.

⚠ Held honest — the counter-weight is real and cuts the other way

The statewide moratorium wave is mostly failing. As of June 2026, 14 states were considering restrictions and bills were introduced in 11 — but Virginia, Georgia HB 1059, South Dakota SB 232 and Wisconsin SB 1061 all failed to advance, and Ohio's proposed constitutional amendment to ban data centers over 25 MW failed to qualify for the 2026 ballot, gathering only ~70,000 of the 413,488 signatures required. Judged on legislation alone, the trend is weak.

What is not failing is the county layer. In Indiana alone: 11 counties with data-center ordinances, at least 17 with temporary moratoriums, and 2 (Marshall and Cass) with outright bans. County gates are more numerous, less legible, and harder to route around than a statewide order — and NY EO 62 explicitly does not pre-empt local permits. Anyone modelling this risk from state legislation is measuring the layer that is losing.

This is a genuine tension and it is not resolved: executive action is succeeding, legislation is failing, and local action is proliferating. They are three different mechanisms with three different durabilities.

Related (added 2026-08-10)

• siting-gate-plus-stranded-capacity-to-demand-side-beneficiary-rotation — the Texas gate satisfies that hypothesis's explicit "a second state adopts an EO 62-style gate" graduation test.
• btm-onsite-generation-to-bloom-fuelcell-gev-turbine — the exemption is the trade.
• legislative-divergence-base-rate — the state-bill failure rate above is a clean instance of that concept: a bill's count carries almost no information about its effect.

---

Update (2026-08-11) — ⚠ tension with yesterday's "neither destroy nor relocate" conclusion

From 2026-08-07-feed-semianalysis-spacex-10gw-in-2027-microsoft-largest-offtaker.

On 2026-08-10 this project concluded that a siting pause neither destroys nor relocates load, because the escape valve written into both the New York and Texas orders is behind-the-meter on-site generation — off-grid, not out-of-state. Today's source partially contradicts the "not out-of-state" half:

• SpaceX built "an onsite generation plant 1km across the border to avoid permitting" — a jurisdictional move, i.e. relocation, undertaken specifically to escape a permitting regime.
• The 2027 plan requires SpaceX to "scout all over the country to find suitable land, with easy permitting and access to gas" — permitting friction treated as an explicit site-selection variable.

How to hold both. The 08-10 finding was about grid-interconnected load facing an interconnection-queue gate, where relocation fails because the alternative regions are themselves capacity-short (PJM 15GW short by 2030; ERCOT's large-load queue 63 → 226GW). Today's finding is about permitting friction on on-site generation, which is a different gate with different geography. So the refined statement is: grid-interconnection gates don't relocate load (no destination has headroom); permitting gates do relocate it (permitting regimes vary by county and state, and the mover only needs land and gas).

That refinement is worth more than either claim alone, because it says which kind of gate produces which kind of response — and the two gates now in force (NY EO 62, the Texas PUCT/ERCOT audit) are both interconnection gates, which is why the off-grid response dominates in those two cases specifically.

⚠ Single paywalled source, and the "1km across the border" case is one anecdote, not a pattern.

Update (2026-08-24) — third large-market gate: Pennsylvania EO 2026-05. Corroboration, not a re-rate.

From 2026-08-24-autoresearch-shapiro-pa-eo-2026-05-pjm-siting-gate. Governor Josh Shapiro signed Executive Order 2026-05 on 2026-08-18, converting the previously voluntary Governor's Responsible Infrastructure Development (GRID) standards into binding conditions for DEP permitting and the Computer Data Center Equipment sales-and-use tax exemption for projects with peak demand > 25 MW — a lower threshold than New York's 50 MW EO 62.

This is not a moratorium. Two tracks remain open. Track 1 (GRID Consent Order) keeps rolling DEP review; Track 2 (no Consent Order) loses rolling review, Fast Track, PAyback, and Permit Decision Guarantee. Local municipal approval is a prerequisite to DEP issuance. GRID requires developers to bring incremental capacity in the same PJM LDA, pay interconnection/network-upgrade costs, and meet a clean-firm share of 10% (2027) → 32% (2035). Fast Track is removed; NDAs banned. Shapiro directs Special Counsel to advocate at the PA PUC that utilities curtail DCs first unless they bring their own capacity, and to assign ER26-3380 Reliability Backstop Program costs to DCs.

DEP: 20 permit applications, 14 tax-exemption sites, 100+ proposed; Shapiro: no AI data center yet operating in Pennsylvania, only 5 have permits to start. The order cites the IMM: data centers = $29.4B / 46% of the last four BRA charges even with the collar.

Replication, 35 days, three instruments: NY (environmental permit, 50 MW, Jul 14) → TX (interconnect audit, Aug 3) → PA (permit-condition + local veto, 25 MW, Aug 18, PJM). Opposite-party / different-legal-pathway replication already moved conviction medium → medium-high on 08-10. A third state is corroboration of the pattern, not a further re-rate. Held medium-high.

The BYOC / same-LDA incremental-capacity requirement is a policy forcing function for btm-onsite-generation-to-bloom-fuelcell-gev-turbine, same sign as the Texas BTM exemption. POWER also names Illinois (Pritzker paused the Data Center Investment Program) and Florida SB 484 — not fetched this pass; do not mint from a roundup sentence.

Political-replication color, independent speaker: chamath-palihapitiya in 2026-08-21-podcast-all-in-podcast-dario-defends-himself-datacenter-panic-ai-doomer was "really shocked" by Abbott and Shapiro; jason-calacanis (same source) reads both as performative responses to constituent opposition ("they just want to make sure you're not tapping into the grid"). david-sacks (same source): "Data centers are going to be 3% of GDP. It's basically creating all the gains in the stock market." Color on durability, not a new chain. Zero net-new ai-infrastructure.

Sources added: 2026-08-24-autoresearch-shapiro-pa-eo-2026-05-pjm-siting-gate, 2026-08-21-podcast-all-in-podcast-dario-defends-himself-datacenter-panic-ai-doomer.

Update (2026-09-16) — SemiAnalysis puts a ceiling on delayed MW. Corroboration and a quantified bound, not a weaken.

From 2026-09-15-feed-semianalysis-everyone-says-datacenter-moratoriums (Maya Barkin, September 15; paid; free portion). The "moratoriums are killing the US buildout" narrative is the thing this page was built to track. SemiAnalysis's Datacenter Industry Model still forecasts +38 GW of US datacenter IT capacity in 2027 (22 GW under vertical construction). Parcel-level mapping of 400+ local instruments: of ~20 GW sitting inside a restricted boundary, only 1,525 MW is actually delayed (three projects: AWS Ohio, a Pennsylvania powered-land campus, Colorado). New York's order touches ~1.4 GW, of which ~0.8 GW is a meaningful delay. Combined genuine delay ~2.3 GW.

Texas ERCOT queue pause is 3–4 months of administrative delay for grid-tied baseload, offset by behind-the-meter acceleration (75 GW of firm BtM equipment orders, more bound for Texas than any other state). That is the same escape valve this page already recorded on 2026-08-10 — now with a vendor count behind it.

Read: this is a ceiling on delayed MW, not a falsifier of the siting-gate as a political variable. Four states have acted; 300+ local instruments exist; voters are net-negative on datacenters (46% unfavorable vs 29% favorable). The new information is that instrument-count is a poor proxy for MW delayed, which is exactly the tension this page already held between executive action succeeding, legislation failing, and local action proliferating. Conviction stays medium-high. Do not mint a tenth AI-infrastructure chain. Do not re-rate CAT / PWR / CEG. Zero net-new ai-infrastructure.

• btm-onsite-generation-to-bloom-fuelcell-gev-turbine — Texas as net BtM positive is independent corroboration of that exemption-as-trade.
• Brownsville de-annexation of 444 acres for an Elon project is one anecdote of the permitting-relocation leg already recorded 2026-08-11, not a pattern.

Update (2026-09-18) — Louisiana as a willing-host compact, not a fourth gate. Counter-weight, not a weaken.

From 2026-09-18-all-in-podcast-meta-s-dina-powell-mccormick-the-case-for-data (All-In, published 17 September). dina-powell-mccormick (Meta) calls Louisiana "our largest data center investment" and names a three-part compact: pay for your own electricity and work to drive costs down; treat water as a real farmland issue; surplus tax dollars going to teachers was a state law, "not our idea." "If a community doesn't want one of our data centers, they get to choose." A Richland Parish superintendent on the same tape: sales-tax collections at the six taxing entities spiked (normal 5–10% → about 60–70%, peaking at about 260%); teachers received a $50,000 check versus $10,000 the prior year, a net of about $45,000. Louisiana commerce secretary color on the same tape: the site was farmland; Meta chose a more expensive cooling system that "uses less water than was being used on that land when it was being farmed"; the state has ~32,000 miles of natural-gas pipelines.

Read: this is a willing-host / compact counter-example, not a fourth siting gate and not a falsifier of NY/TX/PA. Chamath on the same tape cites data centers polling "negative 80" nationally — that is the political weather this page already tracks. A successful Louisiana compact is the other half of the same map: gates in some states, bargains in others. It does not re-rate CAT / PWR / CEG. Conviction stays medium-high. Do not mint a twenty-eighth AI-infrastructure chain.

Update (2026-09-19) — explore-chain on the rotation question: gates still live; rotation still unmet

From 2026-09-19-autoresearch-siting-gate-rotation-graduation (attach on siting-gate-plus-stranded-capacity-to-demand-side-beneficiary-rotation). PA/NY/TX instruments remain in force (PA PUC curtailment-rulemaking 2026-09-10; NY CIF ~$1M/MW + GEIS draft scope; TX ERCOT pause + 2026-09-14 TWDB water-survey). CAT Creed (2026-09-10): BYOP “favors us actually.” Does not graduate the rotation hypothesis. Do not mint. Do not re-rate CAT / PWR / CEG. Conviction stays medium-high.