06-reference/research

data center opposition permitting tracker

2026-08-29·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
data-centerspermittingcompute-supplypower-cyclemonitoring

Does data-center NIMBY sentiment convert into a trackable compute-supply constraint?

The question

"How much of the 70-30 public opposition to local data centers is translating into actual project delays, permit denials, or moratoria — is there a trackable dataset (e.g. Data Center Watch, state legislative trackers, DOE grid-interconnection queue data) that would let RDCO monitor this as a concrete compute-supply headwind rather than a polling data point?"

Raised by the 2026-08-25 Innermost Loop digest, which framed the backlash as "thread lines" (self-inflicted compute queues) and noted RDCO tracks this nowhere. The question has an empirical half (does sentiment convert?) and an instrumental half (can we monitor it cheaply?). The instrumental half is the one that changes behavior.

What we already know (from the vault)

What the web says

Convergences and contradictions

Synthesis for RDCO

Verdict: worth instrumenting, but not as a compute-supply headwind on RDCO's own capability roadmap. Instrument it as the missing demand-side anchor on the power-cycle paper trade. The chain the digest proposed — L4→L5 is downstream of agent capability, which is downstream of compute supply, which is downstream of data-center siting — is directionally true and practically inert on RDCO's horizon. RDCO buys tokens at retail from Anthropic. A Michigan township moratorium does not move Ray's effective compute budget in 2026 or 2027; frontier-lab compute allocation, model pricing, and Anthropic's own capex do, and those are already tracked in the model-mix and hyperscaler-capex threads. Tracking siting fights as a proxy for "will Ray have enough compute" would be instrumenting four links away from the thing we actually care about, which is precisely the failure the [[2026-04-24-targeting-system]] filter is meant to catch.

Where the tracking genuinely pays is one layer down, on the power-cycle v1.1 paper trade, which is built on four supply-side anchors (ERCOT queue, hyperscaler PPA cadence, GEV order velocity, smart-money 13F) and prices in demand materializing on schedule. Stratechery named that hole in May and the vault agreed and then did nothing. This brief supplies the missing input and the source. The differentiated-siting-risk read from that note also survives contact with the new data: the 533 moratoria are concentrated in local jurisdictions with greenfield proposals, which is where VST and greenfield gas-plus-data-center campuses sit, and not where TLN's Susquehanna or CEG's existing nuclear fleet sit.

Concrete minimal setup — quarterly, not monthly, and roughly a day of work. Cadence is dictated by the sources: the two credible ones refresh quarterly and annually, and the underlying phenomenon moves on quarter timescales, so a monthly cron would mostly re-read unchanged files. (1) Moratorium Nation CSV — github.com/mjbommar/moratorium-data-2026, free, CC-BY-4.0, quarterly, documented 44-clause coding taxonomy, direct CSV. This is the policy-friction counter and the single best source in the set. (2) FracTracker resistance/victories tab — free, non-commercial, daily-updated, ArcGIS-backed, the only source with a project-outcome ("denied permits, withdrawn proposals, cancellations") count; note that no CSV or GeoJSON endpoint was documented on the page, so ingestion path is unconfirmed. (3) A denominator — data-center construction starts or put-in-place spending, so a rising moratorium count is always read against whether the aggregate build is actually slowing. Without the denominator this metric is pure advocacy noise. (4) Optional: LegiScan's free API (30,000 queries/month, all 50 states, weekly CSV snapshots) with an RDCO-authored data-center keyword classifier, if state-bill granularity is wanted; NCSL has no data-center database and is not automatable. (5) Data Center Watch: headline-scrape only, with a standing bias flag noting the 10a Labs parentage, the industry-facing risk-product framing, the absent funding disclosure, the blocked/delayed conflation, and the undisclosed dollar-valuation method. It should never be the sole citation for a number in a vault note again.

What would constitute a signal worth acting on. Not the poll — the polls have already moved from 43% to 75% opposition in twelve months and nothing downstream broke, so sentiment is a lagging decoration on this metric. Not the moratorium count alone either, because it is currently rising while construction sets records. The signal is the divergence: moratorium count still rising and data-center construction starts or put-in-place spending declining year-over-year for two consecutive quarters, or FracTracker's defeats-to-active-projects ratio moving materially off its current ~5%. Either would be the first evidence that opposition has crossed from redistributing the buildout to capping it, which is the power-cycle phase 3→4 trigger the thesis is watching for from the other direction. Absent that divergence, the correct posture is a quarterly one-line entry in the power-layer anchor directory and no further attention.

Why this is in the vault

It closes synthesis-queue item 8(c) ("when does demand-side NIMBY pressure convert into actual capacity-online deceleration?") with a dated answer of "not yet as of August 2026," and it supplies the demand-side anchor that [[2026-05-18-power-cycle-v1.1]] is explicitly missing and that [[2026-05-18-stratechery-data-center-discontent]] recommended building three months ago. It also corrects a live citation error: the $156B "blocked or stalled" figure repeated in [[2026-05-12-diamandis-innermost-loop-ai-infrastructure-thesis]] is a Data Center Watch blocked-plus-delayed aggregate with undisclosed valuation methodology, not a measure of lost capacity.

Open follow-ups

Related

Sources

Vault

Polling

Trackers

Grid and construction data