Three Hundred Moratoriums, 2.3 Gigawatts: The Real Math on America's 'Blocked' Data Center Boom

If you have followed AI infrastructure news this year, you have almost certainly heard the story: American towns are rising up against data centers, state capitols are slamming the brakes, and the great GPU buildout is being legislated into gridlock. The numbers sound damning — New York halted environmental permits. Texas paused its major-load interconnection track. Michigan alone has 45 local moratoriums on the books. More than 300 towns, cities, and counties have voted to pause data center construction in the past eighteen months.

A new deep-dive from SemiAnalysis, published September 15, set out to test that narrative parcel by parcel — and found it does not survive contact with the project-level data. After mapping every active restriction in a new moratorium database covering more than 400 local instruments across 17 states, and cross-referencing each one against a facility model that tracks 6,000+ data centers through property records, permits, power-usage data, FOIA requests, and real-time satellite imagery, the firm reached a blunt conclusion: roughly 2.3 GW of planned capacity is genuinely delayed by the policy interventions that dominate the headlines. Against a forecast of +38 GW of data center IT capacity delivered in the US in 2027 — more than double 2026’s delivery — that is a rounding error, not a freeze.

Exposure Is Not Delay

The heart of the analysis is a distinction most coverage misses: capacity that sits inside a restricted boundary is not capacity that slips because of it. Of the roughly 20 GW of pipeline located within a jurisdiction that has enacted some form of restriction, only 1,525 MW — about 7.6 percent — is actually delayed. Three projects account for essentially all of it: an AWS campus in Ohio, a powered-land developer campus in Pennsylvania, and a site in Colorado.

Why does the conversion rate from “moratorium passed” to “megawatts delayed” run below ten percent? Because for a moratorium to genuinely move a project’s delivery date, a long chain of conditions must all hold simultaneously. The restriction must actually reach the parcel — county moratoriums typically stop at city lines, leaving incorporated municipalities untouched. It must be enacted and still in force, not merely proposed, lapsed, or replaced. It must overlap the project’s timeline: a three-to-six-month pause means nothing to a 2028+ project that will not need the affected approval for years, and nothing at all to a project whose approvals were already granted, since these instruments freeze new applications rather than revoke existing entitlements. Many ordinances also carve out projects below a size threshold or apply only to crypto mining. And developers can sometimes simply redesign around the rule — if a moratorium bites at the meter, behind-the-meter generation offers an escape hatch.

Miss any one of those conditions, and the moratorium is irrelevant to the schedule.

The Loophole Economy

The report is salted with examples of how fluid these boundaries are in practice. In one case, a data center planned inside Brownsville, Texas — where a moratorium was brewing — was “deannexed” from the city days before approval, neatly moving the project out of the restricted jurisdiction. (The project’s backer: Elon Musk.)

The Pennsylvania case study is the counterexample that proves how rare a binding moratorium is. NorthPoint’s campus near Scranton sits in a township that adopted a unanimous moratorium resolution in June 2026, specifically covering data centers with no size threshold, before NorthPoint had filed for its special exception — with first buildings scheduled for Q4 2026. Every link in the chain holds: the parcel is reached, the use is covered, there is no federal pathway, no vested rights, no redesign that dodges a use-based restriction. The entitlement process now slides into 2027, and the developer is reportedly still pursuing the project, including a proposed $165 million community benefits package. That is what a moratorium that actually binds looks like — and it is one of only a handful in the country.

State Actions, Same Story

The state-level actions that generated the loudest headlines dissolve similarly under scrutiny. New York’s executive order touches roughly 1.4 GW of pipeline, but SemiAnalysis estimates only about 0.8 GW faces meaningful delay attributable to it. Texas’s pause adds perhaps three or four months of administrative delay for base-load projects in the ERCOT interconnection queue — and is offset by an acceleration of behind-the-meter demand in the same state. For projects that never needed the grid connection in the first place, the pause is a non-event.

Indeed, the broader thesis of SemiAnalysis’s infrastructure coverage this year has been that friction at the grid edge is not stopping the buildout; it is reshaping it. The firm counts 75 GW of firm behind-the-meter power equipment orders, more of them bound for Texas than any other state, and expects BtM generation to power well over half of new US data center capacity from 2028 onward.

Why the Myth Persists

Counting moratoriums is easy — counting megawatts is hard. The number of restrictions is publicly visible and politically satisfying; the parcel-level analysis required to translate a ordinance into slipped delivery dates is not. Election-year incentives compound the distortion: with November 2026 midterms approaching, a temporary pause is a low-cost way for officials to signal responsiveness to community anger without permanently shutting down development. Both major New York gubernatorial candidates have endorsed some form of pause, and in Ohio, Republican gubernatorial nominee Vivek Ramaswamy said in August that no new data center should be approved unless it eliminates electricity bills for nearby residents, pays full property taxes, and meets strict water and farmland protections.

The underlying public sentiment is real, even if its infrastructure impact so far is not. SemiAnalysis’s August polling found Americans net-positive on AI broadly but significantly more negative on data centers specifically, with opponents most frequently citing energy use and higher utility bills — a gap the firm notes is already shaping midterm politics. The local pressure is concentrated where growth has been fastest: Michigan (45 enacted moratoriums, 7 proposed, largely following a December 2024 state incentive law), Ohio (40 enacted, 83% adopted in 2026 alone, trailing the mid-2025 PJM capacity-price shock that pushed average residential bills up 25.7% year over year), North Carolina, and Georgia.

The Bottom Line

None of this means moratoriums are harmless forever. The authors are explicit that the instruments could delay the buildout at scale if they become materially more widespread and more restrictive — and their project-level tracker is designed to catch that inflection the moment it happens. Construction schedules, not ordinance counts, are the signal to watch. But as of September 2026, the arithmetic is unambiguous: 300-plus restrictions, 20 GW of nominal exposure, and barely 2.3 GW of actual delay — while 22 GW of the 2027 forecast is already under construction and the remainder is, in majority, financed and preparing vertical builds. The American data center boom is politically noisy and operationally relentless. Those waiting for NIMBY politics to break the AI infrastructure story are, on the current evidence, waiting for a number that the parcel data refuses to supply.