Diagram showing capital, chips, and power ready to deploy but blocked behind five permission gates: zoning, tariff, queue, water, and community.

Permission Is the New Compute Bottleneck

by

For three years the binding constraint on artificial intelligence (AI) compute was assumed to be capital, then chips, then power. As of September 2026 the evidence points somewhere else: the pace of the buildout is set by permission. Moratoriums, large-load tariffs, interconnection queues, rate cases, and water and land conflicts now decide which projects advance, in what order, and how quickly, more than balance sheets or silicon supply do.

This brief treats permission as an infrastructure layer with its own throughput and failure modes. It sets out what changed in the past 90 days, why the constraint has moved, what the prevailing consensus still misses, and which public decisions will resolve the question.

What changed

On June 18, 2026, the Federal Energy Regulatory Commission (FERC) issued tailored show cause orders under Section 206 of the Federal Power Act to all six regional transmission organizations and independent system operators (RTOs/ISOs) it regulates: PJM Interconnection, MISO, the Southwest Power Pool (SPP), the California Independent System Operator (CAISO), ISO New England (ISO-NE), and the New York Independent System Operator (NYISO). The orders (Dockets EL26-67-000 through EL26-72-000) direct each operator and its transmission owners to justify or reform the tariffs that govern large loads, including data centers, across five categories of reform: application and study processes; cost transparency and cost-shifting protection; co-location and behind-the-meter generation; services for flexible large loads; and study processes for generation serving electrically proximate load. The orders gave the operators 60 days to respond and invited abeyance requests within 45 days.

The abeyance became the story. Reporting indicates all six operators and their transmission owners sought roughly 90 additional days, and FERC granted ISO-NE’s joint motion on August 14, 2026, moving its show cause response and Section 205 tariff filing to November 16, 2026, with answers due December 16. CAISO and SPP have published the same November 16 target, and PJM moved to hold its own proceeding in abeyance while it develops a Section 205 filing through stakeholder engagement. A deadline FERC set at 60 days has become roughly five months, and the operators are writing tariffs rather than litigating the show cause orders.

The same pattern shows up in capacity. On July 14, 2026, PJM reported that its 2028/2029 Base Residual Auction cleared 138,318 megawatts (MW) of unforced capacity, about 6,831 MW short of its reliability requirement, the second consecutive auction to fall short, at the FERC-approved price cap of $325 per megawatt-day, for a total cleared value of $16.4 billion. PJM’s own report notes that without the price collar the auction would have cleared at $554.72 per megawatt-day and produced about $29.7 billion. The shortfall does not mean the lights go out in the 2028/2029 delivery year; it means the system runs with slimmer reserves and more risk.

In Texas, permission became an explicit gate. On August 3, 2026, Governor Greg Abbott ordered the Public Utility Commission of Texas (PUCT) and the Electric Reliability Council of Texas (ERCOT) to audit every data center advancing through the interconnection queue before any new project is approved to connect. ERCOT had been tracking about 474.7 gigawatts (GW) of large-load requests as of June 2026, more than five times the grid’s record peak demand of 91,089 MW set on July 22, 2026; roughly 90 percent of the queue is data centers. ERCOT paused its Batch Zero study, sent survey requests to hundreds of projects, and plans to present its eligibility verification and community impact reports on December 10, 2026.

State and local permission layers kept thickening. The Edison Electric Institute (EEI) counted 23 states with at least one approved large-load tariff and seven more pending as of May 2026; by July it counted 24 approved and six pending. The Smart Electric Power Alliance (SEPA) counted 77 approved or proposed large-load tariffs across 36 states and 60 utilities as of March 31, 2026, with 29 approved in 2025 alone against 14 in all of 2018 through 2024. The Urban Institute counts 313 data center moratoriums enacted as of September 2026, affecting 44 states and more than 200 localities, though tracker counts vary with scope and date. New York became the first state to impose a statewide pause, through an executive order covering facilities of 50 MW or more.

The politics went national at the same time. On September 16, 2026, the U.S. House of Representatives passed H.R. 9340, the Ratepayer Protection Act, by a vote of 417 to 3. The bill would amend the Public Utility Regulatory Policies Act of 1978 (PURPA) to require state regulators to consider a standard making large-load customers above 100 MW pay the full incremental cost of the generation, transmission, and distribution upgrades built to serve them. Senate passage, which would have required unanimous consent, was blocked within a day according to September 18 reporting, and both the House bill and a competing Senate measure remain stalled. Data centers are now a campaign issue in both parties ahead of the November 3 midterms.

Site selection data points the same way. The Site Selectors Guild’s 2026 survey of its members found utility and infrastructure capacity to be the top factor affecting location decisions, cited by 76 percent of respondents, and the leading reason sites are eliminated, cited by 61 percent. Bloom Energy’s mid-2026 survey of 156 data center decision-makers still ranked power availability as the dominant site-selection factor, with 51 percent placing it in their top three considerations.

Why this matters now

Juncture’s energy and infrastructure coverage has tracked a single question through 2026: who pays for the grid that AI needs, and who gets to decide. The ratepayer-protection wave has answered the first half. The second half is the constraint this brief names: permission. A project can have capital, chips, land, and a power purchase agreement and still not be allowed to connect, because a county board has paused permits, a state commission has not approved a tariff, a grid operator has not finished a study, or a governor has ordered an audit.

Capital and chips are globally mobile and increasingly abundant. Permission is local, slow, and adversarial. That asymmetry is why the buildout’s pace is now set by the slowest approval gate rather than by the fastest source of finance. A moratorium cannot be refinanced. A queue position cannot be purchased from a supplier. A rate case cannot be accelerated with a larger check.

The timing matters because the permission layer has just become measurable. For the first time there are docket numbers, filing dates, tariff counts, moratorium counts, and audit reports that together describe how much capacity is allowed through, and when. A constraint that can be measured can be planned around. A constraint that is still treated as background noise will keep surprising project timelines.

The institutional constraint

Permission behaves like infrastructure because it has throughput. Every gate – a zoning approval, a tariff proceeding, an interconnection study, a rate case, a water permit – processes a limited amount of capacity per unit of time. The FERC show cause process is the clearest example: six orders issued June 18, a 60-day response window, then abeyances that moved the substantive filings to November 16. That is fast by the standards of utility regulation, and it still leaves five months between a finding that a tariff is unjust and the receipt of a replacement.

The state layer is where cost discipline is now being set. Large-load tariffs generally require developers to fund the grid upgrades their projects need, sign long-term demand contracts, and post collateral, which transfers real project risk back to the developer. Virginia’s GS-5 rate class, Oregon’s Schedule 96, Pennsylvania’s model tariff, and Ohio’s take-or-pay requirements are different instruments converging on the same principle: cost causation. A tariff and the rate case that produces it are permission instruments, because the economics they set determine whether a project can be financed at all.

The local layer is where permission can be revoked outright. Moratoriums are time-limited, but they stop the clock on permit applications, and in some jurisdictions they have hardened into permanent zoning restrictions. The Urban Institute’s count of 313 enacted moratoriums across 44 states is a snapshot of a veto point that did not exist at this scale two years ago. Community benefit agreements (CBAs) are the negotiated alternative. Cedar Rapids, Iowa, for example, secured a 20-year, 70 percent tax exemption for Google alongside $400,000 per year per data center for 15 years, capped at $6 million per data center and $36 million in total, plus a comparable structure for the QTS campus. New York’s Community Investment Framework and a published NAACP template are attempts to standardize what communities can ask for.

The constraint is not regulation in the abstract. It is that no single authority controls the stack. FERC can order tariff reform but cannot lift a county moratorium. A governor can order an audit but cannot accelerate a stakeholder process. A utility commission can approve a tariff but cannot guarantee a water supply. The buildout now depends on aligning a dozen independent permission holders, each with its own calendar and its own grievance.

What the consensus misses

The consensus still frames AI compute as a capital problem or a chip problem, and increasingly as a power problem. Each framing assumes that once the money, the silicon, or the megawatts are arranged, construction follows. The 2026 record shows otherwise: the gating item is authorization, and authorization does not scale with capital or generation. The national generation queue held 2,061 GW of capacity actively seeking interconnection at the end of 2025, according to Lawrence Berkeley National Laboratory (LBNL); only 13 percent of the capacity that applied from 2000 through 2020 had reached commercial operation by the end of 2025, and the median time from request to commercial operation exceeded five years. Generator queues and load queues are separate systems, but they share the same study teams, the same transmission limits, and the same permission bottleneck.

The second miss is treating opposition as a communications problem. The evidence describes something more durable. Opposition is bipartisan, it tracks the buildout rather than preceding it, and it is strongest in states that welcomed development first. Data Center Watch counted 75 projects worth $130 billion blocked or delayed in the first quarter of 2026 alone, roughly the number affected in all of 2025. A community benefit agreement is a contract, not a campaign, and it is negotiated at the same table as zoning.

The third miss is expecting federal action to clear the logjam by itself. FERC’s orders are the most aggressive federal large-load intervention to date, and their first effect was a five-month extension. The House vote of 417 to 3 is a strong signal of consensus; the bill stalled in the Senate within a day, and even if enacted it would ask states to consider a standard on a two-year timeline. Federal policy can set direction. It cannot supply local consent.

What the consensus also misses is the cost of the queue itself. ERCOT’s roughly 475 GW of requests are not a pipeline; they are an option market. ERCOT has approved 362 projects representing 191.8 GW for its Batch Zero study and excluded 373 projects requesting more than 300 GW. A permission system that inflates its own backlog makes every serious project harder to see and slower to build.

The Permission-Layer Framework

This is a Juncture working framework, drawn from the constraint described above. It organizes the evidence in this brief; it is not a validated predictive model.

Diagnostic question. When a compute project is proposed, which permissions must be granted, how long does each grant take, and what can take it away?

Layer element Gatekeeper What it sets Failure mode
Local land use and zoning City and county boards Whether a site may host a data center at all Denial or permanent restriction
Moratorium or pause Local councils; state executive or legislature Whether new applications can be filed or permits issued Delay measured in months or years; project relocation
Large-load tariff and rate case State utility commission Who pays for grid upgrades; contract terms, collateral, and exit fees Project economics fail; financing withheld
Interconnection study and queue RTO/ISO and transmission owners When and how much capacity may connect Queue delay, withdrawal, or disqualification
Water, land, and community agreement Local government and community Access to water and land; political consent Permit denial; litigation; negotiated CBA terms

Throughput. Permission throughput is the amount of authorized capacity a gate processes per unit of time. A buildout moves at the pace of its slowest gate, not its average. The measure to watch is not how many projects are announced but how many megawatts clear each gate per month.

Decision sequence. Map every gate first. Then measure the slowest one and treat it as the project schedule. Then price the option created by delay, including queue positions, land control, and cancellation costs. Then negotiate the community agreement before, not after, the incentive package. Then sequence filings so that no gate depends on a decision that has not yet been made.

Failure modes. Speculative queue positions inflate the backlog and slow serious projects. Tariffs that push all upgrade and exit risk onto developers can make otherwise viable projects unfinanceable. Moratoriums end the application clock rather than the project, pushing capacity to other jurisdictions. Rate case losses reprice an entire portfolio. Water and land conflicts convert technical projects into political ones.

Applied to this brief (Juncture’s reading). ERCOT illustrates a queue gate that has become the primary constraint. PJM illustrates the tariff and capacity gate. New York illustrates the moratorium gate. Cedar Rapids illustrates the community agreement gate. The same project can face all four.

Where the framework does not apply. Projects that serve load behind the meter or self-generate can bypass some gates, though they still face air, water, and zoning authority. The framework is about grid-connected, utility-scale development.

Transparency caveat. This is a working framework under development under Juncture’s methodology. It organizes the evidence in this brief and requires historical validation to demonstrate predictive power.

Resolvable outcomes

This brief registers qualitative resolution conditions, not a probabilistic forecast. No probabilities are assigned because the record does not support base rates or a single resolution date. Each condition resolves on the public record.

Outcome question Resolution condition Status as of publication
Do the six RTO/ISO show cause proceedings produce replacement tariffs? FERC accepts, rejects, or modifies Section 205 filings, beginning with the November 16, 2026 target date. ISO-NE abeyance granted August 14; responses and filings targeted for November 16; other operators in abeyance or development.
Does Texas convert its audit pause into durable permission rules? PUCT and ERCOT release the Batch Zero eligibility verification and community impact reports and resume, condition, or deny queue approvals. Audit ordered August 3; Batch Zero paused; reports scheduled for December 10, 2026, with a PUCT presentation December 17.
Does the state tariff wave keep expanding? EEI and SEPA counts move beyond 24 approved and 77 tracked tariffs. 24 states approved and 6 pending as of July 2026 (EEI); 77 approved or proposed across 36 states as of March 31, 2026 (SEPA).
Do moratoriums reduce permitted capacity? Tracker counts stabilize, or project-level data show permitted capacity falling in affected jurisdictions. 313 enacted moratoriums across 44 states as of September 2026 (Urban Institute); counts differ by tracker and scope.
Does Congress set a federal large-load standard? The Senate passes H.R. 9340 or a substitute, or the bills remain stalled through the 119th Congress. House passed H.R. 9340, 417 to 3, on September 16, 2026; Senate passage blocked September 17; no Senate action scheduled.
Do community benefit agreements become standard terms? Published agreements spread beyond early adopters and appear as conditions in zoning or tariff proceedings. Cedar Rapids agreements operational; New York framework published; NAACP template released; Lancaster agreement public.

No probabilities are assigned. Juncture registers probabilities only when the underlying record supports base rates and a defined resolution date; this brief carries neither.

Forecast record

Field Entry
Forecast status Not registered
Reason The record supports scheduled public events and qualitative resolution conditions, not a probabilistic forecast; no base rates or single resolution date are available for the permission-layer question.
Resolution authority Juncture Policy editorial desk, on the public record: FERC dockets and orders, PJM and ERCOT filings and reports, state commission and tracker data, and the reporting cited below.
Update history v1 2026-09-18; qualitative conditions only, no forecast registered.

What to watch

  • November 16, 2026: substantive Section 205 filings from ISO-NE, CAISO, and SPP, and any parallel filings from PJM, MISO, and NYISO. December 16: answers to ISO-NE’s filing.
  • PJM’s Reliability Backstop Procurement, proposed for September 30, 2026, and how FERC treats it; whether it closes any of the 6,831 MW shortfall.
  • December 10, 2026: ERCOT’s Batch Zero eligibility verification report and community impact review, and the December 17 PUCT presentation. The question is whether the pause becomes stricter conditions or denials.
  • The next EEI and SEPA tariff counts, and whether any of the states still without a large-load tariff adopt one. Only 13 states lacked any large-load tariff as of July 2026, according to Utility Dive’s reading of the SEPA database.
  • Moratorium tracker divergence and the New York executive order’s implementation, including the Community Investment Framework and local advisory committees.
  • Senate action on H.R. 9340 or the GRID Savings Act after the November 3 midterms, and whether data center politics change the composition of state legislatures.
  • Whether CBAs appear as conditions in permits, tariffs, or rate cases rather than as voluntary side agreements.

Bottom line

The binding constraint on AI compute is no longer capital or chips, and increasingly it is not even power. It is permission. Moratoriums, tariffs, queues, rate cases, and water and land conflicts now set the pace of the buildout, and the 2026 record shows that layer processing capacity in months to years rather than quarters. Investors and governments that treat permission as an infrastructure layer, with measured throughput, mapped gates, and negotiated terms, will build; those that treat it as noise will keep announcing projects that do not connect.

Evidence and sources

Primary and institutional sources

Figure note on moratorium counts. Published counts differ by methodology and date. The Urban Institute’s 313 figure includes all enacted instruments as of September 2026. Other trackers report roughly 300 or more adopted moratoriums, 321 instruments tracked with 261 in force, or 40 states with at least one active local restriction as of late August 2026. This brief uses the Urban Institute count and labels it as one count among several.

Figure note on tariff counts. EEI counted 23 states approved and seven pending in May 2026, then 24 approved and six pending in July 2026. SEPA’s DELTA database counted 77 approved or proposed tariffs across 36 states and 60 utilities as of March 31, 2026. The two counts measure different things: states with at least one tariff, versus individual tariff instruments.

Figure note on the ERCOT queue. ERCOT reported approximately 474.7 GW of large-load interconnection requests as of June 2026, commonly shortened to about 475 GW and sometimes reported as 476 GW. Of the total, 420.8 GW, or 90.2 percent, was identified as data centers.

Figure note on the PJM value. The $16.4 billion figure is the cleared supply multiplied by the clearing price. It is not the total cost to load, because self-supplied and bilaterally contracted capacity is not exposed to the auction clearing price.

Figure note on Cedar Rapids. This brief uses the primary development agreement terms of $400,000 per data center per year for Google. Later local reporting refers to $350,000 per building; the primary agreement and the city’s economic development page state $400,000.

Load-bearing claim map

Claim in this brief Source Confidence
On June 18, 2026, FERC issued Section 206 show cause orders to six RTOs/ISOs in Dockets EL26-67-000 through EL26-72-000, directing them to justify or reform large-load tariffs across five reform categories within 60 days, with 30-day resource adequacy reports and up to 90 days of abeyance available on request. FERC orders and fact sheet High
All six operators sought roughly 90 additional days; FERC granted ISO-NE’s joint motion on August 14, 2026, moving its response and Section 205 filing to November 16, 2026, with answers due December 16; CAISO and SPP published the same target; PJM moved for abeyance and is developing a Section 205 filing. FERC/ISO-NE abeyance order; PJM motion; CAISO motion; trade reporting High for ISO-NE, PJM, CAISO; Medium for the all-six characterization
PJM’s 2028/2029 Base Residual Auction cleared 138,317.8 MW at a $325 per megawatt-day cap, 6,831.3 MW short of the reliability requirement, with a cleared value of $16.4 billion and an uncapped counterfactual of $554.72 per megawatt-day and about $29.7 billion. PJM auction report and news release High
Texas ordered an audit of data centers in the ERCOT interconnection queue on August 3, 2026, pausing approvals; ERCOT was tracking about 474.7 GW of large-load requests as of June 2026, 90.2 percent of it data centers, against a record peak of 91,089 MW; 362 projects representing 191.8 GW were approved for Batch Zero and 373 projects requesting more than 300 GW were excluded; reports are due December 10. Texas Tribune; KERA; POWER Magazine; ERCOT PUCT presentation High for the audit and queue figures; Medium for the Batch Zero project counts
EEI counted 23 states with an approved large-load tariff and seven pending as of May 2026, and 24 approved with six pending as of July 2026; SEPA counted 77 approved or proposed tariffs across 36 states and 60 utilities as of March 31, 2026; only 13 states lacked any large-load tariff as of July 2026. EEI; SEPA via Utility Dive; Environment+Energy Leader High for EEI and SEPA counts; Medium for the 13-state figure
The Urban Institute counts 313 data center moratoriums enacted as of September 2026, affecting 44 states and more than 200 localities; Brookings reported at least 15 states weighing pauses and at least 100 localities adopting moratoriums as of July 2026; New York imposed the first statewide pause on facilities of 50 MW or more. Urban Institute; Brookings; Governing High for the named counts; Medium as a precise national total, given tracker divergence
Data Center Watch counted 75 projects worth $130 billion blocked or delayed in the first quarter of 2026, roughly as many as in all of 2025. Brookings, citing Data Center Watch Medium
LBNL found 2,061 GW actively seeking generation interconnection at the end of 2025, a 13 percent completion rate for requests submitted from 2000 through 2020, and a median request-to-commercial-operation timeline exceeding five years; the dataset excludes load interconnection. LBNL, Queued Up 2026 Edition High
Site Selectors Guild members ranked utility and infrastructure capacity as the top factor affecting location decisions (76 percent) and the leading site-elimination factor (61 percent) in 2026; Bloom Energy’s mid-2026 survey ranked power availability as the dominant site-selection factor, with 51 percent placing it in the top three. Site Selectors Guild; Bloom Energy via Latitude Media High
Cedar Rapids secured a 20-year, 70 percent tax exemption for Google plus community betterment payments of $400,000 per data center per year for 15 years, capped at $6 million per data center and $36 million total, with a comparable QTS structure. Cedar Rapids development agreement and economic development page High
The House passed H.R. 9340, the Ratepayer Protection Act, 417 to 3, on September 16, 2026; the bill covers large loads above 100 MW and would require states to consider a full-incremental-cost standard under PURPA Section 111(d); Senate passage was blocked within a day and no Senate action is scheduled. House Energy and Commerce; CBS News; Utility Dive; Roll Call High for the House vote and bill mechanics; Medium for the Senate block, pending fuller primary confirmation
The Reuters/Ipsos findings that 64 percent of respondents oppose rapid data center construction, 77 percent worry about electricity rates, and 14 percent would live near a data center; and that data centers are a campaign issue in both parties ahead of the November 3 midterms. Brookings, citing Reuters/Ipsos Medium