The growth of AI-driven electricity demand has so far been discussed as a supply shortage: not enough generation, not enough transmission. But what the rule changes that took shape in the United States and Japan between June and July 2026 actually address is not the volume of electricity. It is which customers get scarce connection capacity, in what order, and who bears the cost.

The conclusion first. For large electricity consumers, the variable that decides siting is shifting from "is there power here" to "can we join that queue, when does our turn come, and can we carry the cost and risk of network upgrades."

And data centers are not the only ones standing in that queue. New electric arc furnaces, hydrogen direct reduction plants, chemical facilities and battery factories — all of them power-hungry heavy industry — are in the same line. Designing the connection rules for AI infrastructure is, directly, designing the siting conditions for manufacturing.

This is a hypothesis held by Itajima & Co., not an established fact. What follows separates what we could verify, the interpretation we draw from it, and the hypothesis itself.

What is happening — the verifiable record

United States: the regulator put the interconnection tariff itself on the table

On June 18, 2026, the Federal Energy Regulatory Commission (FERC) issued show cause orders under Section 206 of the Federal Power Act to all six Regional Transmission Organizations (RTOs) and Independent System Operators (ISOs) under its jurisdiction, and to their transmission owners. The dockets map to regions as follows: EL26-67 to PJM, EL26-68 to SPP, EL26-69 to NYISO, EL26-70 to MISO, EL26-71 to CAISO, and EL26-72 to ISO-NE.

These are not final rules. They open a proceeding in which the parties must either justify or reform their tariffs, based on FERC's preliminary finding that existing open access transmission tariffs appear unjust and unreasonable because they do not account for how data centers and other large loads connect to and use the transmission system. This distinction matters for everything that follows, so we place it first.

FERC directed five categories of tariff provisions to be justified or reformed.

  1. Application and study procedures for large load transmission service, together with ongoing operational requirements
  2. Transparency on network upgrade costs and protection against cost shifting, including a pro forma cost recovery agreement
  3. Rates, terms and conditions applying to large loads served by dedicated, co-located generation
  4. New transmission services for flexible large loads willing and able to adjust their use of the system
  5. Rules for generation owners seeking transmission service to serve nearby or adjacent large loads

FERC's proposed definition of a "large load" is a commercial or industrial customer at a single site, with a peak load on the order of 50 MW connecting to the transmission grid, that is not part of a co-location arrangement.

Two qualifications belong here. First, this is FERC's proposed definition, not a settled scope. RTOs and ISOs may propose their own definitions suited to regional conditions — as the Rocky Mountain Institute puts it, "RTOs/ISOs may use FERC's suggested definition of large load in their updated tariffs or develop their own definitions that satisfy FERC's concerns" — and SPP's own definition of a "high impact large load," covering new facilities above 10 MW interconnecting at or below 69 kV and above 50 MW interconnecting above 69 kV, has already been accepted by FERC.

Second, the thresholds are reported inconsistently. Orrick gives "at least 50 MW" and "at least 69 kV"; the Rocky Mountain Institute gives "over 50 MW" and "above 69 kV." Because we could not check the text of the orders themselves, this essay treats neither figure as settled.

On substance, the most important element is that FERC called for a pro forma cost recovery agreement among the RTO/ISO, the transmission owner and the customer, sized according to the amount of transmission service (in MW) requested to serve the large load. Its purpose is to place the cost and risk of network upgrades on the large load itself, and to keep them from being shifted onto existing customers.

FERC did not, however, prescribe specific formulas — minimum contribution ratios, levels of financial security. What it set out is a principle for where cost and risk belong, not an amount.

The procedural deadlines run 30 days for an informational report on resource adequacy, 45 days for a request to hold the proceeding in abeyance, and 60 days — August 17, 2026 — to justify the tariff or file revisions.

United States: a state began taxing electricity, self-generated power included

Virginia imposed a consumption tax of $0.011 per kWh on electricity used by data centers, effective July 1, 2026. The legislature passed the budget on June 22, 2026 and Governor Abigail Spanberger signed it on June 30, the day before it took effect. The authority is H.B. 30, Item 3-5.24, of the 2026 Special Session I.

Three features of the design are worth noting.

  • The tax is not limited to utility-supplied power. It covers self-generated electricity as well, with no carve-out for on-site solar or wind.
  • It applies to data centers with power and cooling capacity of 1 MW or more, excluding facilities whose primary purpose is internet access or communications services.
  • Revenue above $600 million in a fiscal year is refunded pro rata according to each operator's share of payments, from July 1, 2027. Absent an extension the tax sunsets on June 30, 2028 — a two-year measure.

Japan: an efficiency standard, and a demand outlook built on connection applications

In Japan, an efficiency standard for the data center industry was introduced under the Energy Conservation and Non-Fossil Transition Act. A document from the Energy Conservation Division of the Agency for Natural Resources and Energy, "Institutional responses toward data center efficiency" (June 9, 2025), sets out as a proposed rule that data centers newly built from fiscal 2029 onward must reach a PUE of 1.3 or lower two years after entering service. The existing benchmark program separately targets an operator-average PUE of 1.4 or lower by fiscal 2030. PUE (Power Usage Effectiveness) is total facility power divided by the power drawn by ICT equipment such as servers; the closer to 1.0, the more efficient the supporting plant.

These are figures from the proposed rule as of June 9, 2025. We could not obtain the text of the guideline adopted on April 10, 2026, though secondary sources indicate the finalized version retains PUE 1.3, the fiscal 2029 threshold, and the two-year grace period. This essay treats them as proposal-stage values.

Now the demand side. The Organization for Cross-regional Coordination of Transmission Operators (OCCTO) publishes an annual "Nationwide and area-by-area demand outlook" consolidating the projections of the general transmission and distribution utilities, and within it accounts separately and by name for new and expanded data centers and semiconductor plants. These itemized figures are not total demand; they are the increment added to the base year for new and expanded facilities that regression-based demand projection does not capture.

In the latest edition (for the fiscal 2026 supply plan, published January 21, 2026), the itemized total for data centers and semiconductor plants rises from 830 MW in fiscal 2026 to 7,620 MW in fiscal 2035, and energy demand from 6.7 TWh to 56.8 TWh. At the fiscal 2034 point, data centers alone account for 6,150 MW and 46.0 TWh.

The difference between editions is worth pausing on. In the previous edition (published January 22, 2025), the itemized figure for data centers alone ran from 470 MW in fiscal 2025 to 6,160 MW in fiscal 2034, and energy from 3.0 TWh to 44.0 TWh, with peak demand including semiconductor plants at 7,150 MW in fiscal 2034. In the latest edition OCCTO states that "for data centers, reflecting operators' revisions to their construction plans and the observed performance of existing data centers, the trajectory runs below the previous projection through fiscal 2033," while exceeding it from fiscal 2034 onward. The near years were revised down, and the growth pushed later. The connection queue is not a line that only grows.

What matters most about this outlook, for our purposes, is not the magnitude of the increase but how it is built. OCCTO judges the likelihood of new construction by "ascertaining the status of grid applications in line with demand facility grid access operations," and states that "projects that have advanced to the stage of concluding and invoicing a construction cost contribution agreement are assessed as highly likely and are always itemized," while earlier-stage projects are itemized when concrete engineering study, subsidy awards, press releases and the like indicate high likelihood. The itemized values themselves are calculated "on the basis of the contracted capacity applied for with the general transmission and distribution utility."

In other words, in Japan the queue that industry has formed at the grid is itself the foundation of the national demand outlook. But joining the queue is not what gets you counted. Likelihood is filtered by whether a project has advanced as far as a construction cost contribution agreement. Standing in line and moving up the line are treated as different things, statistically as well.

The conventional view, and what it misses

The dominant account runs like this. Generative AI has driven a surge in data center electricity consumption; generation capacity and grid buildout cannot keep up; therefore power is short. The remedy is to add generation, build lines, and raise efficiency.

That account is not wrong. The Belfer Center at Harvard Kennedy School (February 2026) frames AI data centers as a watershed moment for the US electric grid, and official projections in several countries do show demand rising faster than expected.

But when you line up what institutions actually did between June and July 2026, the contested question turns out not to be volume.

FERC's orders do not order power plants built. They are orders about how connection applications are studied, who pays for upgrades, and how co-located and behind-the-meter generation is treated — procedure and cost allocation. Virginia's tax is not a supply measure either; it draws revenue from large loads and is designed to close off avoidance through self-generation. Japan's measures impose an efficiency floor while building the demand outlook out of grid application status and how far each application has progressed.

What is scarce is not only electricity. It is the institutional capacity to decide who gets limited connection, and on what terms. And this queue does not distinguish by end use. An application from a data center and an application from an arc furnace or a battery plant stand in the same line, on the same grid.

The electricity constraint of the AI era appears not as a shortage of generating capacity but as a shortage of institutional capacity to allocate connection.

Spillover into heavy industry and industrial structure

From here on, this is interpretation drawn from the record above.

First, once cost and risk are clearly assigned, connection changes from "something you receive by applying and waiting" to "something taken by whoever can carry it." If priority follows the ability to absorb upgrade costs and schedule risk, then capital strength and creditworthiness are reflected in queue position. That is a statement about electricity and, at the same time, about the terms of entry into an industry.

Second, exit via self-generation or co-located supply stops being a complete escape the moment the rules bring self-generation inside the tax and the tariff. Virginia taxes self-generated power; FERC placed co-location and behind-the-meter generation among the areas to be reformed. The premise that owning your own generation frees you from the grid's institutions does not hold in these two cases. That said, this is an inference from two observations, and other states and countries may not adopt the same design.

Third, the binding constraint on heavy industrial capital plans changes. For electric arc furnaces, hydrogen direct reduction, chemicals and batteries, if the available connection date falls later than construction completion, the gap becomes idle capital for its full duration.

Not productive capacity, but allocative capacity

The capability of an industrial civilization has often been measured as productive capacity: crude steel output, installed generating capacity, wafer starts. What the rule changes above point to is a different capability — the institutional capacity to decide in what order, at whose cost, and to whom scarce connection is allocated.

Power plants can be built. Lines can be strung. But the procedure that decides who gets connected first moves at the speed of institutions, not technology. That FERC set a 60-day deadline is itself evidence that the speed of this procedure has been recognized as the problem.

If we call this the operating capability of a civilization, its level is observable — not as an abstract ideal but as the length of the connection queue and the way that queue is made to move. Keeping that observability is the condition for not lapsing into unfalsifiable civilizational commentary.

The Itajima & Co. hypothesis

What follows is a hypothesis held by Itajima & Co. It is neither established fact nor industry consensus.

Grid connection is changing from one procedural step in procuring electricity into a de facto permitting regime for industrial siting. Increasingly, the location of power-intensive industry will be decided not by "is there electricity here" but by "can we join that queue, when does our turn come, and can we carry the cost and risk."

"Permitting" here does not mean a permit in the legal sense. It denotes functional equivalence: that connection comes to perform the role of allowing or refusing a site. This point connects directly to the objections below.

The causal chain behind the hypothesis

  1. Responsibility for network upgrade costs and risk is clearly assigned to the demand side
  2. Parties able to absorb that cost and risk effectively occupy the front of the queue
  3. Connection capacity begins to behave as a scarce good, acquiring a price and a rank
  4. The dominant variable in siting decisions moves from the price of electricity to the date connection becomes available
  5. As a result, the design of the connection allocation regime determines the geography of industry

The strongest objections

Objection 1: this is nothing more than transitional congestion. Once transmission investment and distributed generation catch up, the queue clears and connection reverts to ordinary procedure. Historically, infrastructure bottlenecks have been resolved again and again.

Objection 2: "permitting" is an abuse of metaphor. FERC is not refusing connection; it is clarifying where costs belong. Clarifying costs is procedural transparency — the opposite of discretionary permitting.

We have no decisive rebuttal to either at present. Objection 2 in particular lands precisely on the weakness of this essay's vocabulary. The claim we can defend is only the limited one: that connection functions as permitting.

What remains uncertain

  • FERC's show cause orders are orders, not final rules. Responses are due August 17, 2026, and how far this ends in uniform rules is undetermined.
  • Japan's measures center on efficiency regulation; the design of connection allocation itself is still under study.
  • Industry media report multi-year connection waits in major data center clusters, but we could not verify this independently and therefore use no figures for it.
  • That the US and Japanese changes crystallized at the same time is an observation, not evidence that they are linked. And this essay looks at only two cases. Germany moved from first-come-first-served to a maturity-based allocation procedure in April 2026, so our observational range is not comprehensive.
  • In OCCTO's latest outlook, itemized data center figures fell below the previous projection through fiscal 2033. The connection queue does not lengthen monotonically; it can contract when operators revise their plans. Because our hypothesis assumes a long queue, this downward revision counts against it.
  • "Connection availability becomes the dominant variable in heavy industrial siting" is a prediction derived from the hypothesis. We have not confirmed a source showing that any individual company has decided on that basis.

How to test the hypothesis

Falsification conditions

Any of the following would refute the hypothesis or require substantial revision.

  • If, by the end of 2028, large load connection lead times shorten significantly across major RTO regions and connection availability ceases to be cited as a factor in new large manufacturing siting decisions
  • If FERC's proceeding ends with the principle of demand-side cost and risk responsibility withdrawn or substantially relaxed
  • If regions where self-generation and co-location offer an easy institutional exit expand, and a substantial share of large load actually leaves the grid queue

Indicators to watch

  1. Queue depth and average lead time for large load connection, by RTO/ISO
  2. Levels of financial security, prepayment and termination charges required in connection agreements
  3. Lead time from grid connection application to commercial operation in Japan
  4. The number of states and municipalities imposing electricity taxes on data centers, and their rates
  5. Whether heavy industrial siting announcements disclose power and connection terms
  6. The share of large loads choosing co-location or self-generation

"Connection rights" is not a new term

We initially considered "connection rights" — the state in which queue position for grid connection carries a price and a rank as a scarce good — as a candidate for a new concept. Having checked prior usage, it is not a new term. Below is the existing usage, and how ours differs.

  • Connection rights is already established in academic and regulatory usage. Pollitt, Duma, Mitchell & Covatariu, "Managing Great Britain's electricity distribution connection queue: lessons from auction theory and a potential position trading system" (Utilities Policy, 2025), treats the queue problem as the design of "the initial (primary) allocation of connection rights," and further examines "a secondary trading of connection rights to increase efficiency." What this essay described as connection rights has already been named and formalized in auction-theoretic terms by that literature.
  • Naming connection capacity as a scarce resource is already established practice. In Germany, transmission system operators replaced first-come-first-served with a maturity-based allocation procedure (Reifegradverfahren) on April 1, 2026; the law firm Noerr's commentary (April 28, 2026) calls it "the scarce resource 'grid connection capacity'" and notes that a secondary market in connection reservations already exists.
  • Japan has its own precedent for rights concepts in grid use — though of a different kind. Use of interregional interconnectors was governed by first-come priority and a prohibition on speculative holding until an indirect auction was introduced in October 2018, followed in fiscal 2019 by indirect transmission rights as a hedge against price-difference risk (Japan Electric Power Exchange and Agency for Natural Resources and Energy, "Report of the study group on the design of indirect transmission rights," March 2025). Indirect transmission rights are a financial claim on interregional price differences, not a right to a place in the connection queue.

Where we differ. Existing usage deals chiefly with the generation-side queue and with improving allocative efficiency — auction design, secondary trading of queue positions. What this essay observes is the demand-side queue, and our interest is not efficiency but the point at which that allocation begins to function as the effective decision on whether an industrial site is possible at all. We have therefore not invented "connection rights"; we are extending an existing term into a demand-side, industrial-siting context.

Note also that Germany's April 2026 change precedes the movement we described as crystallizing between June and July 2026. Reform of connection allocation is under way beyond the US and Japan, and our observational range is not comprehensive.

To decision-makers

When you evaluate a site for capital investment, asking "what does electricity cost in this region" reveals less, and later, than asking "can we join that grid queue, can we get as far as a construction cost contribution agreement, and when does our turn come." Given that Japan's demand outlook already filters likelihood by exactly that stage of progress, the question is supported from the statistical side as well.

Itajima & Co. enters the industrial frontline and reads technology, equipment and constraints back into units of capability — what can actually be done. Connection order is currently among the constraints bearing hardest on that reading. We set out the thinking behind it in What Is Enterprise Intelligence?

Frequently asked questions

Won't building more generation and transmission solve this?

Adding supply is necessary, but it is a different question from the one here. What FERC ordered in June 2026 was not the construction of generation; it was a review of how connection applications are studied and where cost responsibility sits in the tariff. Even as capacity grows, if the procedure deciding who connects first is slow, an individual developer's wait does not shorten. That said, the objection that this is merely transitional congestion remains live, and we have not refuted it.

How do Japan and the United States differ?

Their institutional focus differs at present. In the US, connection procedure and cost allocation are themselves the subject of regulatory correction, and taxation has begun at the state level. What Japan introduced is an efficiency standard for new data centers (under the proposed rule, those built from fiscal 2029 must reach PUE 1.3 or lower two years after entering service); the design of connection allocation rules is still under study. What they share is that Japan's demand outlook (OCCTO) is built on grid application status and progress toward construction cost contribution agreements — the queue has become the foundation of policy.

Does this concern industries other than data centers?

It does. FERC's proposed definition of large load covers commercial and industrial customers on the order of 50 MW, and is not limited to data centers. Electric arc furnaces, hydrogen direct reduction, chemicals and batteries can all fall within the same framework. We stop short of saying they will, because the definition is FERC's proposal and each RTO/ISO may adopt its own (SPP's alternative, with lower thresholds, has already been accepted). Nor have we confirmed a source showing that siting decisions in these industries have actually turned on connection availability. On that point this essay offers a prediction, not an observation.

What does "connection rights" refer to?

The state in which queue position for grid connection comes to carry an effective price and rank as a scarce good. We did not coin it. "Connection rights" is already used in the academic literature on queue allocation design (Pollitt et al., Utilities Policy, 2025), where both primary allocation and secondary trading are examined. Japan's electricity system has a concept called indirect transmission rights, but that is a financial claim on interregional price differences, not a right to a place in the connection queue. We extend the existing term to the demand-side queue and to its function in deciding whether an industrial site is possible. It is not a legal right.

Sources

Verification status differs by claim, so sources are grouped by type.

Official primary sources (checked against the original)

  • Organization for Cross-regional Coordination of Transmission Operators (OCCTO), "Nationwide and area-by-area demand outlook, fiscal 2026," published January 21, 2026, Annex 1 "Itemized accounting for new and expanded data centers and semiconductor plants" — basis for the demand figures, methodology, and the difference from the previous outlook
  • OCCTO, "Nationwide and area-by-area demand outlook, fiscal 2025," published January 22, 2025 — the previous edition; source of the 470 MW / 6,160 MW / 3.0 TWh / 44.0 TWh / 7,150 MW figures
  • Commonwealth of Virginia, H.B. 30 (2026 Spec. Sess. I), Item 3-5.24, signed June 30, 2026, effective July 1, 2026
  • Agency for Natural Resources and Energy, Energy Conservation Division, "Institutional responses toward data center efficiency," June 9, 2025, p. 4 — basis for the PUE figures (the PDF on enecho.meti.go.jp was unreachable; the passage was confirmed in the identical document hosted by the Tokyo Metropolitan Government)
  • Japan Electric Power Exchange and Agency for Natural Resources and Energy, "Report of the study group on the design of indirect transmission rights," March 2025

Official primary sources we could not reach

  • Federal Energy Regulatory Commission, show cause orders, June 18, 2026, Dockets EL26-67-000 through EL26-72-000 — the FERC originals remain unchecked. Automated access to both ferc.gov and eLibrary was refused. The issuance date, the mapping of dockets to RTOs/ISOs, the five categories, the deadlines, and the description of the cost recovery agreement rest on agreement among the independent commentaries listed below. Note that those commentaries still differ on the large load thresholds (50 MW and 69 kV, "at least" versus "over") and on the calendar dates for the 30- and 45-day deadlines.
  • Agency for Natural Resources and Energy, "Guideline for measures applying to the data center industry under the Energy Conservation and Non-Fossil Transition Act," adopted April 10, 2026 — automated access to the PDF was refused; unchecked. It is the successor to the PUE figures above.

Law firm and research institute commentary (secondary)

  • Orrick, "FERC Show Cause Orders Signal Broad Reform to Large Load Interconnection Policies," July 2026 — the five categories, the large load definition, the deadlines
  • Rocky Mountain Institute, "Understanding FERC's Large Load Orders," 2026 — that the large load definition is FERC's proposal, the acceptance of SPP's "high impact large load" definition, and financial security under the cost recovery agreement
  • Troutman Pepper Locke (Washington Energy Report), Foley & Lardner, White & Case, Morgan Lewis (Power & Pipes), McGuireWoods, Day Pitney, Holland & Knight, Baker Botts, Bracewell, National Law Review — all June–July 2026 commentary on the orders, used to cross-check date, dockets and deadlines
  • Williams Mullen, "Virginia Budget Creates New Electricity Consumption Tax for Data Centers," 2026; Greenberg Traurig, "Virginia Legislature Approves Tax on Data Center Electricity Consumption," June 2026
  • Noerr, "New rules for allocation of the scarce resource 'grid connection capacity,'" April 28, 2026 — Germany's Reifegradverfahren
  • Belfer Center for Science and International Affairs, Harvard Kennedy School, "AI, Data Centers, and the U.S. Electric Grid: A Watershed Moment," February 2026
  • World Economic Forum, "Is power grid connectivity the strategic bottleneck for AI?", May 18, 2026

Peer-reviewed literature

  • Pollitt, M.G., Duma, D., Mitchell, R., Covatariu, A., "Managing Great Britain's electricity distribution connection queue: lessons from auction theory and a potential position trading system," Utilities Policy, 2025, DOI: 10.1016/j.jup.2025.102099 — prior usage of "connection rights"