Summary

  • PJM's load forecast is the document that sets the peak loads behind capacity obligations, and one multiplication in the tariff turns that forecast into the requirement an auction must satisfy: 164,579 MW times 0.926 equals 152,400.2 MW of unforced capacity for the 2027/2028 delivery year.
  • In the large-load screening PJM presented in late 2025, only projects holding an electric service obligation or a construction commitment were marked "Firm" and allowed to move that forecast, under a default 36-month data-centre ramp and a 70% utilisation factor unless the supplier supplied better data.
  • On 9 October 2026 the regulator accepted the load's name and its registry entry and rejected the interim-service, designate-capacity and curve-exclusion routes. The register records the load. It does not compute it.

One multiplication, published in advance

On 26 August 2025, PJM posted the planning parameters for its 2027/2028 Base Residual Auction. The document contains a sentence and a table, and between them they fix what the region's capacity will cost before a single data hall is energised: "The forecast peak load for the PJM RTO for the 2027/2028 Delivery Year is 164,579 MW which is 5,250 MW, or 3.3%, above the forecast peak load of 159,329MW for the 2026/2027 BRA" (2027/2028 RPM Base Residual Auction Planning Period Parameters). The narrative paragraph prints the result as "152,400 MW". The parameter table on the same posting prints 152,400.2 MW UCAP. Both printings exist; I quote the table and note the paragraph.

That figure is not a measurement of any customer. It is a forecast of a regional peak multiplied by a reserve factor. My own arithmetic, labelled as mine: 164,579 times 0.926 is 152,400.15, which reproduces the posted 152,400.2 at the printed rounding. The multiplication is not hidden, disputed or approximate in the way a rumour is approximate. It is the operating rule, published seventeen months before the delivery year began and still public now, while that delivery year has not started.

The manual says what the forecast is for

PJM's load-forecasting manual is blunt about the job. "The PJM Load Forecast Model produces 20-year monthly forecasts of unrestricted peaks assuming a range of weather conditions for each PJM zone, locational deliverability area (LDA) and the RTO," it states, and then the sentence that matters: "It is used to set the peak loads for capacity obligations, for reliability studies, and to support the Regional Transmission Expansion Plan" (PJM Manual 19: Load Forecasting and Analysis, Revision 39, effective 30 June 2026).

A twenty-year horizon is longer than most of the assets it describes will last in their first configuration. Where a zone is expected to do something the historical model does not capture, the manual allows PJM to adjust model inputs or to make "an explicit adjustment to the modeled forecast" — and for load that has not yet occurred, PJM works from what the electric distribution companies report in answer to its annual request about large load changes, weighing certainty and magnitude case by case. Read that twice.

The one input that cannot be measured is supplied by the utility that wants the load, assessed by the operator that runs the auction.

The capacity-market manual then puts the forecast inside the clearing process: the RTO and zonal peak load forecasts are used in RPM auction clearing and to determine the RTO Reliability Requirement (PJM Manual 18: PJM Capacity Market, Revision 63, effective 24 September 2026). It also sets out the re-posting sequence, which is the only scheduled moment the number moves: a preliminary forecast goes out with the Base Residual Auction planning parameters, updated forecasts before the first and second incremental auctions, and a final forecast before the third.

The sentence that converts a forecast into a bill

The tariff does not leave the conversion to inference. In the currently rendered definitions, "PJM Region Reliability Requirement" means, for the Base Residual Auction, "the Forecast Pool Requirement multiplied by the Preliminary PJM Region Peak Load Forecast, less the sum of all Preliminary Unforced Capacity Obligations of FRR Entities in the PJM Region", and for the incremental auctions the same product against the updated forecast (PJM Open Access Transmission Tariff, definitions). The page carries an effective-date stamp of 1 April 2026 and Docket ER26-846-001, which is how a live tariff record looks rather than a stakeholder proposal. What I read is the definitions page. Section 5 of Attachment DD, where the forecast is established on a preliminary and final basis, was not read in currently effective form, and I say so rather than pretending otherwise.

Down at the level where an obligation actually attaches to a company, the arithmetic grows a third factor. Under the reliability agreement's Schedule 8, "Daily Unforced Capacity Obligation = OPL x Final Zonal RPM Scaling Factor x FPR", where the Obligation Peak Load is the daily summation of the zone's weather-adjusted coincident summer peak, net of operating behind-the-meter generation and not below zero, "as adjusted further to include the Obligation Peak Load associated with any Large Load Adjustment that may be allocated to the Party" (RAA Schedule 8 revisions, Members Committee clean package; the same formula and definition appear in the merged Reliability Assurance Agreement).

Two different quantities live inside that one defined term. One is a coincident summer peak — a thing that happened, at a meter, in a zone. The other is an adjustment — a thing forecast, at zone/area level, summed by zone, "as further detailed in PJM Manuals". A data centre enters the obligation through the second half. That is why talking about the load's "demand" in the singular misses the mechanism: the obligation is computed on a zone's peak plus an allocation, and only the first of those is a measurement.

The same term, printed two ways

Here is where a careful reader earns their keep. The 2024 clean package and the merged agreement both print "Large Load Adjustment". The regulator's order of 23 September 2024 in 188 FERC para. 61,200 accepted the underlying revisions effective 31 August 2024 and directed PJM to strike the word "Large", and PJM's compliance filing of 21 October 2024 says so in its own words: "consistent with the Commission's directive, PJM submits this compliance filing to amend the two sentence description of 'zone/area' ... and to amend the term 'Large Load Adjustment,' and all proposed references thereto, to remove the word 'Large'" (PJM compliance filing, Docket No. ER24-2447-002). The filing renames the abbreviations with it: ZLLA becomes ZLA, LLAOPL becomes LAOPL, LLAMW becomes LAMW.

So the same defined term is on the public record twice, with and without its adjective, and the regenerated agreement still carries the older one. I print both and adopt neither as the operative wording. But the divergence is itself instructive: a regenerated PDF is not the amendment record. When a document and the filing that was ordered to change it disagree about a defined term, the filing tells you what the Commission did and the PDF tells you what has not yet been reflowed.

The screening that decides how much of a project counts

The part of this system that is easiest to miss is the part that decides how many megawatts of a named project ever reach the forecast. PJM's Stage 3 committee package sets the inclusion tests: a large load coming online in three years or less with a construction commitment or an electric service obligation "will be considered for inclusion in the PJM forecast", while a load more than three but less than eight years out without construction commitments is considered only if it has cleared demonstrable project milestones or is de-rated for uncertainty; absent a supplied ramp rate, a default three-year ramp applies, and inclusion requires that non-disclosure agreements let the relevant information reach PJM (PJM CIFP Package Stage 3: Large Load Additions). A committee package, not adopted manual text — the distinction matters and I am keeping it.

Then the preliminary screening summary presented to PJM's Load Analysis Subcommittee in November 2025: "Worked to identify those that had ESO/CC and marked those as 'Firm'. Only these projects will be allowed to impact RPM," with a minimum 36-month ramp imposed on data centres going from online to full demand and a 70% utilisation factor "unless otherwise supported" (Load Adjustment Requests Summary for 2026 Load Forecast - Preliminary).

Put the two documents together and a plain consequence follows. Two campuses with identical nameplate capacity can arrive in the same forecast at different megawatts, and the difference is documentary rather than physical: one holds an electric service obligation or a construction commitment on paper and the other does not. Nothing in either document says the difference is wrong. It says the difference is decided in a filing, at a ramp rate and a utilisation factor, by an operator exercising judgement case by case.

PJM's own long-term load forecast report shows the judgement landing in specific zones. Its table of adjustments lists AEP as adjusted for "growth in data center load and a chip processing plant", COMED for data-centre load and an electric vehicle battery plant, DOM for data-centre load and a voltage optimisation programme, and APS, ATSI, BGE, DAYTON, PECO and PL for data-centre load alone (PJM Long-Term Load Forecast Report (2025)). That is the closest the public record comes to saying data centres moved the regional number. The report names which zones were adjusted. It does not print the megawatts of each adjustment.

What the auction then did

The 2027/2028 Base Residual Auction report states that the auction "cleared 134,478.1 MW of Unforced Capacity (UCAP) in the RTO from annual, summer-period and winter-period matched resources", and that for that delivery year "all prices cleared at the cap ($333.44)", with no constrained locational deliverability area among the RTO, EMAAC, SWMAAC and MAAC (2027/2028 Base Residual Auction Report, dated 17 December 2025).

PJM's own news release of the same date tells it in rounder numbers: 134,479 MW cleared, an additional 11,299 MW under the Fixed Resource Requirement for 145,777 MW in total, the price at "the FERC-approved cap, $333.44/MW-day (UCAP) for the entire PJM footprint, a slight increase (+1.3%) from the 2026/2027 Base Residual Auction", and committed supply 6,623 MW short of the reliability requirement (PJM news release, 2027/2028 Base Residual Auction results).

Two arithmetic checks, both mine and both labelled. First: 152,400.2 minus 145,777 is 6,623.2, which reproduces the printed shortfall of 6,623 MW once the Fixed Resource Requirement megawatts are added to what the auction itself cleared. That is the cleanest closure in the set, and it is the reason I trust the pairing of these two documents. Second: 152,400.2 minus 134,478.1 is 17,922.1 MW, the gap between the posted requirement and what the auction alone procured. That is a different quantity from the 6,623 MW shortfall and the two must never be added, subtracted or blended into one number.

The release also prints a 14.8% reserve margin where the report's summary prints 14.4% for RPM and 14.9% in total. Rounding or basis, I cannot tell from what I read, so all three printings stand as printed and none is averaged.

The regulator's October decision

On 9 October 2026 the Commission decided PJM's August 2026 filing in Docket ER26-3515-000. It "accept[ed] PJM's proposed Tariff revisions to establish a definition of Large Load and a Large Load Registry, effective October 12, 2026, as requested", required a compliance filing within thirty days, and then: "We reject PJM's remaining proposed Tariff revisions, as discussed below" (PJM Interconnection, L.L.C., 197 FERC para. 61,019).

The rejected remainder is the interesting half. It comprised an interim resource adequacy service for new large load, a requirement that a supplier serving new large load designate new capacity at least equal to that load's peak in order to keep it out of that interim service, the rules for bringing your own new capacity, and the exclusion of eligible new large load from the variable resource requirement curve from the 2029/2030 delivery year.

Read the dispositions together and the shape is plain: the load acquires a name and an entry in a register, and it does not acquire a route out of, or down within, the obligation that the forecast produces for it. That reading is mine, drawn from the accept-and-reject list. The order's reasoning beyond those dispositions is not characterised here, because what I read is the dispositions.

Eighteen days earlier, a separate order had accepted for filing and suspended a different PJM instrument. The Reliability Backstop Procurement was described as "a one-time procurement of new capacity to address significant resource adequacy challenges primarily driven by an influx of large loads" (PJM Interconnection, L.L.C., 196 FERC para. 61,245). Its suspension terms were not read in full here. What the two orders say side by side is that the shortfall attributed to large loads is being handled partly outside the ordinary auction and partly by naming the loads inside it.

What a reader can check, and what nobody can

A reader can check the posted parameters against the cleared result, because both are published. A reader can check the definitions page against the dockets that amended it, because the dockets are public. A reader can check which zones PJM adjusted and for what stated reason, because the load forecast report says so.

A reader cannot check what any single registry entry contributed to its zone's forecast. Nothing in the material I read links one named large load to the megawatts its host zone carried, and no per-zone breakdown of adjustment megawatts in the 2027/2028 parameters was obtained. That is an absence in the reading of 18 November 2026, not a finding that no such record exists. It is also the whole accountability question in one line: the register can tell you who was named, and the formula can tell you how a number was made, and nothing public joins the two.

A queue would have told a developer its place. A forecast tells it its number, and the number was written before anyone could read its meter — by the same body that runs the auction the number feeds.

What I read, and what I did not

Ten limits travel with everything above, load-bearing first.

  1. The 26 August 2025 parameters may be re-posted with final Fixed Resource Requirement data; nothing I read shows the final figure for 2027/2028.
  2. Section 5 of Attachment DD was not read in currently effective form; the conversion above is quoted from the rendered definitions page, which carries an effective-date stamp of 1 April 2026 and Docket ER26-846-001.
  3. Manuals 19 and 18 were read through provider-reported excerpts of public PDFs, not by page-checking every page; revision numbers and effective dates are as those excerpts print them.
  4. The obligation formula and the adjusted-term definition come from a 2024 Members Committee clean package and from the merged reliability agreement, neither of which is proven here to be the currently effective tariff sheet.
  5. The Commission's 23 September 2024 order in 188 FERC para. 61,200 was not read; its direction is known here through PJM's compliance filing, quoted as PJM's account.
  6. Two public documents still print different names for the same defined term; both are reported and neither is adopted.
  7. The 9 October 2026 order was read through retained passages; its reasoning beyond the accept and reject dispositions is not characterised and no motive is assigned to it.
  8. No document I read links an individual Large Load Registry entry to the megawatts its zone's forecast carried, and no per-zone breakdown of adjustment megawatts for 2027/2028 was obtained.
  9. The Stage 3 package and the November 2025 screening summary are committee and subcommittee material describing methodology in progress; they are not adopted manual or tariff text.
  10. Everything here is a reading of documents obtained on 18 November 2026, and a later revision of any one of them may print something different.

Two further limits belong beside the numbers rather than after them. PJM's two documents of 17 December 2025 print different clearing totals and different reserve margins, all reported above and none reconciled. And nothing in the material read states that any named data-centre project was mis-forecast, favoured, disadvantaged or billed incorrectly, so no such claim is made here.