Summary

  • The Large Load Registry, effective 12 October 2026, records the address, the in-service date, the ramp schedule, the backup fuel and the permit limit of every load of 50 MW or greater, and its only sworn statement attests to the distance between affiliated facilities.
  • The commitment filter that decides whether a load moves the capacity requirement, "Firm," is defined in a load-forecasting manual presentation rather than the filed attachment, whose first section is marked [RESERVED]; the 2027/2028 auction still raised the RTO reliability requirement by 6,295 MW UCAP, including 5,249.9 MW of forecasted large load.

A registry is a list that other arithmetic is allowed to trust. PJM Interconnection's Large Load Registry was accepted by the Federal Energy Regulatory Commission on 9 October 2026 and took effect on 12 October, and almost everything in it is a fact somebody can look up: an address, an in-service date for each increment of load, a ramp schedule, the fuel a backup generator burns, the permit limit on that generator, a telemetry object ID. One instrument in the file is signed by a person under penalty of perjury. It concerns how far apart two buildings stand.

The same word, written twice

The operative definition is not vague. FERC accepted PJM's text, which reads:

end-use customer(s) load that has a cumulative peak load of 50 [MW] or greater at a single electrical site behind one or more points of interconnection, where for configurations involving multiple affiliated load facilities or multiple points of interconnection with affiliated load facilities behind them, all affiliated load facilities within a one-mile radius will be considered a single electrical site as further described in the PJM Manuals

That is the order of 9 October 2026, and the population it fixes is a polygon: one electrical site, one or more interconnections, affiliated facilities inside a one-mile radius folded in together. Eleven months earlier the PJM Board's decisional letter had proposed a different object altogether: "individual additions of load at or above 50 MW at a single Point of Interconnection." One point of interconnection, one addition, no radius.

The wording is still drifting. A filing docketed 16 November 2026 describes Large Loads parenthetically as "loads over 50 MW at a single location" — looser than the accepted threshold, and not yet operative text. The pending docket asks for an effective date of 16 March 2027, which means the loose paraphrase is a proposal, not the rule. But it tells you where the attention sits inside PJM: on the boundary of the category, not on the substance of the claim.

What the form actually collects

Read the interim Manual XX tabled at the 18 November 2026 Members Committee meeting and the registry is a well-built inventory. Description and identity of the load, locational and modelling information, backup or on-site generation, telemetry including ICCP object IDs, contacts. The compliance filing adds the operating rules: entities submit by 1 March 2027 for loads in service before 1 June 2027, and before the in-service date for everything later. The submitting entity "will ensure that all information contained within the Registry remains accurate and will update the Registry whenever material changes occur." PJM "may require periodic verification of data."

Then comes the sentence the whole piece turns on, and it is permissive:

Duly-signed officer declarations, signed under penalty of perjury, for affiliated loads attesting to the distance between the facilities may be relied upon to calculate the one-mile radius

Read it twice. The oath exists to settle whether two facilities count as one site. Whether the load exists, is financed, has signed anything, or will ever run is not what the officer swears to. Distance is a geometric fact with a precise answer, and PJM took the trouble to make it swearable. The commercial fact underneath it — that there is a customer who will actually draw 50 MW — is not sworn at all in this instrument.

The section that is reserved

The filed tariff attachment carrying the registry is short, and its first section is a placeholder. Attachment DD-4, as reproduced in the compliance filing's transmittal, marks section I "[RESERVED]"; section II says the Office of the Interconnection shall maintain the registry pursuant to Reliability Assurance Agreement Schedule 11; section III is the distribution company's duty, and its standard for what to hand over is self-assessment:

Electric Distributors shall coordinate with the Office of the Interconnection in developing and maintaining the Large Load Registry by providing the Office of the Interconnection any information or data the Electric Distributor believes appropriate to facilitate an accurate Large Load Registry

"Believes appropriate." That is not a data-quality floor; it is a discretion. And the compliance filing's own account of what it changed is subtraction: the two fields tied to the rejected package came out — the amount of capacity a large load brought with it, and the allocated backstop capacity. What remains records the load's attributes, not its capacity position.

Two documentary classes are doing different work here, and they should not be flattened. The tariff attachment is filed text. Manual XX is manual text, interim, tabled eight days before this article. I read Attachment DD-4 as reproduced in the compliance transmittal and did not verify the field-by-field language of Reliability Assurance Agreement Schedule 11.B.1(a) from the filed document, so the field list used here comes from FERC's description of the proposal and from Manual XX.

Where the word that filters actually lives

The word that decides whether a forecast megawatt moves the capacity market is not in that attachment. It is "Firm," and its definition is best found in a slide deck. The February 2026 Manual 19 presentation on load forecast adjustments says:

Firm will mean having an ESO or CC plus a meaningful/verifiable financial commitment (such as credit/collateral support, financially significant infrastructure investment, or long-term supply commitment with new or existing resources)

A February 2026 workshop presentation states the consequence plainly: only data-centre projects identified with an executed service obligation or construction commitment and marked "Firm" are allowed to impact RPM; the rest are labelled Non-Firm; PJM imposed a minimum 36-month ramp and a 70% utilisation rate unless otherwise supported. The same deck records PJM reviewing economic data "to mitigate double counting load growth." That is a real screen, with real thresholds, sitting in Manual 19 Attachment B, Load Forecast Adjustment Guidelines, under issue identification and issue verification. It is manual and stakeholder material, not the filed attachment.

The timing matters and I will not paper over it. The workshop is dated 25 February 2026 and the 2027/2028 auction parameters were set weeks earlier, so those controls describe the 2026 long-term forecast process rather than the exact derivation of the number below. What the documents do establish is the split: the thing that makes a load count for reliability is governed by a financial-commitment test in a manual, while the thing that makes a load visible is a registry whose sworn instrument is geographic.

What the load did to a requirement

The 2027/2028 Base Residual Auction report prints the chain in one line: the RTO Reliability Requirement rose from 146,105 MW to 152,400 MW UCAP, "including a 5,249.9 MW increase in forecasted load, largely due to additional Large Loads." The installed reserve margin went from 19.1% to 20.0%, the forecast pool requirement from 0.917 to 0.926, and every price cleared at the temporary cap of $333.44 per MW-day UCAP.

My arithmetic, not PJM's: the requirement rose 6,295 MW, and 5,249.9 MW is 83.4% of that increase. PJM's auction release then attributes "nearly 5,100 MW" of a roughly 5,250 MW forecast increase to data-centre demand — 97.1% of the printed load increment. The auction cleared 134,478.1 MW UCAP, with 809.6 MW uncleared because offers exceeded the cap, and FRR areas committed 11,299 MW for a total of 145,777.1 MW, which the report measures as 6,516.6 MW short of the requirement.

Two dollar figures, two documentary kinds

PJM prints $16.4bn as cleared supply times clearing price, and then prints the qualifier that matters: that figure "does not equate to a total cost to load because load that is hedged through self-supply or bilateral contracts is not exposed to the clearing prices in the auction." My division: 134,478.1 MW at $333.44 per MW-day times 365 days is about $16.37bn. The headline is a settlement base, and PJM says so.

The Independent Market Monitor produces a different quantity from a different method: including existing and forecast data-centre load in the 2027 peak forecast raised capacity-market revenues by $6,497,653,512, or 65.5%, for this auction, $7,271,197,971 or 82.1% for 2026/2027, and $23,100,955,341 across three auctions. My arithmetic again: 65.5/165.5 puts data-centre load at 39.6% of auction value, and $6,497,653,512 over $16.4bn reproduces that share. The Monitor's counterfactual uses an "uplift MW" construction whose definition the passage I read does not fully specify, so the dollars are an estimate, not a billing record, and dividing them by 5,249.9 MW would produce a quotient that is a scale check and not a rate any load pays.

One number I will not tidy: the report measures the shortfall at 6,516.6 MW and 14.4% IRM, while the release says 6,623 MW and 14.8%. The difference of 106.4 MW is a difference of bases, not an error, and the two are not interchangeable.

What the chain does not close

The report does not itemise 5,249.9 MW by project. No individual data centre can be traced into that requirement from these documents, and I read no operator's own filing. What a 2023 Markets Implementation Committee presentation shows is the shape of the exposure: the zonal capacity obligation is actual zonal peak load times a zonal scaling factor times the forecast pool requirement, and during the forecast period a large load addition is spread across all load-serving entities in the zone rather than assigned to the load that caused it — until it comes online and joins that entity's actual peak load. That presentation predates later adjustments to large-load allocation, so read the formula as the printed mechanics, not necessarily current settlement practice. Still, the consequence is legible: a forecast megawatt that never materialises has already been charged, for a while, to someone else.

Where the accountability sits

PJM maintains the registry. It does not warrant it. The accuracy duty in Manual XX belongs to the submitting entity; the information duty in filed Attachment DD-4 belongs to the distribution company and is bounded by what it "believes appropriate"; and PJM's own check is discretionary periodic verification with no stated frequency, method or consequence in the text I read. The one place the record is hardened is the fence line.

That is the finding, and it is narrower than an accusation. Across the eleven documents I read for this piece — the Board letter, the FERC order, the compliance filing, the interim manual, the Manual 19 presentation, the auction report and release, the Monitor's section, the 2023 presentation, the 2026 workshop deck and the pending November filing — none conditions registry reporting or inclusion in the requirement on a sworn or certified statement about a load's commitment, financing or likelihood of construction.

Another instrument may exist; I did not read the Reliability Assurance Agreement in full, and I make no claim beyond this corpus.

But the asymmetry is the story. PJM built a registry that can place a load to the metre and cannot require anyone to swear it will be built, then used the resulting forecast to lift a reliability requirement by 6,295 MW into an auction that cleared at its cap. The oath went where the boundary was contested and stayed away from where the money was. If the registry is going to carry that much weight, the sworn statement should attach to the thing that moves the requirement, not to the distance between two fences.