Large-load disclosure · one operator, one file · a dated series

One U.S. grid operator publishes how much of its interconnection queue is AI data centers.

FILENYISO-Interconnection-Queue.xlsx · edition July 31, 2026
DECLARED AI LOAD (DAT-AI)5,290 MW
SERIESre-read against every new edition — an alarm fires if the arithmetic stops matching
CUSTODYread August 21, 2026 · SHA-256 b3026855903e… · how this page is governed
Free to cite · CC BY 4.0 · no permission needed
How much declared AI load?
5,290 MW — 12 requests
39% of the 13,568.9 MW that NYISO publishes as classified load in its queue, under the operator's own DAT-AI code — which its key defines as “AI Datacenter” — in the edition dated July 31, 2026. The operator states the scope of that total in the table itself: it “excludes withdrawn and in-service load projects”. The end-use code appears on 51 of the 73 requests: withdrawn rows are left blank, and a few others are.
Who says so?
The operator itself
The breakdown is NYISO's own table, titled MW By End-Use on the Load Project Tracking tab. The MW and the labels are the operator's; the request counts are ours, from the individual rows. Here it is, whole:
DAT Data Center (general)
5,877 MW · 43%
DAT-AI AI Datacenter
5,290 MW · 39%
M-CH Microchip fabrication
1,056 MW · 8%
DAT-CM Cryptocurrency mining
745 MW · 6%
RD Research and Development
395 MW · 3%
M-CG Consumable Goods
106 MW · 1%
M-IN Industrial Goods
50 MW · 0%
O Other
50 MW · 0%
The full table as published, and its scope+
CodeEnd-use, as publishedMWRequestsShare
DATData Center (general)5,876.92443.3%
DAT-AIAI Datacenter5,290.01239.0%
M-CHMicrochip fabrication1,056.027.8%
DAT-CMCryptocurrency mining745.045.5%
RDResearch and Development395.052.9%
M-CGConsumable Goods106.020.8%
M-INIndustrial Goods50.010.4%
OOther50.010.4%
Total published by NYISO13,568.951100%

Two classes in the published taxonomy carry no load in this edition: M Manufacturing (general), HP Hydrogen Production. We list them because the taxonomy is wider than what currently appears in it.

Taken together, the three data-center classes account for 11,911.9 MW88% of the classified load in this table. The denominator is NYISO's, and so is the caveat on it: the table “excludes withdrawn and in-service load projects”. It is not the whole queue, and we do not call it one.

Can you compare AI vs crypto “seriousness”?
No — that measures age, not intent
95% of the AI-datacenter class entered the queue in 2025; 100% of the cryptocurrency-mining class entered between 2018 and 2021. A request filed in 2019 has had years to move through studies that a request filed in 2025 has not yet had. That is arithmetic about time, not evidence about either industry.
16
18
19
20
21
22
23
24
25
26
DAT-AI MW by submission year
16
18
19
20
21
22
23
24
25
26
DAT-CM MW by submission year
NYISO publishes end-use totals and submission-year totals as separate tables and does not cross the two — the cross above is ours, computed from the individual rows.
The comparison to avoidMaturity is not comparable across these classes without controlling for the age of the request. In this file the AI-datacenter class is 95% 2025 vintage, while the cryptocurrency-mining class is entirely 2018–2021. Any comparison of how far along the two classes are measures queue age, not credibility.
The exact MW by submission year (our cross, in full)+
Code2016201820192020202120222023202420252026
DAT300···120300·9073,2001,050
DAT-AI·······2505,040·
DAT-CM·6050435200·····
What’s missing from the file?
The signature date
NYISO’s status ladder names the executed interconnection agreement explicitly — 11, IA Completed — and in this file no load request sits at 11. The IA Tender Date column is empty on all 73 rows.
The full finding — the gap that survives all of it+

The ladder carries four further rungs beyond the agreement, through construction to commercial service. Four requests stop one rung short, at Accepted Cost Allocation/IA in Progress, while eight sit further along and so plainly cleared that stage — and FS Completion Date is empty as well. So the most specific large-load disclosure we have found follows a request from first study to commercial operation without ever recording when the agreement was signed. We are describing the record, not the world: those agreements plainly exist. They are simply not in the file. The wider survey of what each operator publishes is at index.loadstar.energy/queues.

What this record cannot tell you

One jurisdiction+
NYISO is New York. Nothing here supports a national figure, and we do not publish one from it.
Who assigns the code? Not established+
The file publishes the taxonomy and the values; it does not say whether the developer selects the class or the operator assigns it. Attributing the classification to the applicant would go beyond the record — ours and anyone else’s.
The list is not a census+
NYISO’s interconnection process reaches load above a threshold set in its tariff; requests below it go to the transmission owner’s own procedure and never appear in this file. We have not verified that threshold and do not state one.
Withdrawn rows carry no end-use code+
Every withdrawn row in this file has the column blank. A withdrawal rate by load type cannot be computed from it — not by us, and not by anyone else working from the same file.

How we read it — the arithmetic, reproduced

73 rowsthe workbook's load tab, 16,922.9 MW
− 2,978.8 MWwithdrawn, by NYISO's own count
= 13,944.1 MWnot withdrawn (our reading)
13,568.9 MWthe operator's classified total — matches ours to the decimal
Δ 375.2 MWfive rows with no end-use code, four already in service
Why we show the reconciliation+

The workbook has 73 load rows totalling 16,922.9 MW. Subtracting the 2,978.8 MW NYISO reports as withdrawn leaves 13,944.1 MW not withdrawn. The operator’s own end-use table totals 13,568.9 MW, and our independent reading of that classified population matches it to the decimal. The 375.2 MW difference is five not-withdrawn rows carrying no end-use code, four of them already in commercial service.

We report the reconciliation rather than a single number because a survey that cannot reproduce an operator’s own arithmetic has no business describing that operator’s file. A collector repeats this comparison against every new edition, and raises an alarm if it ever stops matching.

Source. NYISO Interconnection Queue workbook, Load Projects and Load Project Tracking tabs, in the edition dated July 31, 2026, published at nyiso.com/interconnections. NYISO republishes this file and overwrites the previous edition; we keep the dated editions.

Chain of custody. Downloaded by a person on August 21, 2026 and byte-compared against an independent web-archive capture of the same edition; the two are identical. SHA-256 b3026855903e58e689d532c5fbbec18fa09c4f3568eb0b4a4010f69cab84ce54.

What we do not publish. No project or company name appears on this page. It is not needed for anything stated here.

Independence. NYISO is not affiliated with LOADSTAR and does not endorse this analysis. The end-use codes and totals are NYISO’s published data; the reading of them is ours.

Context, dated. On July 14, 2026, New York’s Executive Order 62 paused discretionary state environmental permits for new data centers of 50 MW and above. The order does not reach grid interconnection and does not alter the queue file this page reads — its own text cites the load requests sitting in the NYISO queue.

Cite as: LOADSTAR, Declared AI Load in the NYISO Interconnection Queue, the edition dated July 31, 2026, read August 21, 2026, index.loadstar.energy/ai-load. Free to cite and reuse under CC BY 4.0.
← The full seven-operator survey