AI brief, Saturday 31 October 2026
Exelon has published the discount on its own data-centre pipeline — about 11 GW it calls high probability against about 23 GW of future pipeline, with roughly 45% of the first figure backed by FERC-approved agreements and about $1.2bn of collateral. Three days ago Texas answered the same problem by pausing its queue. Alliant raised its four-year capital plan 19%, and nVent's sales grew 44%.
Exelon has published the discount on its own data-centre pipeline
Exelon reported its third quarter on 30 October. The earnings themselves are a utility's earnings — GAAP net income of $0.70 a share, adjusted operating earnings of $0.87, full-year guidance narrowed to $2.85–2.91, growth affirmed near the top of a 5–7% range through 2029. The interesting document is the presentation filed beside it, because it splits the data-centre pipeline into two tiers and puts a number on each.
- About 11 GW of "High Probability Load" — roughly 9 GW at ComEd in northern Illinois plus roughly 2 GW in the Mid-Atlantic — defined as load in "advanced phases of design or backed by FERC-approved TSAs", of which about 45% has a TSA, backed by about $1.2bn in collateral.
- About 23 GW of "Future Pipeline" — roughly 14 GW at ComEd plus roughly 9 GW in the Mid-Atlantic — load "within current or future cluster studies" concluding in 2026 and 2027.
- Exelon says its Large Load Adjustments submitted to PJM in 2025 and 2026 "were fully approved", and that its 2027 submission "reflects enhanced commitment criteria and aligns with ~11 GW of high probability load" — in other words, it is filing against the smaller number.
- Data-centre load in northern Illinois has grown at a roughly 28% compound annual rate between 2022 and 2026, on a year-to-date-September basis.
- And the framework: Exelon describes "protecting customers through FERC-endorsed Transmission Security Agreements while improving queue quality and certainty", and notes FERC "has endorsed Exelon's TSA framework as a model for protecting customers, improving queue certainty, and enabling efficient large-load growth".
Why it matters. Read this against this site's brief of 28 October. Texas faced exactly this problem — an interconnection queue nobody could believe — and answered it by stopping: ERCOT paused Batch Zero classification and said it would not authorise large loads to energise until a Governor-ordered audit of data-centre projects of 25 MW and above was complete, with about $2bn of security held against 45 GW. Exelon and FERC have answered the same problem by pricing entry instead of halting it: a contract, collateral, and a published split between load you should believe and load that is still a study.
Three days ago this site wrote that the most useful figure anyone would publish about AI power demand was the difference between a queue and its verified subset. Here is a utility publishing that difference voluntarily, as 11 against 23. Two cautions. The tiers are Exelon's own definitions, not a regulator's audit, and "advanced phases of design" is doing real work inside the high-probability number — only about 45% of it has a signed agreement behind it. And one quarter's collateral figure, $1.2bn against 11 GW, is a lighter deposit per gigawatt than Texas demanded, which is the price of not stopping.
Sources: Exelon Corporation, third-quarter 2026 results, Exhibit 99.1 to Form 8-K, 30 October 2026, and the earnings call presentation, Exhibit 99.2, which carries the load figures (both SEC EDGAR).
The two-tier distinction travels beyond Exelon's own territory, because it is really a distinction about paperwork: load with a signed service agreement behind it, and load that has a proposal and a place in a queue. Ark7's Project Soperton publishes its own power status in those terms — a written supply proposal still awaiting a signed service agreement, which on this taxonomy is the second tier rather than the first. That is the ordinary condition of a project at this stage rather than an exception, and it is the kind of thing worth checking on any single-site project. Ark7 is an advertiser on this site and we are paid if you open that link; see About.
Alliant Energy raised its four-year capital plan 19%, and froze base rates to 2030
Alliant Energy reported on 29 October, filed 30 October. Third-quarter GAAP earnings were $1.52 a share against $1.09, and the company narrowed 2026 guidance to the top half of its range, $3.41–3.46, while issuing 2027 guidance of $3.63–3.73 and a 2027 dividend target of $2.25.
The data-centre numbers:
- Its two utilities, Interstate Power and Light and Wisconsin Power and Light, have 3.4 GW of contracted demand from data centres.
- Alliant expects peak demand to rise by approximately 60% by 2031, which it calls an industry-leading growth rate.
- It has increased its four-year capital expenditure forecast by 19%, to $15.9bn, to meet that demand.
- And it is extending IPL's retail electric base rate stability plan through 2030 while earning its authorised return on equity.
Among the stated assumptions behind the 2027 guidance is the "realization of expected revenues from new data center demand".
Why it matters. The word doing the work here is contracted. Alliant is not reporting a queue or a pipeline; it is reporting 3.4 GW already under contract, and it has raised its capital plan by a fifth against it. That is the same decision AEP's filing warned about three days ago in its own risk language — projected data-centre demand "may not materialize as expected" — taken anyway, with the contracts as the justification.
The rate freeze to 2030 is the other half, and it is the answer to a question this site raised on day one, when PG&E argued that a gigawatt of data-centre load would lower everybody else's bill. Here that argument is a filed commitment: existing customers' base rates are held while the capital plan grows 19%, which only works if the new load pays for the new capacity. If it does not arrive, the freeze is the thing that breaks — and a guidance assumption is a weaker instrument than Exelon's collateral.
Source: Alliant Energy Corporation, third-quarter 2026 results and 2027 guidance, Exhibit 99.1 to Form 8-K, filed 30 October 2026 (SEC EDGAR).
nVent's sales grew 44%, and all of it was organic
nVent Electric, which makes electrical enclosures, power distribution and liquid cooling for data centres among other things, reported on 30 October.
- Record sales of $1.5bn, up 44% — both reported and organic.
- Reported EPS $1.23, up 68%; adjusted EPS $1.52, up 67%.
- Cash flow $303m, up 12%; free cash flow $268m, up 6% — positive.
- Full-year guidance raised to reported sales growth of 48–49% and organic growth of 38–39%.
- New products contributed approximately 30 points of the sales growth, and nVent closed its acquisition of Maverick Power on 1 October, which chief executive Beth Wozniak said broadens "our data center exposure by adding a new power distribution platform".
Why it matters. This is the fifth consecutive day on which a supplier's results have been better evidence about the build-out than any buyer's forecast — Vertiv, Generac, Wesco, and now nVent. The detail that matters is "both reported and organic": 44% growth with the acquisition closing only on 1 October means the growth is demand, not consolidation, which is the usual way a number like this gets inflated. And free cash flow is positive and rising while the hyperscalers it ultimately sells into are running theirs deeply negative.
One caution, the same one as yesterday. Suppliers of components recognise revenue on delivery and carry order books that customers can thin out, so this is a fast signal in both directions. It has been pointing one way all week. The week it turns will be visible here before it is visible in anyone's capital plan.
Source: nVent Electric plc, third-quarter 2026 results, Exhibit 99.1 to Form 8-K, 30 October 2026 (SEC EDGAR).
Three items, from filings of 29 and 30 October. This is the seventh consecutive daily brief, which completes the first week this site set out to publish. What the week actually measured, in one line: the gap between announced data-centre capacity and delivered power is now being priced — in collateral at Exelon, in a 17-month lag at Digital Realty, in $42.4bn of negative free cash flow at Amazon — and the companies selling into it are the only ones with the cash. Where a figure is a company's own definition or a slide rather than an audited statement, we have said so. How this site works.