Texas has a cluster of data center projects sized between 60 and 74.9 megawatts. ERCOT's review threshold is 75. ERCOT has noticed. Pure coincidence, I'm sure. Everybody just loves the number seventy-four. This is The Data Center Daily for Monday, November 9th. Coming up: Exelon asks FERC to scrap PJM's three-year-forward capacity auction, ERCOT goes looking for projects that may have been shrunk to dodge its review, Florida regulators send Duke's large-load tariff back, and OpenAI puts its community promises in writing after last week's ballot losses. One tap on follow, and we'll be back in your ears before you know it.
The Duck Curve, on a FERC complaint filed last Tuesday:
Exelon asked the Federal Energy Regulatory Commission on November 3, 2026 to find PJM Interconnection's three-year-forward capacity auction schedule unjust and unreasonable and to replace it with a "prompt" auction held two months before each delivery year, starting with 2030/2031.
The docket is EL27-11. A prompt auction would move the 2030/2031 Base Residual Auction from May 2027 to around March 2030. Exelon wants fast-track handling and an order by January 31, 2027, with a refund effective date of January 1st. Remember who's filing. This is the buy side. Exelon's six utilities serve more than 10.7 million customers in PJM. The last two auctions cleared at the price cap and still fell short of the reliability requirement, by 6,517 and 6,831 megawatts. Exelon says later auctions are slated to run with no cap at all. So it's trying to get out from under the spring 2027 auction before the uncapped prices show up. The complaint goes after both halves of the 2006 settlement that set the three-year gap. On forecasting, Exelon says load adjustments in PJM's 2026 forecast equal 18.4 and 21.7 percent of summer peak four to five years out, when adjustments in the 2017 through 2021 forecasts were no more than two percent. On building, it cites a 116 gigawatt backlog at GE Vernova and Mitsubishi gas turbine lead times of five years or more. Its best witness is PJM itself: a May 2026 white paper saying the design's foundational assumption has been physically broken. When the grid operator writes your best line for you, that's a good day for the lawyers. Exelon does concede a prompt auction alone won't solve affordability. And its longer-term idea, from an October 30th investor deck, is state-directed planning and procurement of generation, with the capacity market providing residual support. PJM answered the next day, opposing fast-track processing and asking to push its deadline on the merits from November 23rd to December 21st. It argues legal consequences attach when sellers submit binding offers, not when planning parameters post, and it now expects to post the 2030/2031 parameters on or before March 1, 2027. Its board takes up the path forward for that auction on Tuesday, November 17th. The near-term line to watch: the December auction for 2029/2030 is outside Exelon's request and keeps its 337.83 dollar cap either way. Everything after that is now contested.
The San Antonio Express-News, on ERCOT's briefing to Texas regulators last Thursday:
Chad Seely, ERCOT’s senior vice president of regulatory policy, general counsel and chief compliance officer, told the Public Utility Commission on Thursday that there is nearly 10 gigawatts worth of projects classified as medium-size loads working with utilities to connect to the grid.
Medium loads run from 25 to just under 75 megawatts, and Seely said they're concentrated between 60 and 74.9. Under 75, a project can work directly with its utility and skip ERCOT's large-load approval. ERCOT has sent 132 requests for information to transmission and distribution providers for each project's size, location and energization date, to work out whether developers downsized to stay under the line. Everybody designs to a threshold. But ten gigawatts of medium load is a lot of load for the grid operator not to be studying, and Seely told commissioners it's a transmission planning and operational risk. He also said the findings could push the commission toward rules below 75 megawatts. So the gap may close right as people squeeze through it. The same briefing covered the bigger audit Governor Abbott ordered in August. After a first round in September to more than 257 developers, ERCOT is sending a second round to about 80 percent of them, on site control, long-lead equipment and financial security. They get five business days to respond. Five business days to prove you can actually get your transformers. Honestly, that's the right test. And Seely said himself that anyone found ineligible for Batch Zero will likely dispute it, so expect the complaint pile to grow. For scale: a separate questionnaire went to about 465 projects, and the 433 that responded requested 177 gigawatts, nearly twice ERCOT's record peak of 91.31 gigawatts, set in July. That's proposed capacity, not confirmed projects. The roughly 150-page audit report is due Thursday, December 10th, for discussion at the PUC's December 17th meeting. And in the Batch Zero complaint fight we've been following, the judge's pause runs until tomorrow, November 10th.
Environment and Energy Leader, on a Florida commission vote last Tuesday:
The Florida Public Service Commission voted Nov. 3 to deny Duke Energy Florida’s proposed large-load tariff. Regulators found the proposal excluded some customers covered by Florida’s 2026 law and that Duke had supplied insufficient evidence that its existing rates and proposed financial protections would recover each qualifying customer’s full cost of service.
Take the two problems in order. Florida's Senate Bill 484 defines a large-load customer by an anticipated monthly peak of at least 50 megawatts at one location, measured as the highest 15-minute average. Duke applied that threshold to firm demand. Commission staff said in an October 22nd recommendation that could exclude a customer that reaches 50 megawatts through interruptible service. That's the customer every planner says they want. The curtailable one. You can't write the flexible data center out of your tariff and then give speeches about load flexibility. The second problem is cost recovery. Duke's April petition proposed a 20-year minimum service term and advance payment of the estimated cost to extend service, with qualifying customers taking service under existing rate schedules. But the state law also requires utilities to account for generation and transmission costs attributable to each large-load customer, and to protect other ratepayers from nonpayment risk. Paying for the line to your fence isn't paying for the plant behind it. A twenty-year term sounds tough until you notice the rate underneath it was designed for somebody else. Duke now has to file a large-load rate schedule and evidence that qualifying customers would cover their costs. Staff recommended letting it refile within 10 business days of the written order. It's the state-level version of the cost-shift question FERC put to every RTO in June.
Grace Okonkwo, in an analysis published Thursday, on what OpenAI's new National Community Compact promises:
OpenAI says it will pay the full cost of the electricity its facilities use. It will also cover the grid upgrades that power providers need to serve them. It promises to throttle consumption during peak hours and send power back to the grid.
The compact also commits OpenAI to closed-loop or low-water cooling, to publishing water and electricity use for each site, and to signing no nondisclosure agreements with public officials, with exemptions for trade secrets, security and competitively sensitive information. A companion policy blueprint asks lawmakers to write the same commitments into law. Here's the catch Okonkwo flags. The compact covers only sites OpenAI owns and operates, and much of Stargate is built and held by partners such as Oracle. So the written promise doesn't automatically follow the partner-built campuses, which is a big chunk of the program. On timing, it came two days after the November 3rd vote, when, per Ballotpedia plus results gathered by CNBC, voters backed restrictions or rejected pro-development measures in all but one of 41 local ballot measures. Chris Lehane, OpenAI's chief global affairs officer, told reporters the blueprint had been in the works since August and that the timing had nothing to do with the result. In the works since August, out the Thursday after the rout. Sure. Bisnow reported OpenAI abandoned the Narvik site in Norway and canceled an 800 megawatt Abilene expansion with Oracle. Abilene is still the only Stargate campus even partly running. Lehane says the goal is momentum for when the new Congress sits in late January.
One pointer before we close. If your week runs through models as much as megawatts, try Lantern AI Daily Briefing: the top AI news for engineers, founders and investors, every weekday, sorting real capabilities from demo hype, quickly. Somebody has to explain what all those GPUs are for.
Three dates for the board: the Batch Zero pause lifts tomorrow, PJM's board meets on the 17th, and ERCOT's audit report is due December 10th. We're back tomorrow, Tuesday, November 10th. And if you're sitting on a 74.9 megawatt project, maybe keep it quiet this week. From Sarah and Matt, thanks for listening to The Data Center Daily, a Lantern Podcast.