Six point eight gigawatts. That's the shortfall PJM's reliability backstop was designed to cover. Jefferies now thinks the final number could land near 630 megawatts. Ninety percent of a capacity crisis, gone in one analyst note. Welcome to The Data Center Daily. It's Wednesday, November 4th. We said yesterday we'd turn to what Tuesday's votes mean for permits, and we'll do that once there's reporting solid enough to quote. This morning: the PJM backstop may be shrinking, Aurora puts a price on the industry's forecasting error, the hyperscalers line up more than 131 billion dollars of new financing, and Microsoft's Indiana campus clears a groundwater vote.
Ethan Howland, writing in Utility Dive:
PJM utilities, such as PPL Electric, FirstEnergy and AES Ohio, recently lowered their overall, near-term large load forecasts by 6.5 GW, the analysts said in an Oct. 28 note, citing recent filings to PJM’s Load Analysis Subcommittee. That reduction could reduce PJM’s 2027 load forecast, which would affect the grid operator’s backstop procurement target, they said.
We covered PJM's refiled backstop on Friday, October 30th. FERC had rejected most of the first design in late September, and the revision, filed Thursday, October 29th, starts the process March 1, about seven months behind the original plan. What's new is the size. Jefferies says that if the 6.5 gigawatt cut flows into the 2027 forecast, the procurement could fall to roughly 630 megawatts from the 6.8 gigawatt target. Bilateral contracts between developers and suppliers could shrink it further. So which number do you believe? On Monday the Load Analysis Subcommittee discussed work from Charles River Associates, which PJM hired for an independent data center forecast, and CRA found PJM could be about 11.5 gigawatts short by 2030. One note says the near-term hole nearly closes, another says the end-of-decade hole is bigger than the original. That's not a forecast. That's a range you could run a transmission line through. It does line up with Exelon's large-load submission, which we covered October 31st, where high-probability load fell from 19 to about 12 gigawatts. On mechanics, Maria Scheller at ICF told Utility Dive the revision sharply reduces risk for providers of last resort, because an obligation and its collateral can now move with a large customer when it switches load-serving entities. Can move, if the states do their part. Scheller's own caveat is that regulators may need new rules on customer-specific allocation, security, and what happens if the load shrinks, defaults or disappears. Direct risk to residential customers goes down. It doesn't go to zero. Collateral costs, regulatory lag, a developer walking away, all of that still finds a bill to land on. Timeline, pending FERC approval: the bid window closes March 28, the final target is set April 28, and results post May 10.
Latitude Media, reporting from Aurora's Energy Transition Forum in Manhattan last week:
According to Aurora’s North America managing director, Oliver Kerr, that forecasting gap could cost ratepayers anywhere between $79 billion to $624 billion in extra capital expenditure costs — and that’s before transmission and distribution costs, which are rising nationwide, especially in PJM, where 3.7 GW of ready-to-build projects have withdrawn specifically because of the price of transmission.
The gap in question: grid operators expect roughly 63 gigawatts more demand by 2030 than Aurora's baseline. Against a US peak just under 760 gigawatts, that's about 8 percent of the entire system. Published data center forecasts run from around 22 gigawatts of growth by 2030 at the conservative end to 94 to 131 at the top, and Aurora has Texas, Virginia and Pennsylvania taking about 45 percent of data center load growth. Here's the part I'd tape to every planner's monitor. Julia Hoos at Aurora says about 58 percent of planned projects are being built by first-time developers, and historically only about 15 percent of planned projects ever materialize. Fifteen. Latitude also reports PJM's actual demand so far in 2026 has come in 21 percent under its 2025 forecast. The risk cuts both ways. Overbuild, and ratepayers carry capital nobody needed. Underbuild, and reliability suffers. Equipment isn't getting easier either. Hoos expects the bulk power system executive order, which blocks grid equipment from 24 countries, most notably China, to push prices up once implementing rules arrive in late December. China provides about 65 percent of US transformer equipment while accounting for only 8 percent of the spending. Transformer waits already run as long as five years, demand is expected to double next year, and the low-cost supply is about to be off the table. Evercore's Jonathan Kaufman called consolidation inevitable as inexperienced developers get forced out. Honestly, some of that 63 gigawatt gap probably leaves with them.
Georgia Butler, writing for DCD:
As per an SEC filing, Amazon's facility is part of its supplier financing arrangements for technology infrastructure purchases and can be tapped into between Q1 2027 and Q3 2028.
That's an up-to 80 billion dollar loan facility, variable rate, each borrowing maturing in five years. It's the largest piece of more than 131 billion dollars in debt and equity that Amazon, Google, Microsoft and Oracle have disclosed in roughly a week. We covered Microsoft's 25 billion dollar bond sale on October 30th; DCD puts the rates between 4.85 and 6.35 percent, with closing expected today, and Microsoft now projects 170 billion in 2027 spending, down from 175. Oracle announced 20 billion on Monday, half common stock and half floating-rate senior notes. Google is offering 6 billion euros of bonds. Read Amazon's wording again. Supplier financing for infrastructure purchases, varying with the extent and timing of purchase orders. That's borrowing that scales with the server orders. Amazon spent more than 77 billion in capex last quarter and still says the year lands around 220. Fine. But when the debt is sized to the orders, it tracks the buildout whether or not the revenue keeps pace. For balance, these are investment-grade borrowers, and DCD draws the contrast with the neoclouds. It cites Fortune reporting that the money well is beginning to dry up, and notes CoreWeave's bond yields hit 15 percent after issuing at 9.67, which the FT called distressed debt. That's the spread I'd watch. The hyperscalers still borrow in the single digits. If the neoclouds are paying fifteen, the risk didn't disappear. It just moved down the stack.
Peter Mwaniki, writing for Construction Review Online:
The St. Joseph County Drainage Board approved the dewatering plan on November 2, allowing groundwater to be pumped from the development site and discharged into nearby Willow Creek and Juday Creek.
This is Microsoft's six-building campus on about 929 acres in Granger, Indiana. Discharge can run up to 200,000 gallons an hour from three wells operating at different times between mid-December and early April. It's construction dewatering for utility work, not the campus's operating water supply. Separately, Indiana's environmental agency has a draft air permit out for public comment through November 25th, with potential emissions it preliminarily puts below major-source thresholds. The neighbors aren't comforted. Steve Francis of the Juday Lake Estates homeowners association told WSBT the project could drain nearly 192 million gallons from one area of the site into a creek that's prone to flooding. A drainage board member told WNDU officials don't expect flooding. People on private wells want the monitoring plan, and Mwaniki notes frequency, baselines and response procedures are still open questions. The number I wanted isn't there. The county says Microsoft will pay for the electrical upgrades, with AEP involved, but no megawatt figure or substation capacity is public, and the air permit notice doesn't give the number or capacity of backup generators. For cooling, Microsoft describes a closed-loop system using fluid brought in from outside the county and circulated for about two years. Six buildings, the first three targeted for mid-to-late 2029, and nobody will put a megawatt on it. I suspect somebody at AEP has a pretty good guess.
If you're close to those PJM utility filings, or you have a better read on that 630 megawatt figure, we want to hear it. Write to us at datacenterdaily at lantern podcasts dot com. Especially if you think Jefferies has it wrong. Those are my favorite emails.
And that's our Wednesday. Tomorrow, Thursday, the Texas PUC's open meeting is set to take up how it handles the Batch Zero complaints. And we'll start working through what the vote counts mean for Texas permits. The Data Center Daily comes to you from Lantern Podcasts. I'm Sarah, here with Matt. Back with you tomorrow.