In partnership with
⚡ THIS WEEK'S SIGNAL
The cost-allocation fight had its biggest week of the year on August 17, when all six grid operators had to answer FERC on who pays for large-load upgrades. Here is what none of those filings can do: move a transformer up a factory queue. Substation transformer lead times have pushed past 160 weeks. And the federal program named for speed-to-power funds conductors, not transformers. Money is the argument this year. Time is the constraint.
━━━━━━━━━━━━━━━━━━━━━━━
📰 MAIN STORY
The filings landed. On August 17, PJM, MISO, SPP, CAISO, NYISO, ISO-NE and their transmission owners hit the 60-day deadline in FERC's June 18 Section 206 show-cause orders, which directed every organized market to justify or reform the rules governing how large loads interconnect. We called that a clock when it started (FERC Didn't Pick Winners, It Started a 60-Day Clock, June 23). The clock has now run. What it produced is a national argument about cost causation — who pays for the network upgrade a data center triggers — arriving in the same year that the physical equipment those upgrades require has become the actual binding constraint.
The players want different things and are optimizing accordingly. Hyperscalers want energization dates they can underwrite; on Q2 2026 earnings calls, Microsoft, Alphabet and Meta shifted their framing away from aggregate capex toward time-to-energy and the speed of converting infrastructure into revenue-generating compute. Utilities want load growth that spreads fixed costs without exposing them to stranded assets. Equipment makers want order visibility long enough to justify a factory. And roughly 80% of U.S. power transformer supply is imported, which means the fastest-moving variable in the whole system sits partly outside anyone's regulatory reach.
Here is the contrast. In a single quarter, GE Vernova's Electrification segment booked $2.4 billion in equipment orders to support data centers — more than the entire prior year. Over roughly the same window, federal research indicates U.S. residential electric rates rose an average of 10.2% between March 2025 and March 2026. One of those numbers is a backlog. The other is a bill. They are not causally linked in any clean way, and nobody should pretend otherwise. But they are drawing on the same constrained supply of steel, copper, factory slots and skilled winding labor, and only one of those two parties gets to reserve a delivery slot three years out.
The data. Substation transformer lead times have stretched from roughly 140 weeks in 2023 to more than 160 weeks in 2026. Wood Mackenzie's survey work put power transformers around 128 weeks and generator step-up units around 144 weeks, with demand for GSUs up 274% and substation power transformers up 116% since 2019, while unit prices rose 77% for power transformers and 45% for GSUs. Manufacturers are responding: Hitachi Energy has committed more than $1 billion across the continent, including a $457 million plant in South Boston, Virginia, positioned to be the nation's largest large power transformer facility by 2028; Siemens Energy is building its first U.S. large power transformer plant in Charlotte; and Eaton has committed $340 million to a three-phase facility in South Carolina. Most of that capacity produces its first unit in 2027 or later. Meanwhile U.S. data center capacity is projected to scale from roughly 24 GW to 100 GW between 2026 and 2030, with the data center electrical equipment market growing from about $20 billion to $65 billion.
Scenario A — the tariffs get specific about time. If the Section 205 packages that follow these show-cause responses start conditioning service on documented equipment coverage rather than just deposits and minimum-take percentages, the effect is a real filter. Infrastructure investors underwriting speed-to-power assets would see project pipelines shrink and quality rise, because a developer without a factory slot can no longer hold a queue position with a letter of credit. Utility program managers get a defensible basis to sequence workplans. Ratepayers get fewer speculative upgrades entering rate base.
Scenario B — the tariffs stay purely financial. If the reforms land as collateral, minimum bills and exit fees without touching long-lead procurement, the cost-shifting problem gets solved on paper while the schedule problem migrates. Utilities pre-order speculatively to protect energization commitments, carrying the interest cost of equipment ordered years ahead of a customer who may not arrive. That carrying cost lands in rate base and gets recovered from everyone. The hyperscaler is protected by contract; the ratepayer is exposed by timing.
One thing to watch: the Commerce Department's Section 232 finding that laminations, stacked cores and wound cores are being imported in quantities that threaten national security — a finding that has not yet produced trade action. If it does, every imported transformer core reprices at once.
━━━━━━━━━━━━━━━━━━━━━━━
🧾 DEAL DECORDER
The Slot Reservation Agreement — GE Vernova's Q2 2026 Disclosure
Who signs what. A slot reservation agreement, or SRA, between GE Vernova and a power customer. It is not an equipment order. It is a deposit-backed contract to hold a place in the manufacturing line, typically three to five years out, and it sits outside reported backlog until it converts. GE Vernova is now taking reservations for 2031 delivery.
Who fronts capital, who gets it back. The customer pays a deposit years before a unit exists. It converts into a firm order and into the purchase price, or it does not convert. Think of it as a restaurant reservation with a non-refundable deposit, for a table in 2031.
Who eats the loss. The reserving party. If the load never arrives, the deposit is forfeited and the manufacturer resells an oversubscribed slot.
Precedent. In Q2, GE Vernova signed 20 GW of new gas equipment contracts — 18 GW of them reservations, 2 GW actual orders — and ended the quarter with 53 GW of backlog against 63 GW of reservations. More than half of the headline number is deposits. Turbines have this instrument in public disclosure because a listed company must report it. Transformers have the same scramble with none of the visibility.
━━━━━━━━━━━━━━━━━━━━━━━
Reply to everything. Edit nothing.
Your inbox is full. Slack is piling up. Client messages need a response yesterday. Typing thoughtful replies to all of it takes hours you don't have.
Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Speak like you would to a colleague — tangents and all — and get polished output. Emails, Slack, LinkedIn, WhatsApp, whatever's open.
89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.
━━━━━━━━━━━━━━━━━━━━━━━
⚡ QUICK HITS
SpaceX Bought $295 Million of Tesla Megapacks in One Quarter. SpaceX disclosed $295 million in Megapack purchases for April–June and $329 million for the first half, with its AI segment's nameplate compute reaching 1.4 GW at quarter-end, up from 1.0 GW in Q1. Batteries are being used here as a substitute for grid firmness the utility cannot deliver on schedule, which is the transformer story wearing a different jacket. [SEC filing coverage — CNBC]
24 States Now Have an Approved Large-Load Tariff. EEI's tracker, updated August 14, puts 24 states with at least one approved large load tariff and six more pending. If you run projects at a utility in one of the remaining states, your cost-allocation framework is now the exception rather than the rule, and your next large-load applicant knows it. [EEI Large Load Projects and Tariffs]
Meta Is Expanding Hyperion Toward 5 GW. Meta described its northeast Louisiana Hyperion campus as expanding into a 5 GW AI supercluster, and separately committed more than $9 billion to a 1 GW campus in Sturgeon County, Alberta. Two announcements, two interconnections, and both of them need transformers from the same order book everyone else is queued in. [Data Center Knowledge]
The Tariff Math on Transformers Changed in April. A April 2, 2026 proclamation effective April 6 moved most derivative products to 25% on the full customs value, replacing the prior 50%-on-metal-content approach, while goods that are wholly or almost wholly metal stay at 50%. For an import-dependent category, that is a repricing of landed cost that eventually reaches a rate case. [Brownstein analysis]
Residential Rates Rose 10.2% in Twelve Months. Federal research indicates residential electric rates rose an average 10.2% nationally between March 2025 and March 2026, with some states considerably higher. Very little of that is directly attributable to data centers yet — the recovery lag runs quarters to years — which is precisely why the tariff decisions being made this fall matter more than this year's bill. [Consumer Reports]
━━━━━━━━━━━━━━━━━━━━━━━
🧭 THE EXPOSURE MAP
If you hold equities. The segment of the value chain structurally strengthened here is grid equipment manufacturing and the engineering firms that sequence it, where scarcity is currently expressed as backlog rather than price alone. The segment under pressure is any business model that monetizes speed-to-power without controlling equipment supply. The reporting events that would confirm or contradict this: Q3 2026 earnings calls from transformer manufacturers and large utilities, October through November, and specifically whether stated lead times extend or compress.
If you run projects at a utility. The question your organization should be able to answer and probably cannot: for every large-load commitment with an energization date inside 36 months, is there purchase order coverage on each long-lead unit, and what is the contractual remedy if the date slips? That answer belongs in your workplan review before it belongs in a customer meeting.
If you set or influence policy. Cost causation just became the organizing federal principle for large-load interconnection. The precedent still unwritten is whether "cost" includes the carrying cost of equipment ordered speculatively. Next venue: CAISO's Section 205 filing and each RTO's follow-on package.
If you pay a utility bill or buy AI compute. The mechanism is rate base. Equipment bought years before a customer arrives accrues financing cost that enters a rate case whether or not the load shows. It reaches your bill through a transmission or distribution rate component, typically 18 to 36 months after the spending decision, and it will not be labeled.
━━━━━━━━━━━━━━━━━━━━━━━
🔧 TOOL / RESOURCE OF THE WEEK
CRS Report R48933, Electricity Distribution Transformers: Supply, Tariffs, and Policy Options. A free, non-partisan Congressional Research Service analysis of transformer supply, domestic manufacturing capacity, and the tariff regime affecting cores and laminations. Retail investors holding utility and industrial names through index funds will find this the clearest available explanation of the physical constraint sitting underneath the capex story they already own. → congress.gov/crs-product/R48933
━━━━━━━━━━━━━━━━━━━━━━━
💬 CLOSING THOUGHT
Cost allocation is a solvable problem. Commissions have been arguing about who pays for a shared asset since the 1930s, and they will land somewhere workable. Lead time is a different kind of problem, because no order changes a factory's throughput this quarter. The optimistic read is that the manufacturing buildout is real and 2028 looks better than 2026. What would need to be true for that to hold is steady demand signal — the thing a wave of cancelled projects would take away.
I keep wondering whether equipment coverage will ever be disclosed the way queue position is. Will that become a normal filing, or stay something you only learn in a room?
