You come east out of the Inland Empire and the land squeezes. San Jacinto on your right, San Gorgonio on your left, both over ten thousand feet, the highway threading the slot between them. Air gets funneled through that slot and dumped into the desert. Most afternoons. Most of the year. That is the entire reason anything got built here.

Round the bend at Whitewater and the turbines are just there. Up the ridges, out across the flats, scattered along the alluvial fans. Some spinning hard. Some dead still. A few obviously dead and left standing anyway, which nobody ever mentions in the brochure.

It is the only place I know where you can see the grid. Normally the grid hides in conduit and easements and somebody’s substation behind a chain-link fence. Here it is on a hillside, moving, forty miles of it.

And the thought that arrives, every time, is: that is a lot of power.

It is not.

Four decades of turbines, and the number nobody quotes

San Gorgonio Pass holds about 667 MW across roughly 667 turbines. The match is a coincidence, not a ratio, and I mention it only because it will stick in your head better than either number alone. Third-largest wind resource in California, behind Kern County’s 3,254 MW and Solano’s 1,031.

It is also not a wind farm. It is about twenty-five separate projects, different owners, different vintages, different contracts, which is why you cannot pull its output the way you pull a plant’s. First turbines went up around 1982. At the 1987 peak there were more than 4,200 of them, small and fast and genuinely loud. By December 2021, about 1,220. Today, about 667.

That collapse is the good news. It is repowering, and the trade is brutal in the right direction.

FIGURE 1 ·

Repowering swapped thousands of turbines for dozens.

Six hundred twenty machines came off Mesa and Alta Mesa in 2023 and 2024. Thirteen went back on. Coachella Hills gave up 363 for 17. Every new turbine eats fifteen to fifty of the ones it replaced, and the hillside empties out a little more each time. This is what actual progress looks like in hardware, and it took forty years.

Now the number nobody puts on the postcard. Apply California’s 2024 fleet-wide wind capacity factor, about 29%, to 667 MW. You get roughly 1,700 GWh a year. Be generous with that, because the pass is an old site in a worse wind class than Tehachapi, so the real figure probably sits under it.

Four decades. Four thousand turbines up, three thousand down. Seventeen hundred gigawatt-hours.

Hold that number.

Forty miles down the road, somebody ran the same math

Keep driving. The road drops the whole way. Date palms, then table grapes, then the valley floor hits sea level and keeps going.

At 52nd and Fillmore in the city of Coachella, a Riverside developer called Stronghold Power Systems drew up the Coachella Valley Technology Campus. Six buildings, three million square feet, somewhere between 240 and 450 acres of working farmland depending on which version of the plan you read. Phase one at 270 to 300 MW. Full buildout at 600.

Run 270 MW at an 85% load factor and you get about 2,010 GWh a year.

Phase one. One phase, of one campus, would have eaten more electricity than every turbine in the pass produces.

FIGURE 2 ·

One campus would have out-consumed the pass that powers it.

At full buildout it is 4,470 GWh, which beats the pass, the 550 MW Desert Sunlight solar farm, and the 800 MW Sentinel gas peaker put together.

Look at Sentinel for a second. Eight hundred megawatts of nameplate, more than the wind farm, and it produces a fraction of the energy, because it is a peaker and sitting idle is its job. Nameplate is a promise. Output is what showed up. People conflate those two constantly, usually on purpose, and one drive through this valley cures you of it permanently.

FIGURE 3 ·

A wind farm runs flat out less than a third of the year. A data center runs flat out almost all of it.

That gap is the whole business. Wind is free. Turning wind into electricity on a schedule somebody else chose is not free, and those are different problems that get filed under the same word.

Here is the detail that should stick with you, though. The campus was barely going to touch the grid. Fuel cells, battery storage, diesel for emergencies, hanging off Imperial Irrigation District distribution, and in February the city signed an agreement to stand up its own municipal electric utility around the thing. The load was going to show up carrying its own generator, on land that currently grows food, in a valley whose entire public identity is those turbines back at the pass.

Then the city killed it

June 6, the council passed a 45-day moratorium, told staff to draft a permanent ban, and tore up the utility agreement. July 9, they stretched the moratorium to ten and a half months. First reading August 12. Second reading August 26. Done.

The ordinance kills large-scale AI data centers citywide and leaves small telecom and storage facilities alone. Stronghold’s CEO said phase one would have thrown off about $54 million a year in public revenue, schools and water district included. The mayor said it came from the will of the people. Both of those are probably true, which is the uncomfortable part.

Coachella is the second California city to ban data centers outright. Monterey Park got there first, by ballot, in June. Ten or so California cities now have a moratorium or a ban on the books, and that list includes Desert Hot Springs and Indio. Three cities in this one valley have said no. Perris, up in the Inland Empire proper, is thinking about it.

The constraint moved and nobody updated the model

For twenty years the story of new load out west has been the interconnection queue. Get in line. Wait for the study. Wait for the upgrade. Wait for somebody to build the transmission you need. Every consultant deck ends there.

Stronghold designed around all of it. Fuel cells, batteries, its own utility, its own distribution deal. It did precisely the thing everyone in this industry says developers ought to do, which is bring your own power and stop making the grid your problem.

Lost anyway. In a room with a hundred people in it, on a land-use vote, forty miles from the biggest pile of renewable generation in Southern California.

The California Energy Commission expects data center load on the CAISO grid to climb 1.8 GW by 2030 and 4.9 GW by 2040. That demand is real and it has to physically land on dirt somewhere.

FIGURE 4 ·

One banned campus was a third of California’s entire 2030 data-center increment.

So the question stopped being can we get power to this parcel. It is now will the people who live near this parcel let us. Different question, different clock, and it fails in ways you cannot model off a tariff. You can win every technical argument in the CAISO stack and still lose on a Wednesday night to a neighborhood petition that picked up 689 signatures in five days.

Nobody’s siting model has a field for that.

Drive it yourself if you get the chance. Forty-four years of turbines at one end of the valley. At the other end, a city that just told 600 MW to go be somebody else’s problem.

Where Stronghold lands next. Banned projects do not evaporate, they relocate. Watch unincorporated Riverside County and the Imperial Valley, which has its own fight and its own moratorium history.

Perris. It is weighing a restriction, and it matters more than Coachella did. I-215, warehouse stock, existing substations, a labor base. It was always the better site. If the ban pattern jumps from desert resort towns to the logistics corridor, California’s siting map gets small in a hurry.

Whether bring-your-own-generation survives contact. Behind-the-meter gas and fuel cells are the industry’s answer to interconnection delay. Coachella is the first clean test of that answer against an objection that was never about the grid.

Sources: CalWEA California wind fast facts (2024 figures, US Wind Turbine Database compilation, March 2026); San Gorgonio Pass repowering and turbine counts per public project records; Desert Sunlight Solar Farm reported generation, 2015 to 2018; CPV Sentinel Energy Project design specifications; City of Coachella council actions of June 6, July 9, August 12 and August 26, 2026, and the February 2026 municipal utility development agreement; contemporaneous KVCR, NBC Palm Springs, KESQ and Patch reporting; Datacenter Map project listing; California Energy Commission data center load forecast as published by CAISO, January 2026. Annual energy figures for San Gorgonio Pass, Sentinel and the Coachella campus are derived from nameplate capacity and stated capacity or load factors, not metered output.

Dynawatt is an independent brief compiled from public sources (ISO/RTO interconnection queues, FERC notices (Federal Register), ERCOT Large Load Working Group, PJM stakeholder committees, NRC ADAMS (reactor licensing), and DOE Loan Programs Office / EDF announcements). It is informational, non-advocacy, and not investment advice.

Two things happened last week at the same address in Covert, Michigan.

On August 28 the NRC gave Holtec permission to start digging for two reactors that do not have a construction permit. Two days later, on the same site, the company began loading fuel into an 800 MW reactor that has been shut down since 2022.

Both are milestones. Only one of them is close to producing electricity, and it is not the one with the new paperwork.

Permission to dig, not to build

The NRC granted SMR, LLC and Palisades SMR, LLC, both Holtec subsidiaries, an exemption from 10 CFR 50.10(c) on dockets 50-616 and 50-617. It lets them build support-of-excavation walls at the Pioneer Units 1 and 2 site and leave them there permanently, before the agency issues a limited work authorization.

Support-of-excavation walls are what keep a deep hole from becoming a shallow one. Diaphragm walls, soldier pile soil-mix walls, perimeter cutoff walls. When the plant is done they get retired in place and the gap between them and the real structures is backfilled to grade. They have no function in the finished facility. They hold back dirt for a few years, and then they are dirt.

Under 10 CFR 50.10 that still counts as construction, so it normally waits for the LWA. Holtec asked for relief on December 31, 2025, filed alongside Part 1 of its phased construction permit application, supplemented the request in July, and got an answer eight months after asking.

The finding is worth reading closely. Special circumstances exist, the NRC wrote, because compliance “would result in undue hardship or other costs that are significantly in excess of those contemplated when the regulation was adopted.” Building temporary walls, ripping them out, and building permanent ones costs money and schedule, and the rule was not drafted with a 2030 SMR target in mind.

That is a defensible call, and the agency does not dress it up. The cost-and-schedule reasoning is stated as cost-and-schedule reasoning. The safety finding is separate and narrow: the walls carry no safety function once the plant exists, and staged geologic mapping during excavation preserves the NRC’s ability to check the site later. Environmental assessment issued August 25, finding of no significant impact.

Two SMR-300 units, roughly 300 MW each. Part 2 of the construction permit application is due no later than eighteen months after the Part 1 filing. Holtec calls the program Mission 2030.

At the same address

Over the weekend of August 30, Holtec began loading 204 fuel assemblies into the Palisades reactor. About 800 MW. Commercial operation since 1971, shut down in May 2022, and now the first attempt in American history to bring a commercial reactor back out of decommissioning, underwritten by a DOE loan guarantee of up to $1.52 billion.

Palisades has not gone critical. It has not synchronized. Holtec missed an end-of-2025 target and a February 2026 target, and now declines to name a date at all, saying the plant restarts when it is ready for long-term operation. The contract to deliver power lands in March 2027. The refurbishment has been running more than two years.

So: regulatory permission to dig a hole for two reactors that do not exist, and fuel going into a reactor that has existed for fifty-five years and still is not making electricity.

Wyoming already ran this play

None of this is new. It has a control case, and the control case has already reported out.

In May 2025 the NRC granted US SFR Owner, LLC, a TerraPower subsidiary, an exemption under 10 CFR 50.10(a)(1)(iv) for Kemmerer Power Station Unit 1 in Lincoln County, Wyoming. Same regulation, same idea: clear the non-safety work ahead of the permit so the schedule does not sit idle waiting on a docket.

It worked. TerraPower broke ground on non-nuclear work in June 2024. Exemption in May 2025. Construction permit in March 2026, the first the NRC has issued for a commercial advanced reactor. Official construction start April 23, 2026.

Twenty-one months from groundbreaking to permission to build. The front end did its job.

The target for first power is now 2030 or 2031. The original target was 2028.

The clock that binds

MISO’s April forecast puts footprint peak load at 121 GW in 2025 and 163 GW in 2035, up 35%, with data centers taking a fifth of the region’s electricity by 2030 and a quarter by 2040 in its mid case. It expects 8 to 14 GW of data center capacity to energize in 2026 and 2027 alone. In August a single 2,015 MW large-load request entered the MISO queue with no customer named on the record.

Against that, in Covert: about 800 MW that was already built, already licensed, already standing on the lakeshore with a federal loan behind it. It is not on the system.

The permitting story is real. Exemptions are being granted, permits are being issued, and the NRC is moving on advanced reactors faster than it has in decades. The Palisades exemption is a small, sensible piece of that. But nothing in the last two weeks suggests the binding constraint on firm capacity is a federal docket. The constraint is fuel handling, welding, steam generator inspections, operator qualification, and the ordinary grinding difficulty of making a large machine run correctly.

Permitting is the fast part.

That is not an argument against permitting reform. It is a warning about what reform actually buys, which is months at the front of a project whose slippage happens at the back. Anyone building a 2030 load-and-resource balance on announced nuclear COD dates should price the difference between a permit and a plant.

Palisades criticality. The next real milestone is a sustained chain reaction, then synchronization. If neither happens by the end of October, the March 2027 power contract gets tight.

Pioneer Part 2. The second half of the phased construction permit application is due to the NRC no later than eighteen months after the December 31 Part 1 filing, so by the end of June 2027. Watch whether it lands early. Mission 2030 has no slack in it.

The Pioneer EIS. The NRC noticed its intent to prepare an environmental impact statement in June. The draft is the next substantive gate on the limited work authorization, and it is a longer pole than any exemption.

Sources: Federal Register, “SMR, LLC; Palisades SMR, LLC; Pioneer Units 1 and 2; Exemption” (August 28, 2026), and the associated environmental assessment and finding of no significant impact (August 25, 2026, ML26230A249); NRC dockets 50-616 and 50-617, including the February 27, 2026 phased construction permit and limited work authorization notice and the June 15, 2026 notice of intent to prepare an environmental impact statement; Federal Register, “US SFR Owner, LLC; Kemmerer Power Station, Unit 1; Exemption” (May 15, 2025), docket 50-613; American Nuclear Society Nuclear Newswire; Holtec International; US Department of Energy Loan Programs Office; MISO long-range load forecast, April 2026.

Dynawatt is an independent brief compiled from public sources (ISO/RTO interconnection queues, FERC notices (Federal Register), ERCOT Large Load Working Group, PJM stakeholder committees, NRC ADAMS (reactor licensing), and DOE Loan Programs Office / EDF announcements). It is informational, non-advocacy, and not investment advice.

One request, two gigawatts

A 2,015 MW large-load request entered MISO’s interconnection queue this week at queue-entry stage, which makes it eligible for study. The record doesn’t name a customer. At that scale the candidates are a hyperscale campus or a heavy industrial site, and the digest reads it as data-center demand, but nothing in the filing says so.

Take the number seriously and the stage lightly. Two gigawatts of firm load is a large city’s worth of demand arriving as a single request, and MISO is not ERCOT: it has no equivalent of the Large Load Working Group process, so a request this size lands in a queue built mostly for generators. What happens next is a study, and the study phase is where most of the attrition in any queue happens. Queue entry establishes a position and a date. It does not establish that anything gets built.

It is also the only demand-side item on the wire this week. Everything else was supply, and most of the supply was paperwork.

Holtec’s two-track week

The NRC concluded Section 106 consultation on Palisades SMR’s exemption request for Pioneer Units 1 and 2 in Van Buren County, Michigan, dockets 50-616 and 50-617, with a determination that no historic properties are affected. That determination went out as individual letters to twenty-seven tribes, which is why it accounts for most of this week’s item count. One regulatory action, twenty-seven records.

The same week, Holtec submitted Revision 1 of its SMR-300 Instrumentation and Control Architecture Licensing Topical Report to the NRC, a generic design submission covering the 300 MW unit.

Those two filings are the same company working two levels of the licensing stack at once. The topical report is generic design review, the kind of document that gets approved once and then referenced by every subsequent application. The Section 106 clearance is site-specific environmental review for the two units Holtec wants to build at Palisades. Running both concurrently is the schedule-compression play, and the phased construction permit with limited work authorization that Palisades SMR filed in January is a second one. Clearing tribal consultation with no adverse finding also removes one category of contention from a proceeding that will attract others.

Worth noting for anyone reading queue data: Palisades is a restarted pressurized water reactor, not a coal site. Some summaries of these filings describe it as a coal-to-nuclear conversion. It isn’t.

The rest of the reactor paperwork

The NRC’s Advisory Committee on Reactor Safeguards is reviewing the construction permit application for Long Mott Generating Station, with a subcommittee meeting scheduled for September 1 and 2. Long Mott is X-energy’s Xe-100 project at Dow’s Seadrift site in Calhoun County, Texas, not South Carolina as some feed summaries have it. ACRS review is a late step, close to the Commission’s decision, which makes the September meeting the more informative date in this week’s set.

Oklo transmitted slides and supporting material from July pre-application meetings on its fuel recycling facility draft license application, filed August 19. That is the front of the funnel, years from a decision, and it is upstream infrastructure rather than generation. Track it as a supply-chain item: fast reactors need fuel, and the fuel needs a facility.

Add it up and none of this week’s nuclear activity puts electrons on a wire before the 2030s. That is not a criticism of the filings. It is the reason the 2 GW request at the top of this issue is the item that matters.

Queue movement, and a caution about queue records

The Menifee Power Bank, 680 MW of battery storage in Riverside County, entered the CAISO queue. Southern California is the constrained end of that system, so storage entering there is worth a note even at study stage.

Two ISO-NE records also changed status this week in a way the queue data reports as reaching commercial operation: a 716 MW combined-cycle unit in Essex County, Massachusetts, and a 135.5 MW wind project in Aroostook County, Maine. Both descriptions also say the projects had been withdrawn from the queue, which is contradictory, and both correspond to plants that have been operating for years. Read those as record housekeeping rather than new capacity. Queue databases log the administrative event, not the physical one, and the two come apart more often than the feeds suggest.

What to watch

The MISO study. A 2 GW request generates a system impact study, and that study is where the number either survives or gets cut. Watch for the interconnection customer to be named, which usually happens before the study concludes.

Long Mott’s ACRS meeting, September 1 and 2. ACRS is the last technical review before the Commission acts on a construction permit. Whatever the subcommittee flags is what the schedule turns on.

Whether Palisades SMR converts. Section 106 is cleared and the EIS scoping opened in June. The construction permit decision is the milestone, and the phased structure means work could start before the full permit issues.

A second large-load request in MISO. One 2 GW entry is a customer. A second inside a quarter would say the queue design is about to become the story, the way it already has in ERCOT and PJM.

Sources: MISO interconnection queue (large-load request, 2,015 MW); NRC ADAMS (Palisades SMR, LLC, Pioneer Units 1 and 2, dockets 50-616/50-617, EPID L-2026-LLE-0011; Holtec SMR-300 I&C Architecture Licensing Topical Report Rev. 1; Long Mott Energy ACRS subcommittee notice; Oklo fuel recycling facility pre-application transmittal); CAISO interconnection queue (Menifee Power Bank, 680 MW); ISO-NE interconnection queue. Long Mott siting and Palisades plant type verified against NRC project records; both differ from the feed summaries.

Dynawatt is an independent brief compiled from public sources: ISO/RTO interconnection queues, FERC notices, ERCOT Large Load Working Group, PJM stakeholder committees, NRC ADAMS, and DOE Loan Programs Office announcements. Informational, non-advocacy, not investment advice.

Three reactor filings, three different decades

The NRC docket had a busy week, and the interesting part is that all three filings are the same story caught at completely different moments.

TVA’s Clinch River went to a mandatory hearing on its construction permit application. Mandatory hearings sound like nothing, and mostly they are: the Commission checking its own staff’s homework rather than refereeing a fight. But it’s the last real step before a permit, which makes this the week an advanced reactor site in the Southeast stopped being a paper exercise. Note where that is. The Southeast is already watching data centers and industrial load bid for the same firm megawatts.

Constellation went the other direction with the Christopher M. Crane Clean Energy Center, pulling NRC inspection reports that cover decommissioning and restart at the same time. Read that twice. One reactor, two opposite stacks of paperwork, running concurrently. That’s what a restart actually looks like from the regulator’s side, and it’s the only thing in this week’s set that puts electrons on a wire before 2030.

Then there’s Blue Energy Global, way back at the mouth of the funnel, walking the NRC through Part 1 design details at an August public meeting. Years from anything. Track it anyway, because today’s pre-application meeting is 2031’s construction permit.

One company describing a design. One utility finishing a permit. One operator restarting steel that’s already standing. Same queue, a decade of spread.

DOE just took an option off the table

The Energy Secretary cancelled three proposed National Interest Electric Transmission Corridor projects. NIETC is the federal crowbar for getting backbone transmission built where state siting has stalled, so this doesn’t stop any construction. Nothing was under a shovel. What it does is shrink the federal appetite for forcing wires through.

Which matters more than it sounds. If you’re siting a 500 MW campus you’ve got two ways to firm power: pull it across somebody else’s wires, or make it inside your own fence. Washington just got less interested in helping with the first one. Everything else this week is the second one.

And here’s the number that explains the whole thing

Reactors are slow. Everyone knows reactors are slow. So the fast path to firm power is a gas turbine, and that’s where the line is genuinely absurd.

GE Vernova’s turbine backlog hit 116 GW in the quarter ending July 23, up from 100 GW three months earlier. About a fifth of what’s under contract belongs to data-center customers. They’re taking reservations for 2031 delivery right now, expect to be at 125 GW of backlog and slot reservations by New Year’s, and lead times on new heavy-duty orders run around three years. As of the end of Q1 there were roughly 10 GW of slots left across 2029 and 2030 combined. Siemens and Mitsubishi are booking on the same horizon.

That’s not a supply chain problem. That’s a sold-out sign.

So look at the actual choice in front of a developer right now. A reactor that clears its permit in the early 2030s, or a turbine that ships in the early 2030s. That’s not a technology decision. That’s picking a line to stand in.

Which is why the small item this week is quietly the most satisfying one. A 150 MW battery project in Jackson County, Minnesota went active in the MISO queue with its interconnection agreement executed. Study cleared, agreement signed, position real. Somebody’s actually moving.

What to watch

Clinch River’s permit. Does it issue on the timeline the hearing implies? That’s the moment an advanced reactor site converts from study to capital commitment.

Crane’s restart schedule. It’s the only in-decade firm power in this week’s set, which makes its dates the most load-bearing numbers here.

A second round of NIETC cancellations. One is a decision. Two is a direction, and it pushes a lot more load-serving strategy behind the meter.

Sources: NRC ADAMS (Clinch River docket 05000615; Crane Clean Energy Center docket 05000289; Blue Energy CPA Part 1 public meeting), US Department of Energy (NIETC), MISO interconnection queue (J2730). Turbine backlog figures from GE Vernova’s Q2 2026 reporting via Utility Dive and Power Engineering, outside Dynawatt’s usual source set.

Dynawatt is an independent brief compiled from public sources: ISO/RTO interconnection queues, FERC notices, ERCOT Large Load Working Group, PJM stakeholder committees, NRC ADAMS, and DOE Loan Programs Office announcements. Informational, non-advocacy, not investment advice.

The one hard number: 12.2 GW out of CAISO

The withdrawals break down as 9,697 MW of gas across 19 projects, 1,900 MW of solar across three, and 650 MW of wind across two. All but one sit in California. The outlier is a 634 MW gas plant in Clark County, Nevada.

Before reading this as 12 GW of live projects collapsing in a week, look at the queue IDs. They run low, from the 250s into the 450s, and several of the names belong to plants that are already operating or were shelved years ago: Marsh Landing, La Paloma, Antioch, Oakley. This reads as CAISO clearing aged, stale positions, not two dozen developers pulling the plug on the same Tuesday. Treat it as housekeeping until the queue-reform status confirms otherwise.

Housekeeping or not, the composition says something. What washed out was overwhelmingly gas, in the ISO where gas has had the hardest time advancing in the first place. California’s queue is where thermal proposals sit and then vanish. This week a batch of them vanished at once.

Nuclear: motion, not megawatts

By item count, most of the digest is NRC activity. Topical reports, withholding determinations, decommissioning filings, security rulemakings, SMR pre-licensing. The pipeline flagged nearly all of it as routine, and that’s the right call. It’s licensing throat-clearing, years away from any electron.

A few items carry more weight. Constellation filed a subsequent license renewal for R.E. Ginna in upstate New York, which is about keeping existing baseload alive rather than adding to it. The NRC opened environmental scoping for two SMR units at Palisades in Michigan, a real if early step. And the single item that touches our beat directly is buried in the stack: the comment period for Constellation’s license amendment at the Crane Clean Energy Center, docket 05000289, with comments due July 8. Crane is the former Three Mile Island Unit 1, the 835 MW restart tied to Microsoft’s data-center power deal, targeted for 2027. The digest doesn’t flag that connection, but it’s the one operating-nuclear-meets-data-center story on the board, and this week it surfaced as a routine comment window. Motion, not megawatts.

DOE reopens the loan window

DOE’s Loan Programs Office is active again. The digest reports its first clean-energy loan guarantee in nearly a decade, a first conditional commitment for advanced fossil energy, a $2B tribal energy solicitation, and a grid-reliability loan closing. Read this cluster with caution. Some items carry stale framing (one still references the “Biden-Harris Administration,” another is the 2024 Ultium battery guarantee), so at least part of the batch looks recycled rather than fresh. The signal that survives: federal financing for firm and grid capacity is flowing again. That’s money, and money sits a long way upstream of a built megawatt.

Reading the signal

One theme runs through the week. It’s all supply side, and it splits cleanly into subtraction and paper. Gas left the CAISO queue in bulk. Nuclear advanced on paper. DOE reopened the checkbook. None of it puts firm capacity on the system in the near term.

Note what’s absent. The large-load demand side, the ERCOT Large Load Working Group we treat as the primary read on data-center demand, surfaced nothing this week, so there’s no fresh demand print to set this supply activity against. What’s left is lopsided: the queue is shedding thermal capacity, the firm-power hope is migrating to SMRs that are years of licensing away, and the one restart tied to a real data-center offtake showed up as a comment deadline. The gap between the load-growth story and deliverable firm megawatts did not close this week.

The one to watch: whether the CAISO withdrawals are a queue-reform purge. If CAISO confirms a batch cleanup of aged positions, the 12 GW is administrative and the real story is how much dead gas was still nominally in the queue. If any of these are genuine live-project cancellations, that’s a harder read on building thermal capacity in California, and worth pulling the individual queue records.

Dynawatt is an independent brief compiled from public sources (ISO/RTO interconnection queues, FERC notices, ERCOT Large Load Working Group, PJM stakeholder committees, NRC ADAMS, and DOE Loan Programs Office announcements). It is informational, non-advocacy, and not investment advice.

Large load & data centers

PJM moves toward “Connect and Manage” for large loads

PJM’s Board has directed a framework that would let large loads, data centers chief among them, connect to the transmission system before the upgrades that normally support them are finished. The trade is explicit. A load that does not bring its own new generation (PJM’s “Bring Your Own New Generation,” or BYONG, test) accepts non-firm status and agrees to be curtailed during system stress, ahead of pre-emergency demand response. It began as an Exelon problem statement at the Markets and Reliability Committee in February. It’s now a Board-directed workstream under a new Senior Task Force, with tariff and operational language targeted for the back half of 2026.

The moving parts worth tracking:

• Loads connecting under the framework would sit outside the capacity market’s Base Residual Auction. That’s PJM’s mechanism for keeping the cost off everyone else.

• A proposed Large Load Registry (50 MW and up per delivery point) would give states and utilities real-time visibility into where demand is landing, shifting enforcement toward state retail regulators instead of PJM mandates.

• A separate EDC and data-center coalition proposal floats a subscription-style backstop procurement, with a price cap near $555/MW-day and a 500 MW locational limit, deferring existing tariff backstops to the 2030/31 delivery year.

• The Data Center Coalition is pushing a state opt-in version: interruptible service classes, plus manual load-shed that curtails large commercial and industrial before residential.

The through-line is that nobody disputes the loads are coming. The fight is over who gets curtailed first, who pays, and whether the state or the RTO holds the lever.

ISO: PJM · Type: large-load · Stage: rule-change · Parties: Exelon, Data Center Coalition

Why PJM is in a hurry

PJM’s February long-term forecast raised the 20-year summer growth rate from 2.0% to 2.4%, with an explicit data-center adjustment across 13 member utilities and roughly 18.6 GW of net increase penciled in by 2040. Near-term numbers actually came down. The back half of the curve jumped. Meanwhile the 2027/2028 capacity auction cleared about 6.5 GW short of the reserve-margin target, driven by new large-load submissions with no matching supply. That gap is the entire reason Connect and Manage exists.

ISO: PJM · Type: large-load · Stage: study · Capacity: 18.6 GW

Nuclear & firm power

Clinch River clears its safety review

The NRC issued the Final Safety Evaluation Report for TVA’s Clinch River Unit 1 construction permit (Docket 50-615) in June. For the GE Hitachi BWRX-300, this is the gate that matters. The FSER means the agency has finished its technical review, and it lands alongside an ACRS safety report and a Revision 1 PSAR filing. Clinch River is the closest thing the US has to a flagship SMR, and it just moved materially closer to a construction authorization.

Type: nuclear · Stage: cod · State: TN · Parties: TVA, GE Hitachi

Palisades, twice over

Two different things are happening at the Palisades site in Michigan, and they’re easy to confuse. The existing 800 MW reactor (Entergy) is being restarted with DOE backing. The Energy Dominance Financing program approved its fourth, fifth, and sixth loan disbursements this period, a steady drip that reads as federal commitment to getting baseload back online. Separately, Holtec’s Palisades SMR LLC has filed a phased construction permit for two SMR-300 units (Pioneer 1 and 2), and the NRC has opened EIS scoping plus Section 106 cultural consultations with more than a dozen tribes. The consultation volume is procedural. Read it as the EIS clock starting, not as a dozen separate milestones.

Type: nuclear · Stage: study + cod · Capacity: 800 MW + 2×300 MW · State: MI · Parties: Entergy, Holtec

The advanced-reactor pipeline keeps filling

Oklo got preliminary NRC feedback on moving Aurora from DOE authorization to full NRC licensing, completed a pre-Phase 2 audit on its combined license application, and filed approved principal design criteria. X-energy cleared a graphite core verification for the Xe-100. None of these are construction events. They’re the licensing queue stacking up, which is the leading indicator for the firm power data-center developers actually want.

Type: nuclear · Stage: study · Parties: Oklo, X-energy

DOE keeps writing checks

Beyond Palisades, the Loan Programs Office closed a guarantee to restart a Pennsylvania reactor, backed a nuclear supply-chain financing portfolio, and reaffirmed support for the Vogtle expansion (~2.2 GW) through conditional commitment and financial close. The NRC also filed its May status report under Executive Order 14156, the national energy emergency declaration that sits over most of this activity.

Type: nuclear · Stage: cod + rule-change · Parties: DOE LPO, Southern Company

Storage & interconnection

CAISO logged three grid-scale batteries. Plano Storage (350 MW, Monterey County) reached commercial operation. Bateria del Sur (350 MW, Imperial County) and Dallas Energy Storage (400 MW, Monterey County) entered the queue. Storage stays the steadiest line in CAISO’s pipeline, the fast-response capacity the grid leans on while firmer resources grind through longer queues.

ISO: CAISO · Type: storage · State: CA

The week in one line. Demand is now the thing the system plans around. PJM is drafting rules to curtail it, the NRC is licensing the reactors meant to feed it, and DOE is financing the restarts in between.

Dynawatt is an independent brief compiled from public sources: ISO/RTO interconnection queues, FERC notices (Federal Register), the ERCOT Large Load Working Group, PJM stakeholder committees, NRC ADAMS, and DOE Loan Programs Office announcements. Informational, non-advocacy, not investment advice.

The queue in one breakdown

The composition is lopsided in a way that matters. Solar-plus-storage hybrids account for 17,996 MW, 52% of the queue across 67 projects. Standalone batteries add another 14,349 MW, 42% across 61 projects. Put those together and storage paired with solar is 94% of every megawatt seeking grid access in California right now. Firm dispatchable generation, the kind that produces power at 3am regardless of sun or wind, comes to 1,236 MW, just 3.6%, across four projects: three gas plants, the largest a 590 MW combined-cycle unit, and a single 53 MW geothermal facility. Wind is the remainder, under 3%.

Reading the signal

Most of this queue is not capacity. It is timing. A battery generates nothing. It moves existing power from one hour to another. The 18 GW of solar-plus-storage and 14 GW of standalone batteries piling into this queue are built for a single job: smoothing California’s midday solar glut into the evening ramp. They firm a renewable grid, and they do it well. They add close to nothing that a flat, constant load can draw on after dark.

The firm power that exists is tiny, and all of it sits years away. Filter the queue to its most material projects and the firm names do rise into view: a 715 MW gas combined-cycle plant, a 501 MW combustion turbine, a 500 MW steam-and-storage hybrid. But every one of them is at queue entry, the very start of a CAISO interconnection process that routinely runs several years. The projects actually crossing into operation this cycle are batteries and solar. So the supply reaching the grid now is intermittent, and the firm supply is still a filing.

For a data-center load, that is the wrong build on the wrong clock. AI demand is flat and round-the-clock, the opposite of the duck-curve problem storage solves, and it is landing in the next 18 to 24 months, faster than a gas plant clears interconnection. California is pouring 34.6 GW into its queue, and the sliver that actually matches the shape of the incoming demand is the part furthest from delivering a watt. The queue is not failing to grow. It is growing in the wrong direction for this particular load.

What’s not in the data

A limit worth stating outright, because the framing depends on it. All 139 of these projects are CAISO. There is no ERCOT here, no PJM, and not one large-load interconnection filing, which means the demand half of the story still is not in the numbers. The closest thing is a single line in one gas-plant summary guessing the unit “may support data-center demand,” and that is speculation, not a disclosed contract. So read this as what California’s supply queue reveals, not as the national load picture. The national picture needs the ERCOT and PJM feeds, and those are not landing in these pages yet.

What to watch

Two markers. Whether any of the firm thermal advances out of queue entry, because that is the first real sign California is treating 24/7 load as a generation problem rather than a storage one. And whether ERCOT’s large-load filings finally surface here, because that is the demand all of this supply is implicitly racing. Until both appear in the same issue, you are reading one side of the market and inferring the other.

Sources

• CAISO Public Queue Report, the full queue these 139 projects are drawn from: https://www.caiso.com/library/public-queue-report

• Browsable, daily-updated view of the same data:

https://www.interconnection.fyi/?market=CAISO

• Queue data ingested via gridstatus.io. Individual projects are listed by queue position in the CAISO report.

Dynawatt is an independent brief compiled from public sources (ERCOT Large Load Working Group, ISO/RTO interconnection queues, FERC & Federal Register filings). It is informational, non-advocacy, and not investment advice.

That sounds minor. It is the opposite of minor, and the reason runs straight through the documents Dynawatt tracks.

Go back to the evening of July 10, 2024, in the stretch of Northern Virginia the industry calls Data Center Alley. A lightning arrestor failed on a 230-kilovolt transmission line. The protection system did exactly what it was built to do, clearing the fault and attempting to reclose, several times in quick succession. Six faults flickered across the system in eighty-two seconds, each lasting a fraction of a second. That should have been a non-event.

Instead, roughly 1,500 megawatts of load vanished from the grid almost instantly. Around sixty data centers across two dozen substations saw the voltage dip and did precisely what they are designed to do: they cut themselves off and flipped to backup power to protect their servers. No utility shed that load. The customers shed themselves. Much of it stayed dark for hours.

Here is why grid engineers lost sleep over it. For a century, the nightmare scenario was losing supply. A power plant trips, and the system is planned and drilled to absorb the loss. Losing a gigawatt and a half of demand in an instant is a different animal, and the grid was never built for it. Drop that much load at once and voltage and frequency swing in ways the system isn’t tuned to catch. The circuit that protects a data center’s silicon can put everyone else’s power at risk.

Now multiply it. ERCOT alone has logged dozens of these load-loss events from large computing facilities in recent years. Dominion, the utility serving Data Center Alley, reported roughly 40 gigawatts of contracted data-center power by late 2024. The 1,500-megawatt scare was a rounding error against what’s coming.

This is the context for the items in Dynawatt’s feed that look like bureaucratic noise. ERCOT’s Large Load Working Group grinding through batch interconnection studies. A line about “electronic load ride-through.” NERC standing up an entirely new category of regulated entity, the Computational Load Entity, for any site over 20 megawatts. None of it makes headlines. All of it is the grid rewriting its rulebook so that the next time a fault hits Data Center Alley, the loads ride through it instead of bailing out. The demand everyone counts in gigawatts is the easy part. Teaching it to behave is the hard part, and it’s happening in working-group slide decks almost nobody opens.

While one stack of filings re-engineers the demand, another shows what’s lining up to feed it. Read the CAISO interconnection queue across the last two weeks and the answer is monotonous. Solar paired with batteries. Batteries on their own. Centennial Flats, Athos, PLUM, Darden, Umbriel, Wingtip. Project after project, the same shape: variable generation plus storage, trying to behave like firm power.

Then watch what left. In a single week, four projects over a gigawatt each withdrew from the CAISO queue. Two of them were exactly the kind of firm, dispatchable capacity that an around-the-clock load needs most: a combined-cycle gas plant and a 1,300-megawatt pumped-storage project. They walked out. The solar-and-battery hybrids kept walking in.

So here is the bet the system is placing, in plain terms. A new kind of load, large and twitchy enough to make the reliability regulator issue its rarest alarm, is being matched with a new kind of supply that is intermittent by nature and leaning on batteries to fake firmness. Maybe it works. Storage is improving fast and the economics keep getting better. But it is a bet, and the interconnection queue is where you can read the odds before anyone announces them.

This is the part of the AI-and-energy story the loud coverage skips. There is a whole genre now of maps with angry pins, organizing neighbors against the data center down the road over water and noise. That work has its place. It is not this.

Dynawatt reads the filings. The queue withdrawals, the NERC alerts, the working-group decks, the tariff dockets spread across seven grid operators in seven incompatible formats. That is where the signal lives, and it is almost always months ahead of the headline. The job is to tell you what the people who actually run the grid are worried about, in their own documents, without an agenda about whether any of it should get built.

The demand was never the interesting question. The demand was always coming. The question is whether a grid engineered to survive the loss of its largest power plant can survive loads that disconnect themselves by the gigawatt, and what’s queuing up to keep the lights on while it figures that out. The filings are already answering. Most people just aren’t reading them.

Dynawatt is an independent brief compiled from public sources (NERC alerts and reliability filings, the ERCOT Large Load Working Group, ISO/RTO interconnection queues, and FERC and Federal Register filings). It is informational, non-advocacy, and not investment advice.

Dynawatt is a weekly brief on the collision now reshaping the American power system: demand for electricity from AI and data centers is growing faster than the grid can add supply, and the gap is being negotiated in real time — in interconnection queues, working-group filings, and tariff proceedings that rarely surface in the headlines.

This brief reads those primary sources so you don’t have to. Each week it surfaces the most consequential developments in large-load interconnection, generation buildout, and the rules being written to govern both — compiled from public ERCOT, ISO/RTO, and federal filings, in a plain, non-advocacy register. The aim is signal: what happened, and why it matters to anyone building, financing, or planning around grid capacity.

It is informational, independent, and not investment advice. This inaugural issue covers the last seven days.

Large-load & data-center

⚡ ERCOT updates LEL-SSO power-variation framework; proposes three regulatory options

ERCOT presented updated guidance on managing sub-synchronous oscillations caused by AI training loads’ rapid power fluctuations, which can stress synchronous generators and cause shaft fatigue. The framework quantifies maximum continuous terminal variation (MCTV) per location; ERCOT is now soliciting industry feedback on three compliance models — project-specific limits, tiered/hybrid, or system-wide uniform caps — to balance interconnection speed, flexibility, and grid stability.

ISO: ERCOT · Type: Large Load · Stage: Rule-change · State: TX

Source

⚡ ERCOT studies 2-second LEL reconnection time, finds better grid stability

ERCOT evaluated voltage ride-through requirements for large electrical loads, testing reconnection times of 1 vs. 2 seconds post-fault across scenarios with 15.2 GW baseline load and up to 19.2 GW under higher-load conditions. Results indicate 2-second reconnection improves frequency settling and rate-of-change-of-frequency trends, supporting the NOGRR282 compliance framework for data-center and industrial loads.

ISO: ERCOT · Type: Large Load · Stage: Study · Capacity: 19.2 GW · State: TX

Source

⚡ ERCOT Large Load Working Group charter approved by TAC

ERCOT’s Large Load Working Group charter received Technical Advisory Committee approval on April 23, 2025, formalizing governance and operational procedures for coordinating hyperscale data-center and AI-compute load integration into ERCOT’s system planning and interconnection processes.

ISO: ERCOT · Type: Large Load · Stage: Rule-change · State: TX

Source

Grid, generation & regulatory

Atlas Complex: 3,200 MW solar + storage enters CAISO queue (La Paz County, AZ)

A 3,200 MW hybrid solar and battery storage project, Atlas Complex, has entered the CAISO interconnection queue in Arizona’s La Paz County — among the largest hybrid entries this cycle.

ISO: CAISO · Type: Hybrid · Stage: Queue-entry · Capacity: 3.2 GW · State: AZ

Jove Solar: 2,000 MW solar + storage enters CAISO queue

Jove Solar, a 2,000 MW photovoltaic-plus-storage hybrid in La Paz County, Arizona, has entered the CAISO interconnection queue, positioned to support California’s constrained summer-peak environment.

ISO: CAISO · Type: Hybrid · Stage: Queue-entry · Capacity: 2.0 GW · State: AZ

Angeleno Solar Farm: 1,150 MW hybrid enters CAISO queue

A 1,150 MW solar-plus-storage project in Los Angeles County has filed its initial interconnection request with CAISO — a major transmission-addition request amid rising state solar and storage targets.

ISO: CAISO · Type: Hybrid · Stage: Queue-entry · Capacity: 1.1 GW · State: CA

Windwalker Offshore: 1,000 MW offshore wind enters CAISO queue

Windwalker Offshore, a 1 GW offshore wind project in San Luis Obispo County, has filed a new interconnection request (Queue ID 1750), underscoring continued developer interest in offshore wind despite permitting and supply-chain headwinds.

ISO: CAISO · Type: Wind · Stage: Queue-entry · Capacity: 1.0 GW · State: CA

Bouse Solar + Storage: 1,000 MW hybrid enters CAISO queue (La Paz County, AZ)

Bouse Solar and Storage Plant, a 1,000 MW hybrid combining photovoltaic generation with battery storage, has entered the CAISO interconnection queue (Queue ID 2052), adding significant seasonal and diurnal balancing capability.

ISO: CAISO · Type: Hybrid · Stage: Queue-entry · Capacity: 1.0 GW · State: AZ

Dynawatt is an independent brief compiled from public sources (ERCOT Large Load Working Group, ISO/RTO interconnection queues, FERC & Federal Register filings). It is informational, non-advocacy, and not investment advice.