The Trace

It was just an average Wednesday morning in the Dominion power system. Then at 7:56:10 AM EDT, a transmission line tripped in the Dominion power system. That sentence should end right there. There is nothing terribly unusual about a line trip. But it initiated events that caused a disturbance that rolled through the entire Eastern Interconnection. It did not trip anything, but it showed up on meters as far away as Chicago.

Robert Orndorff, a part-time energy consultant, pulled this trace off of his home frequency monitoring system. We owe Robert a thank you for sharing it with us, because this trace is the heart of this story.

Here’s the trace:

Why It Went Everywhere

This chart shows system frequency immediately increasing to almost 60.1 Hz directly after the event. This is not the first time this has happened because of a line fault in the Northern Virginia area. What is causing this is almost 3,000 MW of data center load switching off because the UPS systems that protect the computer systems detected a voltage dip, a “brownout” due to the line fault. To protect the computers, those UPS systems immediately switch to battery and disconnect from utility. The issue is that there was no real power outage; the voltage dip lasted only a few cycles, but it was enough to trip the protection. The power system simply cannot absorb 3,000 MW of lost load in a single step without causing problems. The problem it caused was high frequency, and frequency is virtually the same across the entire Eastern Interconnection.

Confirmed in Ohio

This was verified by historical frequency data I pulled from my own frequency monitor I have under development here at my home. As you can see from this trace, the frequency not only went high in Virginia, but it went high in Ohio as well (UTC Time).

Enough to Make Reuters

The disturbance showed up on enough instruments to make Reuters. A piece came out after 6 PM detailing all the places the disturbance showed up. Orndorff caught it in Virginia. I caught it, later, in Ohio. Turns out we weren’t the only ones. Ting Labs runs a network of 1.4 million home sensors, and their CEO, Bob Marshall, told Reuters they picked up voltage disturbances reaching from the Eastern Seaboard all the way to Chicago. A grid disturbance normally corrects itself in milliseconds. This one took about ten minutes to fully settle, and Ting’s own customers in northern Virginia reported flickering lights and noisy air conditioners while it did. That’s three separate instruments, at three very different scales, all telling the same story. The message is: if a simple line trip can cause this much disturbance, one needs to ask what happens if there is a bigger fault in that Northern Virginia area.

NERC Isn’t Waiting

NERC isn’t waiting to find out. On May 4^th, they issued a Level 3 NERC Alert, brought about largely from the last disturbance in this same region. NERC is looking for ways to mitigate this issue, and they want feedback by August 3^rd. This is just the first step in their action plan. Mitigation is never easy and rarely a straight path.

An Object Lesson

In many ways this mirrors my own frequency station development. The reason I pulled data from history and didn’t post my own report is that my frequency station missed this event. It triggered, but I killed the report by restarting the machine to fix another problem. Once we (Claude and I) pulled the historical data and ran it through the configuration we had, a whole basket of problems came to light, and we completely rebuilt the report engine. That’s how development goes, and how it will go for NERC.

I wanted to seize this event as an object lesson while we had the chance. Operating a power system is not easy. New problems come up all the time. We learn, we fix, and we move on. Tell us about your development story.

Another eventful week in power system operations. PJM declared an EEA1 this week even though load came in about 6 GW below the level that triggered the record scramble two weeks ago. MISO declared an EEA2 as reserve margins thinned out hard during this second round of heat. All of it a reminder that this is an El Niño year — the heat isn’t going anywhere, and we’ll likely be dealing with system stress on and off through October.

The Root Cause

The “why” behind these lower numbers tripping the same alerts that used to take higher ones is often a mystery. But the reality is that an interconnection is just a machine made up of a lot of moving parts, and like any machine, when you push it hard enough for long enough, things start to break. Anyone running a fleet of equipment — or just running an office — already knows this. The copier will make it through maybe 200 sets of that 500-count, 250-page color binder — double-sided, three-hole-punched, collated — before it quietly raises the white flag and goes on vacation. The same story shows up in forklifts, delivery trucks, anything you lean on hard for a big project. Interconnections are no different.

The Data

MISO and PJM are the only two RTOs in the Eastern Interconnection with decent, publicly accessible outage data — and MISO’s is better than PJM’s. ISO-NE has a stated policy of not releasing reliability data to the public. NYISO, as I’ve said before, is an information vacuum. But pull what MISO and PJM do publish and line it up, and a clear pattern shows up: when it’s hot and generation is maxed out, forced-outage counts climb — and they keep climbing for a few days after the heat itself breaks. There’s normally a recovery period after that, as repairs get made, and equipment comes back into service.

That’s where MISO and PJM start to diverge. Over the past five weeks, MISO’s forced outages have genuinely recovered between events, bottoming out at 56% of their own peak during the lull before the record emergency hit. PJM’s haven’t. Over the same stretch, PJM’s forced-outage floor has only pulled back to 72% of its own peak before climbing again — closer to flat than to a real recovery cycle.

Forced outages indexed to each system’s own peak reading. MISO pulls back toward baseline between events; PJM’s floor barely moves. Source: PJM Data Miner 2 (Generation Outage for Seven Days by Type); MISO Multiday Operating Margin Forecast Report.

Derates, the Other Hidden Data

MISO is also the only Eastern Interconnection RTO that publishes derate data on top of outage data, and that’s a meaningful gap. A derate is a different animal from a forced outage — a derated unit won’t show up on the outage list at all, even though it can hit reliability just as hard. A derate simply means a plant can only produce part of what it’s normally capable of. There are plenty of reasons for one: gas turbines lose capacity simply from rising air temperature; nuclear plants are sensitive to cooling-water temperature — France took significant derates during its own heat dome because of river-water limits; coal units get derated when a coal mill goes down and can’t feed enough fuel; hydro units are sensitive to tailwater level. Whatever the cause, derates are a real chunk of the capacity the Interconnection loses during stress events — and the fact we can only see this from one RTO says a lot about how far transparency still has to go across the rest of the fleet.

Wildfire Smoke, the Gas Turbine’s Enemy

In the last 48 hours, a plume of Canadian wildfire smoke has settled over the Eastern Interconnection. Right alongside it, MISO’s derate numbers have climbed — and that’s not a coincidence. It hits gas turbines specifically. A gas turbine is a mass-flow engine: it moves enormous volumes of air to make power, and to protect the precision internals from erosion by airborne grit; it relies on large, high-efficiency filters on its intakes. Wildfire smoke is almost entirely PM10 and PM2.5 — exactly the particle sizes that foul those filters fast. As the filters load up, airflow drops, and so does output. During the Carr Fire, we were changing air filters at REU as often as every 36 hours — each filter ran well over a thousand dollars, and there were at least a half dozen per unit. The smoke fouled the generator cooling-air filters right along with the intake filters.

MISO’s derated capacity troughed at 10,745 MW on July 8 and has climbed nearly every day since, reaching 12,875 MW by July 16 — up almost 20%, with the sharpest single-day jump landing exactly on July 14–15, right as the smoke arrived in earnest. MISO’s own forecast has it easing back down through next week as both the smoke and the heat lift, so treat this as a real hit, but likely a temporary one rather than a structural one.

MISO system-wide derated capacity, actual (solid) and MISO’s own forecast (dashed). Source: MISO Multiday Operating Margin Forecast Report.

Conclusions

Expect utilities to keep struggling to keep their already-thin fleets not just running, but running at full output, as the heat continues through the rest of the summer. That’s exactly why adequate reserves matter as much as they do — that extra margin is what covers the “what if” that always seems to show up eventually. Hope you enjoyed the piece — tell us about your copier going south, or your own reliability story, in the comments.

During the week PJM pushed toward an all-time peak, interconnection frequency spent most of its time running slightly high — a sign, as covered here last week, that the system had real-power balance well in hand even as it leaned hard on reactive support. Zoom out to the interconnection level and the story is almost boring: the grid held.

Zoom into any individual utility’s distribution system that same week, and “the grid held” stops being the whole story. Con Edison cut voltage across four boroughs and into Westchester County, then cut power outright to nearly 10,000 customers in Queens for a completely different reason. Baltimore Gas & Electric ran emergency load-cycling events while cable faults tripled. Commonwealth Edison — running some of the most sophisticated voltage-optimization software in the country — still posted the worst voltage numbers on the entire Whisker Labs/Ting network. And in Lakewood, Ohio, thousands of FirstEnergy customers lost power for days, for a reason the utility and its regulator both already knew about before the heat wave started.

Five utilities, one heat event, four genuinely different failure mechanisms. None of them shows up the same way in the numbers utilities are required to report — and one of the biggest doesn’t show up in a formal metric at all.

Two Different Numbers

If last week’s note left you thinking about frequency, it’s worth pausing on why frequency and voltage aren’t the same failure mode, because the two get conflated constantly.

Frequency is a system-wide number — one value, shared essentially instantaneously across the entire synchronous interconnection, telling you whether generation and load are in balance right now. When PJM’s frequency ran high through the emergency, that wasn’t relief; it was the reactive side straining while the real-power balance held.

Voltage is a local condition. It decays with distance, current, and impedance, and it’s set by conditions at a specific point on a specific feeder. A system can sit dead-on 60.000 Hz across the entire interconnection while individual customers three miles down an overloaded feeder are looking at 108 volts at the wall outlet. That’s the story this piece is about — the big number stayed fine. The local number didn’t, and across the five utilities below, it didn’t fail the same way twice.

Con Edison: One Utility, Two Failures

Con Edison did two distinct things this heat wave, and conflating them costs you the actual story.

The first was a decision. Beginning July 2, ConEd cut voltage 8% across northern Manhattan, the northwest Bronx, Queens, Staten Island, and Westchester — a conservation voltage reduction: precautionary, reversible, and the same move it made in June 2025 under similar conditions. It’s a deliberate operational lever — lower the voltage a controlled amount, shave demand across the system, buy margin. Some Richmond Hill and Forest Hills residents reported it felt closer to 20–25% than the announced 8%, a gap worth taking seriously — but that’s a measurement dispute about a chosen action, not evidence of a different mechanism.

The second was not a decision. On July 3, ConEd cut power outright to roughly 9,800 customers across Howard Beach, Ozone Park, Richmond Hill, and South Ozone Park, plus a separate outage in Riverdale. ConEd’s own statement is specific: temporary shutoffs due to equipment problems caused by extreme heat and heavy demand, done to allow faster repairs and head off longer, wider outages. That’s equipment reaching a thermal or capacity limit it wasn’t rated to handle and failing — the same physics that shows up next at BGE.

Neither of these is new for ConEd, and neither is likely to be the last. NYISO has warned of possible NYC-area blackouts as early as this summer, and ConEd’s own reliability modeling shows a projected reliability margin of negative 1,679 MW under a sustained three-day, 95°F event — close to the event that just happened. That number sits in the background of ConEd’s ongoing rate case friction with Governor Hochul’s office, an 11.4% electric rate request currently drawing an audit demand.

BGE: The Same Failure, a Different Territory

Baltimore Gas & Electric spent July 1–2 running emergency cycling events while its cable faults climbed from roughly 1,400 to more than 3,000 affected customers. That’s not a coincidence of timing with ConEd’s shutoffs — it’s the same underlying mechanism showing up somewhere else. Two separate utilities, two separate service territories, the same failure mode in the same week: equipment that runs fine most of the year hitting a thermal limit under the combined stress of extreme heat and heavy demand. One data point is an anecdote. Two independent utilities failing the same way in the same event is a mechanism.

ComEd: When Good Software Meets a Bad Substation

Commonwealth Edison is, on paper, the utility doing this the right way. It runs a regulator-mandated closed-loop voltage optimization system — Aspen VVC/VVO, built around a Distribution State Estimator that fuses SCADA, AMI, and sensor data with a network model to estimate voltage at every node on a feeder, then optimizes tap changers, capacitor banks, and regulators against that modeled profile in real time. That’s a fundamentally more sophisticated approach than the fixed-percentage tap cut ConEd runs — no blind spot past the substation, an actual model of where the worst voltage on the feeder is likely sitting.

And ComEd still topped the entire Whisker Labs/Ting network for the worst voltage performance during this event.

The reason isn’t the software. It’s what the software is fighting. A meaningful share of ComEd’s substations and feeders date to the late 1960s and 1970s, running over their original design capacity, and no control algorithm creates physical headroom that doesn’t exist. ComEd’s own $15.3 billion capital rebuild plan doesn’t finish until 2028–2031. Until then, the closed-loop system is doing genuinely good work holding the line as gracefully as physics allows — it’s just holding a line that’s already too close to the edge.

There’s a regulatory wrinkle worth flagging here too. The Illinois Commerce Commission’s own Performance Metrics 2.0 workshop, in April 2025, confirmed ComEd tracks no formal power quality metric at all — momentary interruptions, voltage sags and swells go essentially unmeasured, even as SAIDI, SAIFI, and CAIDI keep ComEd in the “first decile” nationally in its own filings. ComEd’s 2022 “System Visibility Index” proposal reads, in hindsight, like a tacit admission of the same gap. The metrics that make it into a rate case filing are the metrics that don’t capture this problem at all. (This event’s separate 297,000-customer outage count was storm-compounded — a different, unrelated story, not part of the voltage picture above.)

FirstEnergy: The Failure Everyone Already Saw Coming

Lakewood, Ohio, isn’t a case of adequate equipment meeting an unprecedented extreme. It’s the opposite — a known, specifically identified failure point that simply hadn’t been fixed yet when the heat wave arrived.

Underground lines feeding the Lauderdale substation failed on Lakewood’s southwest side, leaving roughly 1,200 customers without power for multiple days in triple-digit heat-index conditions. FirstEnergy’s public statement attributed the strain to the heat wave driving elevated regional demand. Lakewood Mayor Meghan George’s account, filed as an updated complaint with the Public Utilities Commission of Ohio (PUCO), is considerably less forgiving: this is the same playbook as a 2025 outage event that triggered a formal PUCO probable-noncompliance finding — a finding that explicitly stated the earlier failures were “caused by equipment or line failure,” not weather, and that FirstEnergy’s infrastructure serving Lakewood was “older than infrastructure found elsewhere.” PUCO’s own October 2025 corrective-action letter didn’t specifically address the Lauderdale substation — which is exactly why Mayor George is now asking PUCO to reopen the case and add a financial-penalty component: fines, ratepayer refunds, or an escrow mechanism.

The timing sharpens the point. PUCO denied FirstEnergy’s request to loosen its own reliability standards on June 23 — nine days before this outage began. Whatever else is true about this event, “we didn’t see it coming” isn’t an available explanation for anyone involved — not the utility, and not the regulator.

A Cleveland Footnote

Lakewood sits entirely outside Cleveland city limits and entirely within FirstEnergy’s Illuminating Company territory — worth stating plainly, because it’s tempting to fold this in with Cleveland Public Power’s own troubles (a battery fire at CPP’s Lake Road substation last June left 3,000-plus customers dark for about 18 hours) into a single Cleveland story. Don’t. They’re genuinely separate failures with genuinely opposite root causes: FirstEnergy’s is deferred capital investment by an investor-owned utility that can raise rates and chose not to spend fast enough; CPP’s is four decades of frozen municipal rates that left a public utility unable to fund capital replacement at all. Regular readers know that story — it’s the subject of “Forty Years of Free.”

The two utilities are connected, just not in a way that makes one utility’s transformer the other’s problem. CPP’s own filings state that it obtains substantially all its power through agreements delivered via CEI interconnections — a dependency that traces back through Centerior to at least the 1990s. A public utility founded in 1906 explicitly to compete with the private one still needs the private one’s wires to get most of its power to market. That’s not a shared distribution asset, and it doesn’t merge the two outage stories into one cause — but it’s a fitting irony for a city with two utilities failing independently, for opposite financial reasons, in the same week.

SCE: The One That Doesn’t Fit the Pattern

Southern California Edison is worth naming precisely because it breaks the pattern instead of confirming it. SCE’s voltage and reliability decisions this time of year are driven overwhelmingly by wildfire risk — Public Safety Power Shutoffs, de-energization ahead of red-flag conditions — not by thermal capacity limits on aging equipment. It’s a genuinely different risk calculus, running on a genuinely different trigger, and folding it into the same bucket as the other four utilities would blur a distinction worth keeping sharp: not every utility having a bad week this summer is having the same bad week.

What the Big Number Doesn’t Tell You

Put the interconnection frequency chart from last night’s note next to any of these five distribution stories and you’d never guess they happened the same week. That’s not a contradiction — it’s the entire point. “The grid held” is a true statement about generation and load balance across an entire synchronous interconnection. It says nothing about whether your specific feeder, your specific substation, or your specific transformer held too.

Most of what determines that answer for you personally isn’t in a filing anywhere. SAIDI (System Average Interruption Duration Index)and SAIFI (System Average Interruption Frequency Index) track sustained outages. They don’t track ComEd’s brownouts, because nobody requires ComEd to measure them. They don’t distinguish ConEd’s deliberate 8% cut from ConEd’s equipment failure three days later, because both just look like “ConEd had a bad week” from the outside. And they say nothing at all about whether the fix for your neighborhood’s specific failure point has actually been funded, or just promised.

If you want to know which of these five stories is happening to you, the honest answer this summer is: check your own outlet. My own plug-in Ting sensor — the same Whisker Labs network cited throughout this piece — told me more about my service quality this month than any regulatory filing did. Thanks for reading, please share your outage story.

Last week was one for the record books in power system operation. Massive heat, massive load, and for a lot of utilities, the first real test of load growth we’ve seen since the early 2000s. The good news: operationally, we made it — no firm load shed. The bad news: we danced with the devil most of the week. Making it through doesn’t mean the system is fine; it means the stop-gaps held. Add another 20 GW of load next summer, and they might not.

At 5:01 PM (17:01) on July 2nd, the Locational Marginal Price (LMP) for the PJM RTO reached $1,139.32/MWh. Shortage Pricing was active, a Potential PAI flag was attached to the emergency action in progress, and instantaneous load was still climbing against every forecast curve that said it should be falling. That produced a flurry of activity in the Northern Virginia/District of Columbia corridor to hold the system together. The system operators earned their pay that day. No AI in control here.

So why the title? We won’t get the final, studied after-action reports — the ones where every fact is checked, every timestamp cross-referenced, the math run and re-run — out of NERC, PJM, MISO, ISO-NE, and NYISO for at least sixty days. It takes that long to gather the facts, verify the timestamps against each other, run the numbers, and convert an engineering report into a legal document fit for public consumption. Some of the language gets softened along the way to make it politically palatable, but that’s genuinely how long it takes to get to the facts. So, let’s set up the week with what we know right now.

Where It Starts — Monday, June 29th

PJM knew things were getting hot and started making preparations early.

1. They issued a maintenance recall — the early warning that gives facilities time to button up whatever they’re working on and get equipment back in service. If you don’t, your outage gets converted to a forced outage, and that counts against your reliability calculation. Dollars and cents.

2. PJM requested authority from DOE to call on standby generators, citing the new Manual 13 Revision 98, which adds Attachment O — a procedure that officially wasn’t considered effective until July 23rd. The new procedure puts a formal path in place for future requests like this one. This is the third time PJM has requested and been granted this authority in 2026. This isn’t going away.

3. That same day, a derecho tore through central South Dakota — 131 mph straight-line winds near Highmore, in Hyde County — and critically damaged two ENGIE wind farms, the 250 MW Triple H project and the 200 MW North Bend project. The same storm system tracked east, causing large outages across Michigan’s Upper Peninsula, Ontario, and Québec. It reignited into severe thunderstorms along nearly the same path every afternoon for several days running, stretching restoration out for days in some places.

All of this was happening while Europe was emerging from its own multi-week heat dome. The human cost of that heat dome has been well documented here; the energy impact on Europe’s grid is a separate story I’ll cover soon.

The Dominion/Northern Virginia Zone — The Structural Trap

You’ve heard me say this before: if the PJM power system ever collapses, the most likely point of origin is the region around Washington, DC. Why, and how did we get here? The short answer is politics, in the most political zone in the nation. Let’s look at it piece by piece.

1. Loudoun County, Virginia has the largest concentration of data centers on Earth. It’s not a slight majority either — this is the epicenter. They’re there to serve data services for the federal government, the military, and the contractors who support those agencies. Data centers invite more data centers, and the buildout has simply exploded. That’s a massive amount of load sitting in a relatively transmission-poor region.

2. Political pressure against coal shut down Brandon Shores power station years ahead of its planned retirement. Normally a utility runs engineering studies years in advance of a retirement, and the required system support is already in place and operating before the plant comes off the grid. Politics doesn’t work that way, and a badly needed asset got pulled out of an already-challenged point on the interconnection.

3. The Coastal Virginia Offshore Wind (CVOW) project is years behind schedule. This project was cited as a reason Brandon Shores was no longer needed — even while it remained incomplete. CVOW won’t provide ancillary services like voltage support, but it would have helped with energy in the region. It could also introduce transmission issues of its own once it’s actually in service.

4. The PATH 765 kV project died the death of a thousand cuts, mired in well-funded NIMBY litigation until its promoters finally surrendered and canceled it.

What comes through all of this is that the area is served by three utilities that don’t coordinate well on their shared problems — Baltimore Gas & Electric, Dominion, and Delmarva. Cooperation across the region is fractured, and every one of these competing agendas gets handled as a standalone project instead of a systems-engineering problem. The result, more often than not, is that nothing gets solved.

The Escalation — Tuesday, June 30th

On Tuesday, the stress starts showing up.

1. PJM issues an EEA1. MORRISVL-LOUDOUN4 hits 2,913 MVA, and PJM posts a Post Contingency Local Load Relief Warning: No Generation or Switching Available. Line-loading issues in the Loudoun corridor start to appear.

2. Secretary Wright signs both 202(c) orders, effective 11:59 PM (23:59) — a 50 MW threshold, 15-minute switchover requirement. PJM gets the demand response authority it asked for, in its pocket.

3. Offline capacity margin context: PJM is running 15,000–22,000 MW offline all week against a ~34,700 MW historical average. The cushion was thin before the peak even hit.

The Climax — Thursday, July 2nd

Before PJM’s afternoon even started, New York took its own hit. The Champlain Hudson Power Express — the brand-new 1,250 MW HVDC tie bringing Hydro-Québec power into Astoria, roughly 20% of NYC’s peak demand — tripped at the Hertel converter station on the Québec side at approximately 5:30 AM. The line had only gone commercial on June 1st, and NYC had already leaned on it as a reason to let some oil-fired generation retire. It stayed off for roughly 24 hours while Hydro-Québec made repairs on their end — right in the middle of the same heat wave that was building toward PJM’s afternoon.

One footnote worth naming: PJM’s emergency board gives a timestamped, ID-numbered account of every action it took that week, which is how this whole piece can be reconstructed to the minute. NYISO’s public footprint on the CHPE outage amounted to a spokesperson’s quote and a vague “sometime between Wednesday and Thursday morning.” Same industry, same continent, two very different disclosure cultures — and readers used to the minute-by-minute PJM timeline below will notice the difference the moment we cross state lines.

Then the day itself, in the Dominion zone:

1. Between July 1st and July 2nd, PJM’s Emergency Procedures board named 14 different Dominion facilities in Post Contingency Local Load Relief Warning messages — 22 messages total. Four facilities were named a combined 12 times over that stretch. Extend the window to July 3rd and the total climbs to 31 messages against those same 14 facilities — no new fault points showed up, the system just kept re-hitting the ones it already had. A few of those facilities hit the identical MVA figure on both days, which isn’t a coincidence of fluctuating load — it’s a fixed equipment limit getting re-tested at nearly the same hour on back-to-back afternoons.

2. At 5:00 PM (17:00), PJM declares EEA2 and fires all three demand response tiers — Long_120, Short_60, Quick_30 — simultaneously. The load trend flattens for a few minutes, then starts climbing again.

3. At 5:18 PM (17:18), a Maximum Generation Emergency Action for MIDATL/SOUTHERN goes out, requiring any generation in the area to ramp to full load. This is the emergency, price-doesn’t-matter order.

4. At 5:32 PM (17:32), an Attachment O warning goes out — Transmission Security, not capacity. It’s not clear from the record whether this warning ever crossed into an actual directive to switch specific loads to backup generation, but the message itself tells you how close PJM was willing to go. We’ll learn more once the 60-day review is in.

5. The entire PJM footprint is pricing above $500/MWh simultaneously — there is no unconstrained pocket anywhere on the map.

6. LMPs go to extremes across the RTO: PEPCO $2,980/MWh, BGE $2,861/MWh, Western Hub day-ahead settling at $1,222.75/MWh.

   

7. At 5:55 PM (17:55), load is still trending up. Every demand response tier has been exhausted. PJM is out of tools.

8. At 6:02 PM (18:02), the LMP market breaks and load starts trending down. We made it.

9. PJM’s own framing, after the fact: roughly 162,700 MW instantaneous, suppressed by DR, “likely to have surpassed” the 165,600 MW all-time record once the mandatory 60-day DR performance review is complete.

The Voltage Story

If you watch the parameters, the system was talking to us all day long. A retired system operator knows the language, and here’s the translation.

1. Frequency ran high all day. That tells you the system had enough prime-mover energy online — more than enough to meet the demand side of the equation. This was never a shortage of megawatts.

2. PJM issued an HLV (Heavy Load Voltage) Warning and an HLV Action, requiring generators to increase voltage support, capacitors to come online, and reactors to go out of service. The system needed every reserve MVAr it could get to hold voltage and maintain a reactive reserve. You’ve heard me push on ancillary services and voltage support before — this is exactly why. What was happening in the DOM area at the very end of the day was a voltage problem, not a power problem. The system was running out of MVArs to hold itself together and was getting dangerously close to voltage collapse.

3. Generation in the right spot unloads transmission; unloading transmission means less MVAr demand. Generation in the wrong spot makes all of it worse. Physics isn’t a subject taught in political science.

The Generation Story

Coal generation stayed pegged at 84–88% of stack capacity, just over 27 GW. It had been at 29 GW earlier in the week, but a likely forced outage took some of that offline. The fact that coal never budged tells you exactly where it sits against gas in the bid stack: coal is what shows up when the system is stressed. Nuclear dipped slightly; likely output derates from cooling-water temperature limits — it was hot. Oil and dual-fuel units came into play in a big way, with oil capacity reaching over 6 GW. PJM was running everything it had. Gas crossed into 70+ GW. Through all of it, wind stayed under 2% of the contributing generation — for the most part it simply didn’t show up, which is typical under a heat dome. Batteries sat at 0% output; there was no surplus energy anywhere to charge them, so they sat this one out.

Rinse and Repeat — July 3rd

The power system carried Thursday’s heat straight into Friday. There was no relief from cooler overnight temperatures — the heat-soaked asphalt and concrete were still hot to the touch the next morning. Load stayed roughly level, with some of it shaved off by early weekenders leaving work. A few things stand out:

1. MORRISVL-LOUDOUN4 hits 3,005 MVA again — the same figure it posted the previous morning, at nearly the same hour (10:56 AM versus 10:11 AM the day before). Not a new high; the same bottleneck reasserting itself at the same time of day, two mornings running.

2. EEA2 is called three hours earlier than the day before — 2:00 PM (14:00) instead of 5:00 PM.

3. A TLR Level 5a goes into effect on Flowgate 50 (AP-South Interface) — actual curtailment territory, not an advisory. It first showed up at 18:00 on the 2nd, was stepped back down that same evening, then came back at 17:30 on the 3rd. Post Contingency warnings ripple westward alongside it, into COMED and briefly even FE-AP. The actual curtailments, if any will wait for the NERC report.

4. The 202(c) authorization expires at 11:59 PM (23:59) without renewal — later requested and extended through July 6th, per PJM’s own July 3rd update.

5. Attachment O is staged again for DOM at 5:22 PM (17:22) but not fully invoked — load breaks over cleanly before it’s needed.

The same warning went out a third time, early Saturday morning on the 4th, ahead of that day’s peak heat. Same tool, same threshold, three separate mornings and evenings in five days. Not once, on the record available, did the system need to go past the warning stage.

On July 5th, the weather largely broke for the Dominion system. Some Post Contingency Load Relief did come across for the Delmarva system, a sign the heat wasn’t entirely finished — but the emergency, for now, seemed to be over.

The Greater Power System

The storms finally tore into the Con Edison system and New Jersey on Friday, July 3rd — winds near 70 mph, roughly 165 utility poles down in New Jersey alone. At the peak, on July 4th, poweroutage.us was showing more than 1.3 million meters out across the states affected by the same storm system that started with the derecho in South Dakota — and one meter can easily mean several people in one household. By Monday evening there were still over 370,000 meters out across the country.

To the north, Hydro One and Hydro-Québec were still working their way back from the massive outages the Canada Day storms caused earlier in the week.

Both NYISO and ISO-NE were burning oil through the heat wave, for the same reason as the rest of the country: wind simply wasn’t there, and neither RTO has enough internal gas-fired generation to cover load at these levels without it.

Closing

All I can say is: wow, what a ride. Part of me misses the adrenaline of being on the desk for days like this. The smarter part of me is glad to be watching from a distance. But I’m glad to be here, telling the story in words that make sense.

At the end of the day, we have to get NIMBY and political obstruction into some kind of reasonable check. This isn’t Dominion’s fault — it’s the environment they have to do business in. These groups deserve their day in court, same as anyone. But when they get powerful enough to block everything and do real harm to infrastructure, the whole region depends on, it’s time for a reality check.

Please share your thoughts.

At midnight on December 15, 1978, Cleveland became the first major American city to enter financial default since the Great Depression.

It didn’t have to happen that way. Six Cleveland banks, holding $14.5 million in short-term municipal notes coming due that night, could have done what they’d always done — rolled them over. They’d done it routinely for the prior administration. Instead, led by Cleveland Trust chairman Brock Weir, they made the rollover conditional on one thing: Mayor Dennis Kucinich had to agree to sell the city’s municipal electric system — Muny Light — to the Cleveland Electric Illuminating Company.

Kucinich said no. The city went into default. He lost his job in the next election.

He was right anyway.

The Founding Proposition

The story of Cleveland Public Power begins not in 1978 but in 1903, when Progressive Era mayor Tom L. Johnson proposed building a city-owned electric plant as a direct competitive check on CEI. The Chamber of Commerce opposed it. CEI opposed it. Johnson lost the fight in his own term, but his successor Newton Baker cleared the legal obstacles, and by 1914 the Cleveland Municipal Light Plant — universally known as Muny Light — was in full operation as the largest municipal electric plant in the nation.

The logic was simple and deliberately non-ideological: not charity, but competition. Private utilities serving natural monopoly markets will price to what the market will bear. A municipal competitor changes that calculus. Muny Light’s job wasn’t to run at a loss — it was to exist, to offer a benchmark price, and to give CEI a reason to stay honest.

By 1927, Muny Light was serving over 42,000 customers. Its advocates claimed it had saved Clevelanders $14 million in its first eight years. Through the Depression, Muny’s profits helped fund municipal relief payments. The utility was, as its founders intended, a civic asset with a financial function.

That didn’t make it beloved by the banks or by CEI.

December 15, 1978

By the mid-1970s, Muny Light was aging and politically neglected — its directors frequently accused of corruption; several later found to have been working effectively for CEI’s interests. Republican Mayor Ralph Perk agreed to sell it outright, a decision so unpopular it ended his political career. He finished third in the 1977 mayoral primary, behind both Kucinich and his opponent, largely on the Muny Light issue.

Kucinich won the general at 31, the youngest mayor of a major American city, on an explicit promise not to sell.

What he inherited was a fiscal trap. The prior administration had been spending bond proceeds for general operating purposes, leaving roughly $40 million in debt to be refinanced. CEI immediately filed a federal lien on city property for the $18 million Muny Light owed them. Kucinich drew from the city’s operating funds to pay it — which drained the general fund further.

Then the banks moved.

The six institutions held their $14.5 million in notes like a loaded weapon. Rolling them had been routine. Now it wouldn’t happen — not unless Kucinich agreed to the sale. The interlocking interests were documented the following year by a House Banking subcommittee: the six banks shared eight directors with CEI, held 1.8 million shares of CEI stock in trust portfolios, and extended CEI $72 million in credit lines. More than 50 officers and directors of four of those banks had contributed to recall campaign committees targeting Kucinich. Cleveland Trust CEO Brock Weir escalated further: agree to sell, and the bank would not only roll the notes but lend the city an additional $50 million.

Kucinich said no. Five banks agreed to wait for a special election on an income tax increase. Weir refused. City Council — led by Kucinich’s political nemesis George Forbes, who had privately met with Weir — gave the mayor a final ultimatum at 11 PM on December 14: sell or default.

At midnight, the city defaulted.

Kucinich’s statement to Council before the clock ran out belongs in the record: “Ultimately the history of Cleveland will reveal that Cleveland Trust and CEI and some members of this City Council brought shame to this city and besmirched its history. But it will also show that there were people who recognized wrong when they saw it and condemned it for what it was.”

He lost reelection anyway. The House Banking subcommittee investigation found the following year that the banks had engaged in “deep animosities and political cross-currents” rather than sound credit judgment — confirming everything Kucinich had claimed. In 1998, City Council formally granted him amnesty and acknowledged he had shown “the courage and foresight to refuse to sell the city’s municipal electric system.” Cleveland Magazine had already done the arithmetic: between 1985 and 1995, Cleveland Public Power saved its customers $195 million compared to what they would have paid under CEI rates.

He was right.

The Name, and What It Was Before

Before there was Cleveland Public Power, there was Cleveland Municipal Light Plant — universally called Muny Light. The 1983 rename happened under Mayor George Voinovich, who had stabilized the utility after the default years and pushed through capital improvements. Same mission, same wires, new nameplate. The competitive rationale was not retired with the name.

Rates Are Set by Council, Not by Voters

Here is a structural point that shapes everything that follows. Cleveland Public Power’s rates are set by City Council ordinance — not by the Public Utilities Commission of Ohio, which regulates CEI and every other investor-owned utility in the state, and not by voter referendum. By the council. The same body that represents the customers whose votes they need.

That creates a predictable incentive structure. Rate increases are politically painful. Frozen rates are politically comfortable. A 2020 consultant’s report, obtained by WKYC, called CPP’s financial situation “not sustainable” and proposed staged increases of 5%, 8.4%, and 10.5%. The administration did not brief council on the unfinished study. Council did not ask.

The rates stayed frozen.

Forty Years

The last base rate increase before 2024 was in 1983 — the same year Muny Light became Cleveland Public Power. A dollar in 1983 is worth roughly $3.15 today. CPP’s rates weren’t just frozen; they were eroding in real terms every year while wages, equipment costs, debt service, and power supply expenses climbed.

By 2024, the operating cost trajectory looked like this: $73 million for the year, rising to an estimated $79 million in 2025 and $83.9 million in 2026. Annual debt service was stepping up from $16 million to $18 million. The capital asset base — $318 million net of accumulated depreciation — shrank by nearly $6 million in fiscal year 2023 alone. CPP was consuming more than it was reinvesting.

The infrastructure visible from the street told the same story. Thirty-six substations. A patchwork distribution system running at 69kV, 13.8kV, 11.5kV, and 2,300 volts depending on the substation — legacy voltage tiers that speak to deferred standardization across decades. In 2019, the Director of Public Utilities told City Council that infrastructure investment “will make a significant difference” on outage issues. Council President Kevin Kelley replied: “I just can’t tell them that everything is going to be okay anymore. I need to know why it’s going to be okay and when.”

In June 2025, a battery fire at CPP’s Lake Road substation took down power for more than 3,000 customers for roughly eighteen hours.

Who They Serve

Cleveland Public Power’s service territory covers roughly 60% of Cleveland’s geographic footprint, concentrated primarily on the city’s East Side.

That geography is not incidental. Cleveland’s East-West divide is one of the most documented in any American city — a direct legacy of redlining and blockbusting that solidified after the 1950s. The East Side is predominantly Black. The West Side is predominantly non-Hispanic white and Hispanic. The wealthy suburban ring — Beachwood, Lakewood, the Heights communities — is CEI/FirstEnergy territory. CPP serves the city proper, and within the city, primarily its eastern half.

The numbers: per the 2020 census, Cleveland’s median household income is $30,907. Its poverty rate is 32.7%. Its population is 47.5% Black American and 13.1% Hispanic or Latino. This is the customer base absorbing 40 years of deferred infrastructure investment.

The utility was created in 1903 to serve the working people of Cleveland against monopoly pricing. By the compounding of geography and demographics, it predominantly serves the city’s lowest-income, majority-minority neighborhoods. There is no dedicated low-income Lifeline rate program. Customers in arrears are directed to HEAP and county assistance programs.

The Performance Numbers

CPP’s tagline is “the utility you can count on.” The numbers from EIA-861 equivalent reporting suggest some caution about that claim. The average CPP customer experiences approximately 1.18 outages per year and sits without power for roughly 162 minutes when they do.

Compare that to CEI — the private utility CPP was created to out-compete. CEI’s Customer Average Interruption Duration Index runs approximately 135 minutes. Its system average interruption frequency runs approximately 1.13 outages per customer annually.

Forty-one years of frozen rates produced a municipal utility whose outage duration is nearly 20% worse than the investor-owned utility it was supposed to keep honest.

This is not a condemnation of the line crews or the field engineers — it is a structural argument about capital investment. A system that cannot raise rates cannot fund capital replacement. A system that cannot fund capital replacement degrades. The physics of aging distribution equipment doesn’t negotiate with political convenience.

A note on the reliability numbers: I couldn’t locate SATI/CATI figures in CPP’s public filings, and they wouldn’t be in PUCO reporting either — municipal utilities aren’t subject to that framework. What’s cited here is drawn from EIA-861 equivalent data. CPP publishes annual reports at cpp.org; if those metrics live anywhere, that’s where to look.

The 2024 Reckoning

In November 2024, City Council voted unanimously to approve Ordinance 957-2024 — CPP’s first base rate increase since 1983.

The average residential customer using 500 kWh per month will see a $4.85 monthly increase in 2025, followed by an additional $3.45 in 2026. The typical summer bill moves from $80.72 to $90.52. In percentage terms: roughly 15% in year one, cumulative 25% by 2026.

Council held three committee hearings and six or seven public meetings before the vote. It passed unanimously. “We don’t take it lightly,” Director of Public Utilities Martin Keane told council. “There’s not a lot of meat on the bone, but we know we need to do better.”

The rate increase answers the immediate debt service problem. The question it cannot answer is whether 41 years of deferred capital investment can be recovered by a 25% adjustment on a 500 kWh monthly base. The honest answer is: not on that timeline, and not with those dollars. The debt service cliff was the forcing function. The infrastructure rebuild is a much longer and more expensive project that has not yet fully begun.

The Company That Tried to Take It

CEI — the Cleveland Electric Illuminating Company that Brock Weir’s banks tried to deliver Muny Light to in December 1978 — did not remain an independent entity. In 1986, CEI merged with Toledo Edison to form Centerior Energy. In 1997, Centerior merged with Ohio Edison to form FirstEnergy Corp.

CEI didn’t help found FirstEnergy. CEI became FirstEnergy. The corporate succession is direct.

That same institutional DNA:

Triggered the 2003 Northeast Blackout. FirstEnergy’s failure to maintain transmission line clearances in Ohio, combined with a software failure that left control room operators blind to cascading overloads on their system, set in motion the largest grid failure in North American history — 55 million people across eight states and Ontario without power. The U.S.-Canada Power System Outage Task Force identified FirstEnergy’s Ohio system as the initiating point.

Nearly caused a nuclear meltdown and then lied about it. In March 2002, maintenance workers at FirstEnergy’s Davis-Besse Nuclear Power Station in Oak Harbor, Ohio, discovered that boric acid corrosion had eaten a cavity the size of a football through the carbon steel reactor vessel head, leaving only a thin stainless steel liner to hold back 2,000 PSI of pressurized reactor coolant. Engineers had been submitting false inspection reports to the NRC for years, including oral testimony from one employee that he was “at peace in his soul” about inspection results that proved to be fiction. The NRC estimated that if Davis-Besse had operated another two to eleven months, a serious loss-of-coolant accident was likely. The incident ranks fifth on the NRC’s list of most dangerous nuclear events in the United States since Three Mile Island. The NRC imposed its largest-ever single fine — $5.45 million. The DOJ settlement was $28 million. Two employees were convicted.

Orchestrated the largest bribery scheme in Ohio history. Between roughly 2016 and 2020, FirstEnergy funneled approximately $60 million through dark money channels to elect and control the Ohio House Speaker, pass House Bill 6 — a $1.3 billion ratepayer-funded bailout of failing nuclear plants — and separately bribed the Chairman of the Public Utilities Commission of Ohio with $4.3 million disguised as a consulting fee. The U.S. Attorney called it “likely the largest bribery, money laundering scheme ever perpetrated against the people of the state of Ohio.” Former Speaker Larry Householder is serving 20 years in federal prison. Former CEO Chuck Jones and former Senior VP Michael Dowling were reindicted on 22 criminal counts in June 2026, following a mistrial in March. The PUCO chairman died by suicide while under indictment.

Kucinich told Council in 1978 that Cleveland Trust and CEI had “brought shame to this city.” He wasn’t wrong about the company.

What the Founders Knew

Tom Johnson built Muny Light as a check on monopoly power. The check worked — for as long as the utility was funded well enough to be competitive. Dennis Kucinich paid a significant political price to keep it in public hands. The city paid a price in default.

What neither the founders nor Kucinich could fully anticipate was that political ownership creates its own form of capture. The same democratic accountability that can prevent a private CEO from extracting monopoly rents can prevent a council from charging what service actually costs. The result, over four decades, was a utility that froze its rates, deferred its capital, and delivered declining performance to the lowest-income residents of one of Ohio’s most economically stressed cities.

The 2024 rate increase doesn’t fix that. It begins the conversation about fixing it. The question for the next decade is whether CPP can translate modest rate revenue into the infrastructure investment that its founding mission has always required — and whether the residents of Cleveland’s East Side will see “the utility you can count on” become something closer to true.

Kucinich was right to save it. The job of making it worth saving continues.

This is the third installment in the When the Clock Runs Out series. The May Mirage examined the gap between nameplate solar capacity and what the grid actually sees when load peaks. The Terminal Irony traced how Anchorage kept drawing from Cook Inlet long past the point where the reserve math still held. This one is about a municipal utility that froze its rates in 1983 and called it a gift. Three different assets, three different jurisdictions, three different clocks — but the same arithmetic. You can push the cost forward. You cannot make it disappear. The longer the deferral, the narrower the options when the bill arrives, and in each of these cases, it’s the people with the fewest alternatives who end up holding the largest share of it.

Cleveland Public Power publishes annual financial reports at cpp.org/About/Annual-Reports. Rate schedules are governed under Part V of the Municipal Utilities and Services Code, Chapter 523. PUCO has no jurisdiction over Ohio municipal electric utilities; reliability performance is not subject to the same mandatory public disclosure framework governing investor-owned utilities. EIA-861 reliability data for CPP is publicly available at eia.gov.

Hello readers. Let’s continue our conversation about poor decisions. Today’s piece links to an excellent article by Sarah Montalbano called “What’s next for Alaska’s Railbelt Utilities?” Sarah does a great job of covering the emerging emergency and where things stand now. But let’s take a hard look at how things got to where they are — and what the clock now says. In Part 1 of this series, “The May Mirage,” we noted a proposed 1.25 GW coal plant for Alaska’s Railbelt as one signal that the energy planning horizon isn’t as renewable-smooth as the headlines suggest. Today, that project gets its full context. The title of this piece has two meanings: the Kenai LNG terminal is the physical object, and the irony is terminal. The same facility that exported Alaskan gas to Japan starting in 1969 is now being pointed the other way. Alaska, floating on North Slope reserves large enough to supply the nation for decades, is preparing to import LNG by tanker to keep the lights on in Anchorage.

The Good Days

In the beginning, Cook Inlet had an excess of natural gas and the future looked bright. The Kenai LNG terminal opened in 1969 as the first LNG export facility in the United States. Natural gas was plentiful, money flowed into the region, and everyone was happy. Cook Inlet gas production peaked around 200 Bcf per year in the mid-2000s, then began a steady decline to the current level of roughly 70 Bcf annually. The first warnings of potential shortfalls appeared in 1998, but Hilcorp, the dominant producer in Cook Inlet, largely ignored them. They engaged in a pattern of short-term fixes and the familiar strategy of kicking the can. If those warnings sound familiar, they should, Matt Randolph – Mr Global, has been warning on Substack for a couple of years that the United States is exporting too much gas and too much oil. He is largely ignored because the money is good and kicking that can is easy. Alaska just got there first.

The Curve Gets Real

By 2015, production had declined to the point that exports were no longer viable. The Kenai LNG terminal went idle — ending an important revenue stream for the region and sending a signal that should have triggered serious action. It didn’t. Hilcorp continued on the same path of short-term supply agreements and cautious investment. The issue became a political hot potato. Local politicians chose not to directly deal with a potentially career-ending problem and left it for the next election. Proposals swirled, but none of them got the support to take root, and the can kept moving.

The Bill Comes Due

So here we are. The Cook Inlet gas supply is on track to fall below Railbelt demand by 2027 — a date the Alaska Department of Natural Resources moved up three years from its prior forecast. We have passed the point of no return. No major permanent infrastructure solution can be completed before the clock hits midnight. Here is where things stand:

1. The Railbelt utilities — serving the corridor from Homer to Fairbanks — depend heavily on Cook Inlet natural gas for power generation. That supply is leaving. Unlike the Lower 48, where a gas shortage can be offset by power imports from neighboring systems, the Railbelt runs islanded. When gas goes short, megawatts go short.

2. Much of the domestic heating is direct natural gas, same story.

3. There are competing proposals to convert the Kenai LNG terminal into a receiving and regasification facility for LNG imports. A second proposal — the Cook Inlet Gateway import terminal — targets the same peninsula. The utilities have split: Enstar is aligned with one project under an exclusivity agreement, Chugach Electric with the other. Both would cost hundreds of millions of dollars. The Regulatory Commission of Alaska is questioning whether the region needs two. Meanwhile, the clock ticks. Neither will be fully operational before the supply gap opens.

4. The North Slope gas pipeline — now branded as Glenfarne’s Alaska LNG project and carrying a $44 billion price tag — has been “imminent” through multiple developers and ownership changes since at least 2013. Note: this is not TAPS, the Trans-Alaska Pipeline System, which carries oil. The gas pipeline is a separate project that has never been built. Construction, once a final investment decision is made, would take years. This is a long-term solution to a near-term problem.

5. The Terra Energy Center, a 1.25 GW coal plant proposed for the Railbelt, has taken a significant step: a roughly $1 billion boiler agreement with Hyundai Heavy Industries Power Systems — the first such order for a U.S. coal plant since 2006. But coal plants take approximately ten years from commitment to commercial operation. Terra Energy Center cannot be online before the gas cliff arrives. The access road needed to move coal from the mine to the plant must cross more than 180 streams, and the environmental and permitting path for that corridor is far from cleared. Most telling: No Railbelt utility has signed a Power Purchase Agreement (PPA) with Terra Energy Center yet. That silence speaks louder than any press release.

6. Diesel may be the last-ditch backstop. Caterpillar and Cummins specialize in responding to situations like this — at a non-negotiable price. Alaskan villages already running on diesel are paying up to $1.31/kWh. For context, Golden Valley Electric Association, connected to the Railbelt, was already at $0.30/kWh in late 2025 against a US average of $0.13. Diesel doubles that again. For a ground-level education on what it takes to keep diesel generation alive in the Alaska environment, check out Wrenchin_The907 on TikTok for an education on what it takes to keep diesel generation running.

The lesson here is that this problem was well known; it was very apparent it was coming, and it was ignored because dealing with it was politically and economically painful. Both the communities and those that serve them took the low road of ignoring the problem, perhaps hoping it would just go away. The can has met the wall, and there is no longer a place to kick it. That doesn’t mean some don’t still want to try. But now the choices are small and expensive. One irony worth naming before we close: the company building the rescue terminal is Harvest Midstream — an affiliate of Hilcorp, the same producer whose refusal to commit to new supply contracts after 2022 started the current scramble. The rescuer and the problem share a parent company. Whether you find that reassuring or alarming probably depends on which side of the rate case you sit on. What’s your lesson learned? Share it in the comments. Next week we talk about another utility that’s been kicking that can down the road, the building is ready to fall in on the can. Thanks for reading!

This headline popped up in the online publication of our fellow Substack member CleanTechnica. The article was titled Despite Trump’s Attacks, Solar Outperforms Coal in U.S.. If you read the fine print at the bottom, it is actually a press release by The Sierra Club, drawing on generation data from Ember Energy, and you without question saw it pop up in other news outlets, I did.

To be fair, on the surface, this press release is factual, in May. What the article wants to lead you to believe is this is a trend, which it is not, and that is the art of a well written piece of propaganda. Let’s be fair, everyone uses propaganda, and everyone tries to paint a rosy picture of their particular position on the issue/industry they want to promote, no crime there, but let’s peel off the wrapping and get down to facts.

Every May and October you can expect to see these breathless claims that renewable energy is winning the war because this month, we won the battle, to continue. The reality is that the industry calls these “shoulder months”. This is the time when system demand is at its lowest, and so are energy prices. So, expect that renewable projects still receiving contracted government support will be able to operate in near zero or below zero energy markets and still make a profit. These facilities must operate to make the energy production quotas built into the subsidy programs. Traditional generators do not have that luxury, unless they are operating under a reliability must run contract. These shoulder months are also plant outage season when much of the US generation fleet comes out of service for maintenance and repairs. Insiders understand the dynamics, but it makes good press.

What this story doesn’t discuss is what this very cold winter brought to the surface. For the first time in five years, the first quarter of 2026, coal actually saw market gain. This was driven by very high Henry Hub natural gas prices. For those that don’t know, Henry Hub is the market that sets the base natural gas clearing price for the majority of the nation, and City Gate prices go up from there at the various delivery points across the country. The only place lower is the Permian Basin because there isn’t enough pipeline capacity to move the co-gas production out. (co-gas is a byproduct of oil production). The dark spread — the margin between wholesale power prices and coal generation costs — reached the point where it was economical to burn coal for the first time in many years and burn coal, they did.

With the shutdown of government subsidies, the future investments in renewables have fallen virtually flat. The new proposed project pool has all but disappeared except in places like New York that provide state funding. This does not speak well to the future takeover by renewable energy. The end of subsidies had to come sometime; the subsidy program was unsustainable. Whether renewables can stand without subsidy support is the test we are now running in real time. What this really speaks to is that we are reaching capacity on our natural gas infrastructure. We are taxing seasonal storage more than ever, we are running pipelines at near full capacity, we do not have room for more production. Yet we continue to add load. One issue is that idle capacity is not profitable, but necessary. We operate in a short-term investment world that wants immediate returns on every dollar invested and operates on a two-year investment cycle. As we continue to put pressure on existing infrastructure, that stress will show up in market price. Investors view high prices as positive, not negative, but that puts us in a very vulnerable spot.

The impact on natural gas is starting to show up in long term planning. Terra Energy Center near Anchorage, Alaska has already ordered coal fired boilers for a new power plant — the first such order placed for a US plant since 2006. That’s beyond just planning; that’s action. The second new-build proposal, TerraSpark at Mount Storm, West Virginia, adds carbon capture to the equation — and just announced on June 8th that Babcock & Wilcox will supply four 400-MW supercritical boilers for that project as well. Basin Electric isn’t far behind with the Dry Fork FEED study fully supported by the State of Wyoming to add on to their Gillette coal plant. Talk about a new coal plant would have been unheard of five years ago, now we are ordering boilers.

What’s driving this is the current investment culture doesn’t fit the infrastructure demands. Power plants, pipelines, even new coal mines take decades of planning. Just the permitting can take almost ten years, especially with the never-ending legal challenges produced by NGOs like the Sierra Club. Yes, the environment deserves its day in court, but one day, or even one year, not one decade of repeated challenges.

Next, there are two case studies in separate pieces talking about the results of short-sighted investment, and short-term political goals. In both cases, it has left quite a mess and utilities in some very tight corners. Tell me what you think about where we are. Thank you for reading!

The Edmonston Pumping Plant

The A.D. Edmonston Pumping Plant lifts water from California’s Central Valley over the Tehachapi Mountains to the Los Angeles basin — a vertical rise of 1,926 feet (about twice the height of the Empire State Building), the highest single-lift pumping operation in the world outside of pumped storage. The plant was built with fourteen 80,000 horsepower (60 MW) synchronous motors. Four of the original Westinghouse units have since been replaced with Hitachi equipment; the reactive power characteristics of those replacement units are pending verification and warrant confirmation from DWR and CAISO as part of any formal reliability assessment of this corridor. The remaining original units are synchronous motors capable of generating reactive power to actively support system voltage.

The station is served by a dedicated 230 kV transmission line from Pastoria Substation in Southern California Edison territory. Pastoria sits on Path 26, directly adjacent to the junction with Path 15 — the primary transmission corridors connecting California to Arizona in the south and to Northern California in the north. This geographic position is critical to the regional reliability argument developed below.

Starting those massive motors without collapsing the transmission system requires a specialized solution. Two Motor-Generator (MG) sets accomplish this by starting each pump motor on an isolated bus, ramping voltage from near-zero through controlled field excitation of the MG generator, and synchronizing the motor to the grid only after it reaches synchronous speed. The grid never sees the locked-rotor inrush. Without this system, starting a single 60 MW synchronous motor could demand up to ten times its running load — nearly the entire capacity of the Pastoria 230 kV feed consumed by a single pump start, almost entirely as reactive current. Regardless of the starting method, the EPP represents a massive synchronous load on the regional power system at all times during operation.

Voltage and Inertia: The Twin Pillars

Much is written about inertia, and inertia is extremely important. But voltage builds the magnetic bonds that hold the power system together. Inertia keeps frequency from dropping out the bottom, or shooting over the top, every time there is a shift in system load. Voltage holds the power system together so that inertia can deliver that stabilizing energy. Without either, the power system tears itself apart.

What happens when a power system loses its ability to control voltage? Unfortunately, we now have a recent and well-documented example, and the parallels to California’s current direction are uncomfortably close.

The Iberian Blackout: A Case Study in Voltage Control Failure

On April 28, 2025, the electric power system on the Iberian Peninsula collapsed, creating the largest blackout in European history in more than two decades. Two earlier analyses — The Iberian Blackout — On That Day and The Iberian Blackout — The Reports — documented the operational sequence before any official reports were released, and those analyses held up. The official ENTSO-E report, published eleven months after the event, was politically filtered to avoid direct assignment of blame. Its headline conclusion — “the problem is not renewable energy, but voltage control, regardless of the type of generation” — is technically accurate in the narrowest sense and operationally misleading in context. The voltage control capability that was missing was precisely the reactive power support that synchronous machines provide inherently, as a byproduct of their physics.

Spain had followed the popular push toward an IBR-based power system without the engineering discipline that responsible high-IBR integration requires. The contrast with Ireland is instructive. EirGrid and SONI — operators of one of the highest IBR penetration systems in the world — recognized early that non-synchronous penetration above 50% required extraordinary engineering investment to manage safely. They built a formal multi-year program called DS3, raising their System Non-Synchronous Penetration (SNSP) limit from 50% to 75% through five incremental operational trials. Demonstrated engineering capability was required before each step was permitted.

Critically, Ireland backed that policy with hardware. EirGrid procured six synchronous condenser projects delivering nearly 7,000 MVA.s of low-carbon rotating inertia — including the world’s largest flywheel at Moneypoint, which alone provides 3,500 MWs of inertia. A hard system inertia floor of 23,000 MWs is enforced at all times. That is the engineering price of high IBR penetration done responsibly: rotating copper and iron, procured deliberately and sized to the task.

Spain reached similar IBR penetration levels with none of those investments in place. There were no synchronous condensers on the Spanish mainland, no grid-scale battery storage, and IBRs were allowed to operate in power factor control mode rather than voltage control mode, removing what little reactive support they could have provided. A political tax on nuclear generation had incentivized nuclear plants — the primary source of rotating reactive energy — to stay offline on the day of the collapse. Spain built a perfect trap. On April 28 it closed, and the grid fell in under two minutes.

The critical takeaway is not that Spain specifically ran out of the ability to reduce voltage. It is that Spain lost the ability to control voltage at all. It does not matter whether control is lost to the high side or the low side. Loss of voltage control means system collapse. Notably, a Bloomberg investigation published in August 2025 found that Spain invested only $0.30 in grid infrastructure for every dollar spent on renewables — against an EU average of $0.70. The UN Secretary General stated the ratio should be one to one. California’s grid infrastructure investment trajectory mirrors Spain’s, not Ireland’s or the UK’s.

The California Parallel

The parallels between pre-blackout Spain and California’s current trajectory are not subtle. California’s SB 100 mandates 100% zero-carbon electricity by 2045. The CPUC’s Integrated Resource Plan proceedings have systematically favored IBR resources over dispatchable synchronous generation. The original decision to close Diablo Canyon — reversed only after the 2020 rolling blackouts made the reliability gap impossible to ignore publicly — was made on political grounds, not on the basis of engineering analysis.

Southern California’s structural synchronous grid inertia (SGI) deficit began with the shutdown of San Onofre Nuclear Generating Station (SONGS) in January 2012. SONGS provided substantial localized reactive support and synchronous inertia to the southern California coastal transmission system. Nearly all generation capacity added to serve southern California since 2010 has been inverter-based, contributing negligible synchronous grid inertia. The claim that distant synchronous generation compensates for this deficit is contradicted by the physics of AC transmission — reactive power loses effectiveness with distance and impedance, and the constraints on high-voltage AC transmission lines from the Pacific Northwest mean that remote generation cannot substitute for the localized reactive support SONGS once provided. The closure of Diablo Canyon would deepen this deficit further.

California is removing rotating inertia and traditional reactive resources from its power system, replacing them with IBRs and batteries on the unproven assumption that they are a fully functional replacement for rotating resources. Like Spain, California has avoided adding the flywheel-equipped synchronous condensers that European grid operators recognize as essential in high-IBR systems — investments that would cost California ratepayers billions of dollars for equipment that produces no electricity. IBRs, even when voltage support is fully enabled, provide extremely limited reactive capability and can never perform beyond their rated current. There is no thermal reserve in a power electronic device. Rotating copper and iron, by contrast, can sustain operation well beyond nameplate ratings for short periods, responding to system disturbances naturally and without programming — delivering both inertia to stabilize frequency and reactive energy to control voltage. Physics does not respond to legislative mandates.

Diablo Canyon, Edmonston, and the Path 26 Corridor

The maps below, sourced from the WECC 2024 Path Rating Catalog, show the transmission geography that defines this argument. Path 15 runs along the California coast, with Diablo Canyon at its southern anchor. Path 26 runs from Midway substation southward to Vincent, passing through the Pastoria area where Edmonston’s 230 kV feed originates. Midway substation is the geographic junction between the two paths — the point where Diablo Canyon’s reactive support and Edmonston’s reactive demand share the same regional transmission picture.

Path 15 including Diablo Canyon. Source: WECC 2024 Path Rating Catalog

Path 26, Midway to Vincent. Pastoria Substation — the dedicated 230 kV feed for Edmonston Pumping Plant — sits along this corridor. Source: WECC 2024 Path Rating Catalog

Diablo Canyon sits on California’s central coastal 500 kV transmission system. Path 26 runs from the Tehachapi area southward toward the Los Angeles basin. Edmonston’s dedicated 230 kV feed originates at Pastoria Substation on Path 26. These facilities share the same regional transmission picture. The reactive support that Diablo Canyon provides to the coastal transmission system is in the same corridor as the reactive demand that Edmonston represents on Path 26. Removing Diablo Canyon’s synchronous generation does not simply retire 2,200 MW of carbon-free electricity — it removes dynamic reactive support from a corridor that already carries a large synchronous motor load whose reactive behavior during a voltage disturbance is significant and fast.

Diablo Canyon is already paid for. It is well maintained and designed to operate for a century. It currently provides more synchronous grid inertia than any other generation facility in California. The alternative — replacing its SGI contribution with flywheel-equipped synchronous condensers — would cost ratepayers billions of dollars for equipment that produces no electricity. The Siemens Energy Rotating Grid Stabilizer whitepaper documents what this hardware costs and how long it takes: even a brownfield conversion of a retired power plant generator takes six months from engineering study to commissioning and runs up to 50% of greenfield cost. Ireland’s experience at Moneypoint and the UK’s Stability Pathfinder program document the scale of investment required across an entire grid. California would be paying billions to partially replicate a service currently provided for free as a byproduct of an already-operating, already-paid-for nuclear plant.

The Fault Scenario: Theory Grounded in Documented Precedent

The most credible failure scenario for the Edmonston corridor involves a three-phase transmission fault with a breaker failure on Path 26. This is not a hypothetical. It has documented precedent in the same regional grid involving another nuclear plant.

On June 14, 2004, at 07:40:55 in the morning, a single phase-to-ground fault occurred on the Western Area Power Authority’s 230 kV Liberty–Westwing line near Phoenix, Arizona. The initiating cause was a heron taking flight from the line. What should have been a routine fault clearance became a regional cascade because two contacts in a single Westinghouse AR auxiliary relay — manufactured in 1974 — failed to close. Breaker WW1022 did not trip. Because both the trip signal and the breaker failure initiation signal shared the same failed contacts, the breaker failure scheme also did not fire. The fault could not be cleared locally within the 230 kV yard. It propagated to the 525 kV system through the Westwing transformers, and the remote ends of every line feeding Westwing began tripping to isolate what the protection could not contain locally. All three Palo Verde nuclear units tripped. Redhawk generating units tripped. Approximately 55,000 customers across Arizona Public Service and Tucson Electric Power lost power. The entire cascade originated from two relay contacts that failed to close in a 30-year-old electromechanical relay.

That is the initiating event topology for the Edmonston scenario. A single protection failure on a 230 kV line in the Path 26 corridor, breaker failure signal lost, fault propagates to the 525 kV system. What follows next depends entirely on what reactive resources are available to recover voltage as the system fights to contain the disturbance.

In 2004, Palo Verde was online. Its synchronous generators responded. The system recovered. In the IBR-dependent system California is building, that reactive recovery response does not exist in the same form.

Whether the EPP synchronous motors — including the Hitachi replacement units — are equipped with Automatic Voltage Regulators (AVRs) is a detail that warrants immediate verification by CAISO and DWR as part of any formal reliability assessment of this corridor. An AVR-equipped synchronous motor can act, within limits, as a synchronous condenser — actively supporting voltage recovery after a fault clears. Without AVRs, that capability is absent and the motors become a pure reactive burden on a system already struggling to recover.

Once the fault clears, the power system requires an immediate injection of MVARs to recover voltage and arrest the advancing power angles. The synchronous generators at Diablo Canyon would respond naturally — increasing excitation and injecting reactive energy into exactly the corridor that needs it. In a system without that synchronous generation, that reactive response does not exist. BESS systems can only deliver rated current, split across competing demands for voltage support, frequency response, and energy delivery simultaneously. As the EPP pumps continue to torque toward loss of synchronism, reactive demand increases further. A fixed field current delivers no additional reactive energy to pull power angles back. As voltage falls at the distribution level, motor loads demand more reactive power to avoid locked rotor. IBRs begin tripping on self-protection protocols, accelerating the collapse. There is a moment at which the system either recovers or it does not.

Conclusion

The glue that holds a power system together during a major transmission event is available reactive energy for voltage control. IBRs do not have it in sufficient quantity, and that deficit cannot be engineered around with current technology at the scale California requires. On a system without adequate synchronous generation, the Edmonston Pumping Plant can pull the regional grid apart.

The CPUC should require CAISO to measure synchronous grid inertia in real time rather than rely on models. Technology to do this exists today. GridMetrix, developed by Reactive Technologies, provides direct real-time measurement of grid inertia using a power pulse technique validated on grids worldwide. Currently, grid inertia is modeled, not measured — causing operators to make consequential decisions based on estimates rather than data. The structural SGI deficit in southern California that this paper describes is measurable. California is making irreversible infrastructure decisions affecting 40 million people and one of the world’s largest economies without knowing the real-time state of its own grid’s stability margin. The CPUC has the authority to require CAISO to deploy real-time SGI measurement and to demonstrate that any proposed generation retirement or IBR addition maintains adequate system strength. That requirement should be in place before any further synchronous generation is retired.

There is a place for all types of energy on the power system. But the composition of that mix must be determined by engineering physics, not political mandate. The Iberian Peninsula demonstrated what happens when that boundary is crossed. Ireland demonstrated what responsible high-IBR integration actually costs in hardware and engineering discipline. Diablo Canyon is already paid for, already operating, and already providing the synchronous grid inertia that California cannot afford to lose. The CPUC has the authority — and the obligation to ratepayers and to public safety — to ensure that California’s energy transition is governed by engineering analysis, not by legislation alone.

I will be offline the next ten days while we say goodbye to my brother, and take care of family/work needs out West, I do have a piece queued to drop on Sunday. We are taking time to visit the North Coast California Redwoods while we are out there, my happy place.

Holding the Voltage — How Power Plants Learned to Regulate Themselves

Colgate-Folsom Series, Part 3: From Handwheels to Vibrating Contacts

Hello readers, today we continue the conversation about the formative days of power system development. Let’s venture back onto the plant floor again in those pioneering powerhouses.

The Plant Floor

The conditions were not a pleasant day in the office. The machinery, even in hydro plants, gave off a loud din of mechanical noise. The laminations in the load windings hummed at 60 Hz in tune with the system frequency; if the system slowed down, they could hear it, and they could feel it through the vibrations in their feet. The generator windings, open to the air in the generator hall, threw off tremendous heat, making the room a hot place to work. The room was filled with the smells of warm insulation, warm lubricating oil, and working equipment. Smell alone could warn an operator when something was amiss and needed to be addressed. To the operators, these were living, breathing entities that they loved and served.

The Central Bus System

Early on, voltage—the thing that makes light bulbs bright and causes them to dim when voltage drops—was an entirely manual operation. The early plant had a “central bus system” with a single standalone, or two separate, DC “exciter” generators, often just called exciters, serving all the large AC generators in the plant. The DC current was used to supply the field windings in the main generators. Folsom had two standalone water turbines with DC generators; one was capable of running the entire plant. DC current to the main generator field winding was controlled by large rheostats (adjustable resistors), in series with the main field. The DC voltage for the entire DC bus was controlled by adjusting the field coil rheostat for the DC exciter.

One thing to make clear: adjusting the voltage to an individual AC generator connected to the same bus as other AC generators will produce circulating currents between the units. All generators connected to the same bus need to be adjusted to the same relative voltage to minimize these circulating currents. That would be the job of a very experienced senior operator.

Exactly how voltage was controlled at these old central-bus, multi-unit, manual plants is lost to history. But we can guess that once individual machines were set, the entire plant was controlled with adjustments to the exciter generator, and the main generator settings stayed static.

For more about this, take a look back in the early part of my Substack where I talked about basics, in Power Systems — Playing With Magnets.

The First Automation

Automation came much quicker to voltage control than it did to frequency. Voltage changes fast, and low voltage has a detrimental effect on system operation. So, voltage became the natural first target, and the ability to get there was a much straighter path.

The first step was to automate the rheostat on the exciter generator. This method was crude and is poorly documented, but abandoned equipment shows up in photos suggesting it may have been more common than we believe. A DC motor was used to drive a reduction transmission, and the output of the transmission drove a chain sprocket. A chain was looped around that sprocket and around a sprocket attached to the input of the manual rheostat. The motor was controlled by a contact-making voltmeter: if voltage got too high or too low, the motor was energized and slowly moved the rheostat in the correct direction until the meter moved enough to open the contact. This system is documented on a synchronous condenser installed at North Tower on the PG&E 60 kV line to Oakland.

The Tirrill Breakthrough

Next came the General Electric Tirrill automatic voltage regulator (AVR), which came into service around 1902. The Tirrill AVR was a massive leap in technology. Instead of using series resistance, it used vibrating contacts. Each time the contacts closed, current flowed to the exciter field winding; each time they opened, the current was interrupted. What mattered was the average field current over time. As voltage sagged, the average current increased; as voltage swelled, the average current decreased. The downside was the constant vibration wore the contacts, so the Tirrill required regular maintenance.

The Tirrill also contained the internal circuits to automatically balance out the circulating currents between two generators on the same bus. This became known as reactive droop compensation, and it is still used today. The Tirrill became the widely adopted standard for almost two decades. However, adopting the Tirrill was more than just plugging in a new regulator.

The End of the Central Bus

Adopting the new regulator brought an abrupt end to the central bus single-exciter-generator system. By the nature of how it worked, the Tirrill needed to work with small currents, so trying to control a single exciter for an entire plant was out. The Tirrill needed each main generator to have its own exciter.

This new requirement forced changes to the generators themselves. New equipment came with exciters driven by flat belts off the main shaft; this was often an additional belt drive to the one driving the flyweight head on the governor. These new belt drives needed tensioner mechanisms because the exciter ran under load. An alternative was a motor-generator (MG) exciter mounted close to the main generator. The trouble with MG exciters was that they offered no black start capability. It is likely in practice that MG exciters were mostly used to convert old central bus plants to Tirrill AVRs.

The Current Wall

As metallurgy continued to improve, exciter mounting was moved to the main shaft behind the main field winding. Then, as machines continued to grow, the exciter field current got too high for the AVRs, so a “pilot” exciter was added behind the main exciter to add a third stage to the control loop.

What Came Next

The Tirrill design was eventually superseded by the Brown Boveri / Allis-Chalmers rolling contact regulator, then the Westinghouse Silverstat silver button design. The evolution continued, but that reaches beyond the scope of this early design series.

Conclusion

This piece completes this series on the early days of the power station industry. Next week step back into the modern era with a requested piece by our own Green Nuke. We talk about voltage support, and go over some examples of what happens when voltage support is not sufficient on a dynamic power system, as well as how modern IBR based resources are playing into this potential problem. If you have some stories about vintage AVR systems you could share, or comments on something i missed in this series, I always love your feedback.

References

Bragg, Geo. H. “The Voltage Regulator at North Tower.” Pacific Gas and Electric Magazine, Vol. II, No. 12, May 1911, pp. 441–444.

Earlier in This Series

Power Systems — Playing With Magnets — The basics of generator excitation and voltage control.

Where It Started — Power System Interconnections

Where It Started — The Men Who Held the Frequency

Welcome back to the early twentieth century, where all of this began. We’re still in the PG&E system, mostly because it’s so well documented. As I said in Part 1, it’s easy to claim “first,” but harder to prove. Let’s just say it’s the earliest documented interconnected power system I’ve found. This time, let’s look at what it was actually like to operate it.

The Signal Lamp

Put yourself inside the Colgate Powerhouse on the Yuba River, circa 1903. You’ve drawn the rotation as the control operator on the frequency unit for the Bay Counties Power Company grid. Your generator is coupled to a Pelton impulse wheel, controlled by a Lombard hydraulic governor — state of the art for its time. Like all governors of the era, the Lombard operates on a set of flyweights, belt-driven off the main generator shaft. It uses a small pilot valve operated by the flyweights to drive a pressurized hydraulic cylinder that adjusts the Pelton needle valves. As customers switch electric load on and off in the distant Bay Area, the Lombard responds by adjusting water flow to keep frequency relatively close.

But close isn’t close enough. Like all governors of this era, the Lombard works on a droop-feedback loop. It holds speed in a narrow band, but it can’t return frequency precisely to 60 cycles on its own after a load change. That’s your job. You are the missing control loop.

To make the fine adjustments the governor can’t, you have a collection of hardware — nuts, washers, bolts of various weights — that you hang on the governor balance shaft to bias the speed set point up or down. You’re watching a signal from the switchboard operator on the floor above: a red incandescent lamp, flashing to tell you which way to go. Add a washer. Remove a bolt. Watch the lamp. The frequency drifts, you adjust. It drifts again, you adjust again. There is no pause button.

This was a mentally exhausting job. Every load change on the entire interconnected system — every motor start, every light switch, every streetcar leaving the barn — showed up as a speed disturbance on your machine. You couldn’t look away. You couldn’t take a break. And this was not a safe working environment by modern standards. The generators were not enclosed. It was noisy. There were few guards over large pieces of unforgiving rotating equipment. Getting injured was easy if you lost focus, and workman’s compensation was decades away.

That’s why the frequency control position had to rotate. Nobody could sustain that level of concentration for a full shift. The operators tag-teamed — not as a luxury, but as a necessity. The job would break you if you held it too long.

The method improved over the next few years. By 1909, the collection of scrap hardware had been replaced by a small electric motor mounted on the governor arm, driving a traveling weight back and forth. The switchboard operator no longer flashed a lamp — he operated a two-direction switch handle to drive the motor and nudge the governor’s speed set point up or down. The operator was still the control loop, still making every decision, but now he was throwing a switch instead of stacking washers. It was the same job with better tools, and it was still exhausting.

The Dance on the Plant Floor

You weren’t alone on that plant floor. While you chased frequency on the lead unit, other operators managed the remaining generators — the follower machines running in droop mode. Their job was equally critical: they watched your output and adjusted their machines to keep you near the center of your operating range. If your unit was running up near its high limit, they picked up load on another machine to pull you back down. If you were running too light, they shed load off another unit to give you room below.

This was the inside-the-plant coordination layer — a continuous two-person dance. The frequency operator chased the moment-to-moment load swings. The second operator managed headroom, so the frequency unit didn’t hit its ceiling or its floor. They’d swap positions because the frequency role demanded a level of concentration the headroom role didn’t. It was a tag team born of operational reality.

Why did frequency matter so much? Because of the proliferation of AC motors across the served load. As the frequency changed, so did the rotational speed of every motor on the system. Customers had a reasonable expectation that their motors would run within a narrow speed range. The width of that range decreased over time as control methods improved, but even in 1903, “close enough” had to be pretty close.

The System Explodes

The size of the PG&E system exploded over a few short years. In 1899, it was a small three-station system under the Yuba Electric Power Company. By 1909 — just ten years later — it had grown to 67,000 kilowatts across ten hydro-electric powerhouses, a 9,000 kW General Electric Curtis steam turbine in Oakland, additional steam plants in San Jose and San Francisco, and a reciprocating gas engine station at Martin near Brisbane.

The Oakland steam plant is worth noting. F.H. Varney, PG&E’s Engineer of Operation and Maintenance, described its construction in the same July 1909 issue of PG&E Magazine. Ordered in July 1908, the plant was running under its own steam 162 days later — in time for the Christmas load. Varney made the case that the steam turbine occupied one-tenth the floor space of a comparable reciprocating engine plant, and it didn’t need to be jacked over by hand for maintenance. This was PG&E betting on the future of prime mover technology.

The gas engine station at Martin was the opposite bet, and it lost. Martin Station sat on the property of the Oakland Gas, Light and Heat Company — a PG&E subsidiary — and ran its reciprocating gas engines on the same coal gas that fueled the city’s gas lighting system. Coal gas was produced by heating coal in retorts, and its thermal quality varied with every batch. Feed an engine fuel that’s changing its energy density in real time and the engine speeds up on a rich slug, slows down on a lean one. The governor chases a moving target it can never catch. Add the corrosive byproducts of coal gas combustion — hydrogen sulfide, ammonia, tar vapors — eating through valve seats, cylinder walls, and piston rings, and you had a station that pulsated, corroded, and was abandoned by 1910. Martin Station didn’t fail because gas engines were bad. It failed because coal gas was an impossible fuel.

Each new source that connected to the system made the frequency control operator’s job harder. Every generator added another set of electrical dynamics. The steam turbine in Oakland responded to load changes differently than the Pelton wheels at Colgate. More sources, more lines, more load — and still one operator on one frequency unit, throwing a switch handle to nudge a governor that couldn’t hold frequency on its own.

The Breaking Point

This explosion of growth forced PG&E to literally invent what we now call the power dispatch system — a system that is still the foundation for grid operations to this day.

P.M. Downing, PG&E’s Engineer of Operation and Maintenance, described this evolution in his July 1909 article “The Load Dispatching System of the Pacific Gas and Electric Company” in PG&E Magazine. What Downing documented was a system that had outgrown every organizational structure the company tried.

The Bay Counties Power Company — the nucleus of what became PG&E — originally handled everything from Colgate. Speed regulation, voltage regulation, line switching for the entire system, all controlled by the station operators at the master plant. That arrangement worked when the system was small. Then the extension of transmission lines brought the important center of switching near Davis, and for a time the operators there handled what had formerly been controlled from Colgate. Then PG&E absorbed the Standard Electric Company and had to operate that system in parallel with the Bay Counties grid. By virtue of its location, South Tower in Contra Costa County by the Carquinez Strait was selected as the common control point for both systems.

About a year and a half before Downing wrote his article, PG&E made the decisive move. The regular station operators were relieved of dispatch responsibility entirely, and the company created the dedicated position of Load Dispatcher. The job had outgrown the operators doing it at every stage.

The Dispatcher’s Office

The Load Dispatcher’s office became the nerve center of the entire system. From that office, every powerhouse and switching station received its orders. No superintendent could take a powerhouse, transmission line, or any other part of the system out of service without first receiving authority from the Load Dispatcher.

But the dispatcher was not controlling frequency. That was still the job of the operator on the frequency unit at the designated control plant. The dispatcher’s job was managing the system, so the frequency control plant had room to work. He distributed load among the powerhouses to keep the frequency unit near the center of its operating range. He tracked the status of approximately 125 oil switches and 350 air switches using physical dummy models on a large diagrammatic wall board — the first iteration of what we now call a mimic board. Oil switches were represented by circular dummies, air switches by rectangular ones. When an order was given to operate a switch, no change was made on the board until the operator reported the order carried out. A dispatcher coming on watch could tell at a glance what lines were out of service and what switches were open or closed.

The dispatcher also managed the daily rhythm of the system. His office kept careful records of energy delivered from every generating station. Daily load curves were plotted showing hydro generation, steam generation, purchased power, and total system load. That information was compiled between midnight and 8:00 a.m. and delivered to the general office as a daily service report by 8:30 each morning. The report also included interruption details, weather conditions, rainfall, snowpack at gauging stations, reservoir levels, and ditch flows — the raw hydrological data that determined how much generation the system could sustain.

But Downing made clear that load balancing was not the hardest part of the job. Re-establishing service after an interruption was far more difficult, and it demanded quicker, more decisive action. With everything running together on a common 60 kV network spanning sixteen hundred miles, trouble on any line could affect the entire system. The trouble had to be located, the faulted section isolated, and the rest of the system restored — all by telephone. Generators could be thrown out of synchronism. In severe cases, individual machines within a powerhouse could lose synchronism with each other. The system of switching in use would automatically separate the powerhouses when trouble occurred, leaving each running with whatever lines and load it could carry. Then the dispatcher had to put it all back together.

As the system continued to grow beyond what Downing described, the dispatcher’s responsibilities would expand further. With enough generation behind multiple transmission paths to the same load center, power flow divides according to the impedance of each path — not according to what the dispatcher wants. Managing where power flows, not just how much, becomes essential to keeping transmission lines within their thermal ratings. Overload a line and you’re not just tripping breakers — you’re risking conductor sag, physical damage, or worse. In 1909, with limited paths and switching points, line loading was likely implicit in the switching decisions. But the problem was coming, and it would only get bigger.

This critical work was in the hands of three dispatchers, rotating around the clock. Fred R. George served as Chief Dispatcher, assisted by C.P. Pierce and W.D. Skiner. Downing’s wife, Frances Stevenson Downing, contributed a piece called “Men Must Work” to the same July 1909 issue of PG&E Magazine. In it she wrote about life as the wife of a powerhouse engineer at a remote hydro plant — the isolation, the engineering magazines on the library table, the heroes hidden deep in the canyon that the city office seldom heard of. The human cost of keeping the lights on has always been a family affair.

The Wider Interconnection

Further complicating the dispatcher’s job was the fact that PG&E was not operating alone. Downing’s article lists four independent utilities running in parallel with the PG&E system:

Northern California Power Company — Pit River country, Shasta County. Great Western Power Company — Feather River, Big Meadows and Lake Almanor. Stanislaus Power Company — Stanislaus River. Snow Mountain Water and Power Company — Lake County.

All four were separate companies at the time, each with their own generation and their own operators. But they were electrically tied to the PG&E system, and from the information available, PG&E appears to have taken on frequency control responsibility for the entire interconnection. That’s not just load dispatching — it’s the embryonic version of what we now call a Balancing Authority. PG&E was holding frequency for companies it didn’t own and couldn’t directly command, coordinating over leased telephone lines that ran from San Francisco to Oakland to South Tower, Stockton, Sacramento, Marysville, and Chico, with branches to Electra, Davis, Mission San Jose, de Sabla, and Centerville.

Regardless of the contractual details, those independent utilities would have been responsible for keeping the PG&E Load Dispatcher informed at all times of the status of their systems and the estimated power exchange between companies. The phone system wasn’t just a convenience — it was the only real-time coordination channel between the entity controlling frequency and the generation resources it needed to manage. The phone booth still standing at the Folsom Powerhouse State Historic Park is physical evidence of that link. Folsom was on the other end of those calls.

Every one of those independent companies would eventually be absorbed by PG&E. That wasn’t a coincidence. When you’re already operationally responsible for coordinating the interconnection — holding frequency, dispatching load, managing trouble restoration across company boundaries — buying the companies just makes the org chart match the operational reality.

The Long Road to Automation

There were attempts to take the human out of the frequency control loop. In 1913, Thomas Fogalsang wrote about an early concept in his PG&E Magazine article “The Development of Automatic Regulation for Turbo Generators” — using a contact-making frequency meter with a dual-coil design to automatically bias the governor. One coil used non-inductive resistance, the other used inductive resistance. When frequency drifted, the torque imbalance between the coils would close an electrical contact and energize a relay to drive the governor motor automatically. It was an elegant idea on paper.

Fourteen years later, in 1927, Leeds & Northrup and General Electric deployed the first automated frequency controllers in field tests on interconnected grids — an evolution of exactly the concept Fogalsang had described. But these early systems looked only at frequency, and that created a new problem. On an interconnected grid, a load change in one utility’s territory drops frequency across the entire system. A purely automatic controller at a neighboring utility would aggressively ramp its own generation to fix the frequency — accidentally shoving massive unscheduled power across the tie-lines, overloading transmission, and creating financial disputes over unintended power transfers.

The problem wasn’t truly solved until the 1930s, when Nathan Cohn of Leeds & Northrup invented Tie-Line Bias control. Cohn’s system introduced a new metric called Area Control Error — ACE — that considered two variables simultaneously: is the system frequency off, and is the net power flowing across our tie-lines matching what we agreed to exchange? Only when both conditions were monitored together could automatic generation control work safely on an interconnected grid. That invention — AGC operating from centralized analog computers — finally displaced the manual operator at the frequency control unit, and the modern power system truly began to take shape.

But every line of code in a modern AGC system, every Area Control Error calculation, every tie-line bias setting traces its lineage back to a man standing next to a governor in a remote canyon powerhouse, stacking washers on a balance shaft while a red lamp flashed above him. The automation didn’t replace knowledge — it encoded it. And the knowledge came from the men who held the frequency, one washer at a time.

Long-term frequency tracking got more sophisticated along the way. In 1916, the Warren Telechron Master Clock gave the switchboard operator a precise tool for measuring accumulated frequency error. The instrument used three separate dials mounted vertically — a standard 12-hour clock face on the bottom, and two upper dials scaled 0 to 4, manufactured by Telechron Inc. of Ashland, Massachusetts. The upper dials displayed the accumulated time error: the difference between real time, kept by a high-accuracy pendulum reference, and grid time, tracked by a synchronous motor running on the system’s own power. When the grid ran below 60 cycles, grid time fell behind real time and the error accumulated on the dial. When it ran above, the error accumulated in the other direction. The switchboard operator read the error directly and issued correction commands to the frequency control unit operator to bring the accumulated error back toward zero. It was the same measurement concept I’m using today in my Arduino frequency meter project, where a TCXO crystal reference replaces the pendulum, and an optocoupler replaces the synchronous motor. The physics hasn’t changed. The tools got smaller.

I hope you enjoyed this second walk down the early history of power system operations. These were the unspoken heroes who made the interconnected grid possible — not with computers or algorithms, but with concentration, experience, and a box of scrap hardware. Drop me a comment with your own story.

References

Downing, P.M. “The Load Dispatching System of the Pacific Gas and Electric Company.” Pacific Gas and Electric Magazine, Vol. I, No. 2, July 1909, pp. 48–50.

Varney, F.H. “The Steam Turbine Installation at Oakland.” Pacific Gas and Electric Magazine, Vol. I, No. 2, July 1909, pp. 35–38.

Downing, Frances Stevenson. “Men Must Work.” Pacific Gas and Electric Magazine, Vol. I, No. 2, July 1909, p. 53.

Fogalsang, Thomas. “The Development of Automatic Regulation for Turbo Generators.” Pacific Gas and Electric Magazine, Vol. IV, No. 12, May 1913, pp. 426–427.

Martin Station photograph and history. Calisphere, Brisbane Library Historical Photograph Collection, University of California.

Cohn, Nathan. “Tie-Line Bias Control.” Leeds & Northrup Company, Philadelphia, PA, c. 1930s. Cohn’s work established Area Control Error (ACE) as the foundation for modern Automatic Generation Control.

Warren Telechron Company, Ashland, Massachusetts. Master Clock frequency monitoring instrument, c. 1916.

Pacific Gas and Electric Magazine, Vols. I–IV (1909–1913). Digitized by Google, original from University of California. Available via HathiTrust Digital Library.

I’m taking a short break from my Colgate-Folsom interconnection series to launch a document this weekend. I’ll return to the series next week. This one has been a long time coming.

I started my career as a graduate of the Idaho State University Vo-Tech Diesel Electric Program. Straight out of school, I was a field service technician for standby generators. Like most first jobs, those early roots took hold deep and never let go, even as my work moved on to the transmission grid and power plant operations.

In the 1990s, while working as an InterNational Electrical Testing Association (NETA) certified test technician, I wrote a piece for NETA World magazine. That piece inspired a follow-up article in the next edition about why reactive load bank testing mattered. Unfortunately, I didn’t save that issue, and those years didn’t make it into the NETA World archives—so it’s lost to history. But, I never forgot the lesson of that article.

In the years since, I’ve watched for some guide, some reference, some direction on why reactive load bank testing is important and how to do it properly. It never came. NFPA 110—the standard governing standby generators in life-critical applications like hospitals—mentions reactive loading only in passing. The testing requirements are expressed entirely in kilowatts. A generator can pass every code-required annual test with a purely resistive load bank and receive a compliance certificate while its voltage regulator has never been challenged, its exciter thermal capacity is unknown, and its ability to hold voltage under the actual motor and transformer loads it will face in an emergency has never been verified.

So, I decided it was time to write the document that should have existed thirty years ago.

Attached below is my training document and field guide for performing reactive load bank testing on standby generators. It represents several months of work. It is written for three audiences: generator technicians who need to understand the procedure and the math, engineers who need to know what to specify, and facility managers who need to understand what they’re buying and why it matters. The field reference cards in Appendix A are designed to be printed and laminated for use on the job. The specification language in Appendix C can be copied directly into bid documents.

I’m providing this as a free download. Use it, print it, share it with your colleagues. I ask only that you not modify the document itself—if you find an error or have a suggested improvement, let me know in the comments or via private message, and I’ll incorporate it into the next revision. This Substack post will serve as the permanent repository; updated versions will always be available here.

I want to acknowledge the technical review contributions of Chris Atiyeh of Tower Generators in Canton, Connecticut, and Evan Wiese of Quinn Power Systems in Riverside, California. The document is better because of their input, and I’m grateful for the time they invested in it.

Put it to good use. It was written to be a working document, not a shelf decoration.

Brad Panike

NERC Certified Senior System Operator (Retired)

Kilovar Substack — kilovar1959.substack.com

Reactive Loadbank Training V1

1.04MB ∙ PDF file

Download

Download

Enjoying a weekend with my sweetheart as we celebrate another trip around the sun together, be back next week.

Hello readers, a change of direction. We’re going to talk history in this series—jump in the way-back machine and take a look at where this wonderful machine that drives our modern world, the electric power interconnection, began.

Author’s note: always take claims of being “first” with a grain of salt, because someone will always find a way to argue your facts. What I include here are the earliest documented cases I’ve found. If you have some local history that belongs in this story, please add it to the comments.

Before Interconnections

The War of the Currents

Anyone familiar with the early history of electric power knows about the battle of the currents between George Westinghouse and Thomas Edison. Edison staged a massive campaign against the adoption of alternating current, ultimately at the sacrifice of his own personal reputation. Even while that war raged, Edison General Electric was quietly making the change behind Edison’s back. By that point, Edison was only a token shareholder. That’s another story, and there are many books and even a few movies on the subject.

One thing should be understood from the outset: even in the early DC days, generators were run in parallel. Parallel equipment operation was baked into the cake from the very foundation of the industry. What didn’t exist yet was interconnection—tying together separate systems, owned by separate entities, serving separate loads, into a coordinated whole. That’s a fundamentally different problem.

The Firsts

The Niagara Myth

George Westinghouse was a master promoter and played the press like an instrument. He made bold statements about his Niagara project to make it commercially viable, and many of those claims are still repeated today as fact. In that same cloud of promotion, Nikola Tesla has been associated with many inventions that were not his. Let’s set a few things straight.

Tesla’s biggest genuine success was inventing a self-starting AC motor. Up to that point, DC had a decisive advantage: DC motors started on their own, while AC motors needed assistance to begin rotation. Tesla’s induction motor solved that problem. It was huge.

But Tesla’s system was two-phase, not three—two completely separate windings ninety degrees out of phase, requiring four wires. The original generators installed at Niagara were two-phase machines with the stationary stator load windings in the center and the field windings rotating outside the stator, producing 25 Hz output. Tesla did not invent a workable transformer. And Niagara served local load only when it first came online.

The famous Niagara-to-Buffalo transmission line had to wait for the development of the Scott-T transformer connection that converts two-phase to three-phase. Niagara came online serving local load on August 26, 1895. Transmission to Buffalo didn’t begin until November 16, 1896. The plant started with 11 MW of capacity across three units.

Before Niagara

Several milestones were reached years before Niagara’s claimed firsts. This is not an exhaustive list, but it gives a sense of how quickly the technology was moving in the early 1890s—and how much of that movement happened in the American West.

Willamette Falls Station (1890) — Six 80 kW, 4 kV, 125 Hz units sending power over a single-phase line to Portland. Details are sketchy, but the installation was operating before Niagara broke ground.

Lauffen to Frankfurt (1891) — Built for the 1891 International Electro-Technical Exhibition in Frankfurt, Germany, this installation moved 180 kW at 40 Hz over a 109-mile, 20 kV transmission line from a hydro plant at Lauffen to the exhibition. It was a demonstration, not a permanent commercial installation, but it proved the concept of long-distance AC transmission to a European audience.

Ames Hydroelectric Plant (1891) — Near Telluride, Colorado. A 75 kW, 133 Hz plant with 3 kV single-phase transmission to a stamp mill. Small, remote, purpose-built for mining—but it worked.

The Bodie Single-Phase Transmission (1893) — In Bodie, California, the Standard Consolidated Mining Company completed a thirteen-mile, 3 kV single-phase AC line between a 125 Hz hydro plant and a single-phase motor. The motor drove a shaft-and-belt power distribution system typical of the time.

Redlands Power Plant (1893) — Two 250 kW, 50 Hz units in parallel at Mill Creek Hydro, feeding a seven-mile, 2,400-volt three-phase line to Redlands, California. Union Ice was the primary customer.

Folsom Powerhouse (1895) — This one matters for the rest of the story. Four GE 750 kW rotating-armature, stationary-field generators—likely a converted DC design—producing 800-volt, 60 Hz output. Three megawatts in total. Connected via dry-type air-blast transformers to an 11 kV transmission line running 22 miles to Station A in Sacramento. Sacramento became one of the first “electric cities”—streetcars, street lighting, commercial and residential power all served from a single hydro plant. Online July 13, 1895, six weeks before Niagara. (Author’s disclosure: I served as a volunteer docent at the Historic Folsom Powerhouse from 1990 to 1996.)

The First Interconnection

How It Happened

All of these early installations shared a common characteristic: they were isolated systems. One plant, one load, one owner. They proved that hydroelectric generation worked, that AC transmission worked, that you could move power over meaningful distances. But they were islands.

The step from isolated plant to interconnected system didn’t come from grand engineering vision. It came from desperation.

Folsom’s Problem

The Sacramento Electric, Gas and Railway Company—the Livermore family’s consolidation from 1896—was having a difficult time meeting Sacramento’s growing load. The American River was unreliable. Drought reduced Folsom’s output to a fraction of its capacity, and power transmission to Sacramento was being interrupted. The company was scrambling.

They took several emergency measures. They built a steam plant in Sacramento as a backup. They constructed a smaller tailrace powerhouse below the main plant to squeeze another 750 kW out of the same water by exploiting the 25-foot elevation difference at the mouth of the tailrace. They even moved Unit 4 down to the tailrace powerhouse as a stopgap. None of them were enough.

Sacramento’s load kept growing. The American River kept disappointing. The Livermore’s needed more power, and they needed it from a source that wouldn’t dry up when their river did.

Colgate—Built for Sacramento

Meanwhile, sixty miles to the northeast, two men were moving fast.

John Martin and Eugene de Sabla had already built the Rome power station on the South Yuba River near Nevada City in 1896, serving the mines of Grass Valley and Nevada City. In the spring of 1898, they built a 1,220-horsepower plant on the middle fork of the Yuba River to serve Marysville—and completed the entire installation in four months and five days.

They were fast. And they had ambition.

Sacramento Electric, Gas and Railway Company’s woes led to negotiations, and a contract was struck. The Yuba Electric Power Company was formed on January 30, 1899, to build the Colgate Powerhouse near Dobbins, California. The plant was named for its financier, Romulus Riggs Colgate, a New York soap fortune heir. Stanley Electric Manufacturing Company of Pittsfield, Massachusetts, supplied the generators.

In what would be an impressive feat even today, Martin and de Sabla built a three-unit, 2,700 kW power plant and a 61-mile transmission line to Station A in Sacramento—the same Station A that Folsom already fed. Construction photography began in July 1899. The plant came online September 4, 1899, just four days past the agreed-upon deadline.

In the fall of 1899, three generating sources were operating in parallel: Colgate hydro, Folsom hydro, and the Livermore’s’ Sacramento steam plant. Colgate handled speed and voltage regulation for the entire system. A 1912 article in the Journal of Electricity, Power and Gas documents the arrangement: “In the fall of 1899 energy was supplied to the system of the Sacramento company, the steam plant of the latter, the hydroelectric station at Folsom and the Colgate station being operated in parallel.”

What This Meant

This was not two generators running in parallel inside the same plant. This was two separate companies—the Livermore’s’ Sacramento Electric, Gas and Railway Company and Martin and de Sabla’s Yuba Electric Power Company—with separate generating stations fed by separate rivers, tied together at a common delivery point, operating in coordinated parallel. Colgate was designated as the frequency leader.

That required a host of things that wouldn’t become formalized for decades: operating agreements, frequency control responsibility, shared load dispatch, coordinated planning. A system operator at Colgate was holding frequency for generators he couldn’t see, owned by a company he didn’t work for, fed by a river he couldn’t control. Every problem that NERC would eventually write standards about was already present in embryonic form on the Yuba River in 1899.

The tie line operated first at 22 kV, then was raised to 40 kV, and finally to 60 kV as the system grew.

The Race That Made It Possible

Here’s where the story gets interesting.

Colgate was built to serve Sacramento. That was the contract, and that was the plan. But Colgate wasn’t the only hydroelectric project under construction in the Sierra foothills.

Prince André Poniatowski—a genuine Polish-French prince, descended from royalty—had first visited California in 1892 and conceived of harnessing the Mokelumne River for hydroelectric power after seeing Niagara Falls. He and his partners, including banker William Henry Crocker, incorporated the Standard Electric Company of California on November 27, 1897. Their precursor plant, the Blue Lakes Powerhouse, came online in October 1897, serving mines in Amador and Calaveras Counties.

Poniatowski’s big project was the Electra Powerhouse on the Mokelumne River. It would be a 10,000 kW plant with transmission to the San Francisco Bay Area—a distance that no one had attempted commercially. Stanley Electric proposed running the line at 50,000 volts over more than 100 miles. All three major electrical manufacturers—General Electric, Westinghouse, and Stanley—had told Poniatowski that no insulator existed that could handle 40,000 volts. He pressed ahead anyway.

But Electra was slow. Construction planning didn’t begin until 1899. The bunkhouse didn’t move to White’s Bar until November 1899. Water didn’t reach the main canal until February 1902. Electra’s five generators didn’t come online until May 6, 1902, with power reaching San Francisco in November.

Poniatowski had the earlier vision by seven years. He lost the race by three.

After Colgate was running and the Sacramento contract was fulfilled, the Oakland Transit Company came looking for cheaper power to operate its 126 miles of street railway in Oakland, Berkeley, and Alameda. Martin and de Sabla saw the opportunity. They organized the Bay Counties Power Company on June 4, 1900, consolidating their holdings.

Colgate power was first delivered in Oakland on April 27, 1901. The initial voltage was 40,000 volts, raised in 1903 to the 60,000 volts for which the line had been designed. Through a tie line to the substation of Poniatowski’s Standard Electric Company—which had not yet completed its Electra plant—power was supplied to Standard for delivery under contract to the Oakland Gas, Light and Heat Company. Standard also took Bay Counties power for delivery to San Jose, a transmission distance of 184 miles from the generators at Colgate Powerhouse.

The competitor whose plant wasn’t finished was buying power from the company that beat him to the Bay Area—and reselling it under his own contracts.

Within a few more years, the whole web—Bay Counties, Standard Electric, Valley Counties, and the rest—would be consolidated into a single entity: Pacific Gas and Electric Company, incorporated in October 1905.

The Grid Wasn’t Planned

The lesson of Colgate-Folsom is one that every system operator recognizes: the grid wasn’t designed from the top down. It grew from the bottom up, driven by necessity, competition, and opportunity.

Drought on the American River forced the Livermores to find supplemental power. A contract to serve Sacramento put Martin and de Sabla’s transmission line into the same station that Folsom already fed. The Oakland Transit Company’s appetite for cheap power extended the network to the Bay. The competitive pressure of Poniatowski’s Electra project created the market dynamics that pulled it all together. Each step was a response to an immediate need, not a master plan.

But the result was a master plan’s equal: an interconnected system spanning hundreds of miles, with multiple generating sources, multiple load centers, coordinated frequency control, and the organizational infrastructure to operate it. The birth of the grid in miniature.

Somebody at Colgate took responsibility for holding frequency for a system that included another company’s generators. From an operator’s perspective, that’s Day One.

Coming Up

In this piece we talked about the history—how the first interconnection happened and why. Next, we’ll look at what all these changes meant for the technology that quickly developed to make it work: synchronizing equipment, protective relaying, frequency regulation, and the operating practices that had to be invented from scratch because no one had ever run a grid before.

You have a story about your utility’s early history? Please share it. I look forward to your comments.

Sources

Charles M. Coleman, P.G. and E. of California: The Centennial Story of Pacific Gas and Electric Company, 1852–1952 (McGraw-Hill, 1952).

Archie Rice, “The Great Power Plant at Electra,” PG&E Magazine, Vol. II, No. 4, September 1910.

“World’s Largest Transmission System: Pacific Gas & Electric Company,” Journal of Electricity, Power and Gas, June 1, 1912.

California State Parks, Folsom Powerhouse State Historic Park historical documentation

This has been a fun series to write, and I hope you’ve enjoyed reading this as much as I have enjoyed writing it. Researching these pieces I always learn something. While I have a basic understanding of all these rules and policies, it’s when you have to dig in that the real picture comes together. Thanks for sharing the journey.

So, wrapping this up, let’s go full circle. We started out talking about the hidden 35 GW mentioned by Secretary Wright and included in a couple of DOE 202(c) orders. We talked about the difficulties of bringing that capacity to bear, then ventured down the EPA regulation route to discuss what a tangled web that is to navigate if you want to run standby generators beyond a few hours for system support. Which brings us here. Let’s try to round up all the divergent pieces of this conversation.

First, we established that Behind The Meter (BTM) generation is not standby generation and has to play by a completely different set of rules. For clarification, let’s talk about that BTM generation — we’ve largely ignored it up to this point.

BTM Generation: A Different Animal

Robert Bryce touched on it in his piece “Otto Goes Behind The Meter.” It is true that piston engine units seem to be dominating much of the behind-the-meter market, but they don’t own it. If you look at the equipment mix of projects like Socrates North and South in New Albany, Ohio, you’ll find a mix of gas turbines and otto cycle units. The Colossus supercomputer facility in Memphis is powered by gas turbines of very questionable legality. The reasons to choose between the two go well beyond just lead time, as discussed by Robert.

BTM generation is prime power. It runs continuously, sometimes as the sole source of electricity for a facility. That means it must comply with prime power emission standards — the full weight of EPA regulation, not the generous emergency standby exemptions we discussed in Parts 2 and 3. For diesel, that’s Tier 4 Final with all the aftertreatment burden. For natural gas, it’s the non-emergency side of Subpart JJJJ — still far more manageable, but a real regulatory obligation, nonetheless. This distinction alone explains much of the fuel choice migration we’re seeing in BTM data center builds.

Gas Turbines: Power Density at a Price

Gas turbines offer significant energy in a small footprint. That’s what makes them attractive to both utilities and BTM generators. Where otto cycle engines run into their ceiling, is where gas turbines really start to find their footing. They tend to be very reliable, able to run long periods without stopping, with relatively little routine maintenance. The power density is hard to argue with — a single aeroderivative turbine can deliver 30 MW or more from a package you could park in a large garage bay.

But they have downsides.

Using a gas turbine is often an “all your eggs in one basket” approach precisely because they generate so much power per unit. They consume more fuel per MWh than an otto cycle engine, though they also produce more usable exhaust heat — and in a data center context, that heat is far from a waste product. Steam absorption chillers can convert exhaust heat into chilled water for server cooling, which is one of the largest energy draws in any data center. A CHP system paired with absorption chillers can dramatically reduce the electricity otherwise consumed by mechanical compressor chillers. That waste heat recovery advantage is one of the strongest arguments for gas turbines in BTM data center applications.

The real knock comes when it’s time for major maintenance. Gas turbine hot sections require extremely specialized parts and highly trained technicians to service. A hot section overhaul is not something you do in the field with a wrench and a torque spec — it is usually returned to the factory. So instead of days, you’re looking at weeks of downtime.

Due to that extended repair cycle, turbine owners usually belong to user groups — essentially cooperatives that keep one to several spare hot sections in reserve. When your machine comes due for overhaul, the pool hot section goes in, your section gets shipped to the factory for repair, and once rebuilt it joins the pool for the next member’s turn. It works, but it gets extremely expensive. I saw this firsthand at Redding Electric Utility. We had two Alstom GTX100 turbines running combined cycle, feeding an older GE steam turbine that was originally driven by an onsite wood-fired boiler. Hot section maintenance on those machines was coordinated through exactly this kind of user group arrangement — you’re essentially paying into an insurance pool for parts that cost more than most houses.

The gas turbine manufacturer field is extremely small. On the large frame and aeroderivative side, you’re really looking at three companies: GE Vernova, Siemens Energy, and Mitsubishi Power. That’s it. Siemens now owns what was the Alstom turbine line — the same Finspång, Sweden operation that built the GTX100 machines we operated in Redding. Caterpillar owns Solar Turbines, which fills in the gap on the small industrial end of the market with their Saturn, Centaur, Taurus, and Mercury product lines. In East Asia, Doosan is a player out of Korea, and China has some domestic manufacturing capability, but Mitsubishi is the only major worldwide player out of that region in terms of export market presence.

Three companies for the big iron. That’s a thin bench for a technology the world is increasingly counting on.

Otto Cycle Engines: The Workhorse

Otto cycle reciprocating engines have been around for over a hundred and fifty years now. They are a well-known device, relatively easy to repair compared to a turbine, and they offer better single-unit thermal efficiency. Waste heat recovery is possible — recip engines can also drive absorption chillers for facility cooling — but collecting that heat is more complex than with a turbine. The thermal energy is split between exhaust, jacket water, and lube oil circuits rather than concentrated in a single high-temperature exhaust stream. Single-stage absorption chillers driven by hot water are the typical match for recip CHP installations, while turbines can support higher-efficiency two-stage machines with their hotter, more concentrated exhaust.

They are easy to start and stop. They don’t suffer from the same thermal stress damage that plagues gas turbines on short run cycles — a critical advantage for anything that needs to respond to grid signals or load-follow. Their ramp time from start to full load is considerably better than a gas turbine.

There is no need to join user groups and maintain multi-million-dollar spare hot sections. Parts are straightforward. A full overhaul kit — liners, rebuilt heads, bearings, gaskets — can easily be kept in an onsite warehouse. Because of their individual size, recip engines tend to operate in groups, so the loss of a single unit has significantly less impact on overall plant output than the loss of a single turbine. If you have twenty engines and lose one, you’ve lost five percent of your capacity. If you have two turbines and lose one, you’ve lost half.

But they do require quite a bit more hands-on maintenance than a gas turbine during normal operation. The overhaul life cycle is usually shorter, though that is steadily improving with better metallurgy and oil analysis programs. That typically means a larger maintenance staff onsite. And they occupy a considerably larger footprint for the same total output as a gas turbine installation.

Both technologies have legitimate places in the BTM market. The choice comes down to site constraints, operating philosophy, maintenance capability, and — increasingly — how fast you can actually get the equipment delivered.

Who Actually Makes These Engines?

This is where it gets interesting — and where the public, and frankly most of the policy world, has almost no visibility. When we start talking otto cycle engines, the manufacturer field gets larger than the turbine side, but not as much as you would think. Reaching all the way from the small standby engines in your neighbor’s yard to the top of the otto cycle product line powering data centers, it is actually a pretty small group of manufacturers serving a huge market.

And once you start pulling the covers back, you find out that a lot of the names on the enclosure aren’t the names on the engine block.

The Residential Tier: Everybody’s Selling the Same Engine

At the bottom of the chain — the sub-30 kW home standby market — the consolidation is almost comical. Generac, Champion, Honeywell Home, and several other brands are all selling generators powered by the same 999cc Loncin engine out of China. Different paint, different enclosure, same block. Briggs & Stratton’s Fortress units are one of the few outliers with different iron underneath

.

Generac claims to manufacture its own engines for natural gas units up to about 1 MW, and they do have domestic engine assembly in Whitewater and Oshkosh, Wisconsin — but how much of that is truly manufactured versus assembled from imported components is a question worth asking. What we know for certain is that the residential volume product is Chinese-sourced.

This matters because it sets the pattern that repeats all the way up the chain: the block comes from one place, the fuel management and controls come from another, and the badge on the enclosure is a marketing decision.

Light Commercial: The Sourcing Gets Creative (30–500 kW)

Move up into the light commercial and small industrial range and the sourcing picture gets genuinely complex.

Generac runs a tiered sourcing strategy. Their own 4.5L and 9.0L engines cover the small end. At 14.2 liters they’re using Shanghai Diesel (SDEC) — a Chinese engine. From 350 to 500 kW they source from MAN, and above 600 kW the iron comes from Mitsubishi. That’s four different engine suppliers across one product line.

Caterpillar — a name most people assume makes everything they sell — has been known to source engines from Generac and Origin Engines in the 200 kW class, painting over the original manufacturer’s stickers. Cat absolutely makes their own engines at the larger end, and they own EMD (Electro-Motive Diesel, via Progress Rail) for the truly massive 710 series prime movers. But in the middle of the lineup, they’re buying iron like everybody else.

Rehlko (formerly Kohler Power) buys the 14L, 18L, and 22L Doosan V Series engines through PSI (Power Solutions International), fitted with EControls fuel systems for natural gas/LP bifuel capability and Holset turbochargers exclusively. Good engines, well-integrated — but not Rehlko’s own iron.

Cummins stands alone in this range as the only manufacturer with true domestic vertical integration above 150 kW on natural gas. Above that threshold, a Cummins genset has a Cummins engine, period. They also supply engines to other packagers, including Kohler/Rehlko for some gas and diesel applications. Below 150 kW on the residential side, even Cummins sources a Mitsubishi 2.4L for their smaller platforms — and historically used Ford 4.2L V-6 and 6.8L V-10 long blocks, buying them without warranty and doing all the intake and fuel system work in-house.

How Is This Legal? The EPA Manufacturer of Record

You might be wondering how Generac can put a Shanghai Diesel engine in a genset and call it a Generac. The answer is the EPA Manufacturer of Record framework. Whoever registers the engine with EPA for emissions certification is the legal manufacturer, regardless of who cast the block. If you buy an incomplete engine, add your own ignition and fuel management systems, calibrate it, and submit it for emissions certification under your name — congratulations, it’s your engine as far as EPA is concerned. You own the warranty, the compliance obligation, and the liability.

This is not a loophole. It’s how the industry has operated for decades. Cummins used to buy Ford V-10 long blocks without warranty, do all the intake and fuel system engineering themselves, certify and warrant the finished product as a Cummins engine. Same block, completely different machine by the time it went into service. The regulatory framework accommodates this because what matters for emissions compliance is the finished, certified configuration — not who poured the metal.

Understanding this is essential to understanding why the genset “competitive landscape” is much thinner than the brand names suggest.

But there’s a nuance here that matters to anyone who actually has to maintain these machines. Just because two gensets share the same base engine block doesn’t mean they’re the same engine. The buyer’s specification determines what goes inside. One packager might spec forged crankshafts, steel cylinder liners, and replaceable valve seat inserts. Another might take the same displacement block with cast cranks, iron liners, and integrated seats. Same foundry, same casting, remarkably different machine in terms of durability, overhaul intervals, and long-term parts cost. The technicians who open these engines up know exactly what they’re looking at — and they know which spec sheet to trust and which one to question. That kind of knowledge doesn’t show up in a sales brochure.

Large Industrial: Where the Field Narrows Fast (500 kW–4 MW)

At the large industrial and mission-critical scale, the illusion of a broad competitive market really starts to fall apart.

Rehlko actually runs a double diesel line at the large industrial scale. Their KD Series — the flagship platform — is Liebherr iron, co-developed between Liebherr’s operations in Colmar, France and Bulle, Switzerland. That’s the product line for dedicated, loyal Rehlko customers. But for competitive bid work, Rehlko also offers Mitsubishi-powered units — the S12R and S16R platforms that are technically competitive and well-proven. Two different engine sources, two different go-to-market strategies under one brand. Below the KD range on the diesel side, Rehlko uses Volvo Penta engines for their mid-range products. The industry has known about the Liebherr relationship for years, but it was reinforced when insiders noted that Rehlko didn’t even have a test cell large enough for their own 4 MW KD Series unit. You don’t lack a test cell for an engine you designed.

Caterpillar makes their own diesel engines at this scale — the 3500 series and above — and owns the EMD 710 platform through Progress Rail for the truly massive applications. On the gas side, Cat offers well-proven platforms through the mid-megawatt range.

Cummins fields the QSK50 and QSK78 Centum series at the top end — but unlike the gas side, Cummins makes their own diesel iron across essentially the entire range. The same company that puts the 6.7L turbodiesel in your Ram pickup builds the multi-megawatt prime movers. On natural gas above 150 kW, Cummins is vertically integrated. On diesel, they’ve been vertically integrated from the bottom to the top for decades.

That’s essentially it for companies making their own large engine iron for the North American market. Only Cat and Cummins manufacture their own diesel engines at this scale with domestic production. Everybody else is packaging someone else’s block.

Mitsubishi makes technically competitive large engines — the MGS series, S12R, S16R — but has been essentially absent from the US genset market despite strong positions in Asia.

MTU (Rolls-Royce Power Systems) has been developing a 1600 Series gaseous engine platform for the 300–500 kW range, but the rollout has been troubled — reportedly only a handful of units shipped over more than a year of availability. Not yet a factor.

The lead time picture at this scale tells its own story. Generac quotes 50–60 weeks for large industrial units. Caterpillar and Cummins are out to 107 weeks — over two years. When a hyperscale operator needs generation capacity to meet a construction timeline, that delivery gap is a strategic factor, not just a procurement nuisance.

BTM Data Center Scale: Follow the Permits

At the top of the otto cycle market — the multi-megawatt behind-the-meter installations powering data centers — the manufacturer field is dominated by a handful of European engine builders.

INNIO Jenbacher has carved out a strong position here. The PowerConneX facilities in New Albany, Ohio — which I’ve photographed and documented firsthand — are all-Jenbacher installations. These are Type 6 and Type 9 engines running natural gas as baseload BTM generation, configured with closed-loop fin fan cooling systems rather than evaporative cooling towers. That cooling choice alone tells you something about the water consumption politics these operators are navigating.

Wärtsilä competes at the same scale with their 34SG platform, capable of reaching into the high teens to low twenties of megawatts per unit. Strong in the utility peaker market, increasingly showing up in BTM data center applications.

MAN Energy Solutions, part of the Volkswagen Group, builds credible large-bore engines but has limited North American presence in the data center backup and BTM space.

The Stargate facility in Abilene, Texas provides a revealing snapshot of how this actually plays out at hyperscale. The construction permits show 28 Baudouin engines and 34 Caterpillar gensets on the same site. Baudouin is owned by Weichai Power — a Chinese state-controlled enterprise. That means a flagship American AI infrastructure project has Chinese-manufactured engines at its core. The story was broken by Hunterbrook Media in February 2026, and the implications for tariff vulnerability and supply chain security are significant and ongoing.

This Chinese manufacturing exposure isn’t limited to Baudouin. Generac’s residential volume runs on Loncin engines from China. The Shanghai Diesel engines in Generac’s mid-range commercial products originate from the same industrial ecosystem. Even Mitsubishi engines serving the North American market are produced through a Shanghai Mitsubishi Engine joint venture. The supply chain threads lead back to China far more often than the nameplates suggest — and in a trade environment where tariffs on Chinese goods are escalating, that embedded exposure creates real commercial risk that most buyers haven’t priced in.

The Strategic Plays Worth Watching

PowerSecure (a Southern Company subsidiary) uses Volvo Penta engines in their PowerBlock containerized generation systems and has a six-year alliance with Edged (formerly Aligned Energy). The Edged New Albany data center campus uses PowerSecure Tier 4 Final PowerBlocks — a vertically integrated utility-subsidiary model that’s quite different from the independent packager approach.

Rehlko’s European orientation — Liebherr diesel iron, the Clarke Energy distribution relationship, and Clarke’s deep integration with INNIO Jenbacher — positions them for a potential BTM play through the Jenbacher platform rather than trying to compete with Cat and Cummins on domestic diesel manufacturing they don’t have. If Rehlko packaged a Jenbacher-based BTM gas generation solution alongside their KD Series diesel emergency backup through Clarke Energy’s installation and service network, they’d be offering a complete power island to a data center developer through one commercial relationship.

Blue Star Power Solutions operates as a pure integrator — no engine manufacturing, but strong packaging and project delivery capability using engines sourced from multiple OEMs.

The View From Here

So what does all this add up to?

The 35 GW of backup generation Secretary Wright identified is real. It’s sitting behind fences and in basements and on rooftops across the country. But mobilizing it for grid support — even in an emergency — runs headlong into the regulatory thicket we walked through in Parts 2 and 3. Run your standby diesel beyond its permitted hours and you’re staring at Tier 4 Final aftertreatment obligations, test hour budget violations, and potential enforcement actions. Run your natural gas standby unit as prime power and you’ve jumped from the generous emergency side of Subpart JJJJ to the non-emergency standards. The physics of the engines don’t change. The regulatory environment changes completely.

Meanwhile, the BTM market is booming precisely because the grid can’t keep up with load growth. Data center operators aren’t building on-site generation because they want to — they’re building it because they can’t get enough grid power fast enough. And the manufacturer base serving that market is far thinner than most people realize. A handful of foundries in Finland, Austria, Germany, France, China, and a few American cities are casting the blocks that keep the lights on for the infrastructure the modern economy runs on.

Three companies make large gas turbines. Two companies make their own large diesel engines domestically. One company has full vertical integration on natural gas engines above 150 kW. The rest is packaging, sourcing, and — with all due respect to some very competent integrators — badge engineering.

The people who actually understand this — the operators, the maintenance technicians, the field service engineers who open these machines up and know whose casting marks are on the inside of the crankcase — are retiring. They’re taking institutional knowledge with them that doesn’t exist in any database or spec sheet. That’s been the thread running through everything I write here at Kilovar: the gap between what the nameplates say and what the people who turn the wrenches know is getting wider, not narrower.

If this series has done its job, you now understand the regulatory architecture, the operational trade-offs, and the industrial reality behind that number — 35 gigawatts — that sounds so simple when a Secretary of Energy says it out loud.

Thanks for reading. As always, I welcome your comments, corrections, and war stories. This is a conversation, not a lecture

If you’ve been following this series, you’ve now spent two articles inside the regulatory machinery that governs diesel standby generators. Part 1 laid out the tier structure — Tier 1 through Tier 4 Final — and the patchwork of federal, state, and local air district rules that a facility owner must navigate before a diesel genset ever turns a crankshaft. Part 2 walked you through what Tier 4 Final actually means on the ground: diesel particulate filters (DPF) that need regeneration cycles, diesel exhaust fluid (DEF) that freezes at 12°F, a 100-hour annual test budget that barely covers your National Fire Protection Association (NFPA) 110 obligations, and an inducement system that will derate your engine if it decides something is wrong with the aftertreatment.

If you came away from those two pieces thinking “there has to be a better way,” you’re not wrong. There is. And it’s been sitting there the whole time.

The Other Subpart

The diesel generator world lives under 40 Code of Federal Regulations (CFR) Part 60, Subpart IIII. That’s the rule package we’ve been dissecting. But there is another subpart that governs a completely different class of engines, and it operates under a fundamentally different regulatory philosophy.

Subpart JJJJ — Standards of Performance for Stationary Spark Ignition Internal Combustion Engines — covers natural gas and propane engines. If you’re an operator or a facility engineer who has spent the last two articles wincing at DPF regen cycles and DEF logistics, I need you to pay attention to what Subpart JJJJ does not require.

No diesel particulate filter. No selective catalytic reduction (SCR). No diesel exhaust fluid. No regeneration cycles. No inducement system. No derate risk.

For emergency standby spark ignition (SI) engines, the Environmental Protection Agency (EPA) essentially said: meet your emission limits at the engine, and we’re done here. The engine manufacturer handles compliance at the factory through combustion tuning — lean-burn calibration, ignition timing, air-fuel ratio management. The unit arrives at your site certified. You install it, commission it, connect it to your gas supply, and it’s ready to protect your facility.

There is no aftertreatment system to maintain, no consumable fluid to stock, no regeneration cycle to schedule, and no software system monitoring a catalytic converter that can lock your engine into a reduced power mode during the worst possible moment.

Read that list again. Then go back and read Part 2.

Why the Difference?

The answer is combustion chemistry, and it’s not complicated.

Diesel is a compression ignition fuel. You spray fuel into a cylinder full of hot compressed air and it ignites on contact. That process is thermodynamically efficient, but it produces two things that EPA cares deeply about: particulate matter (PM) — soot — and oxides of nitrogen (NOx). The entire Tier 4 aftertreatment architecture exists to capture those two pollutants after they come out of the cylinder, because the diesel combustion process inherently creates them.

Natural gas is a spark ignition fuel. It premixes with air before entering the cylinder, and a spark plug initiates combustion — the same basic process as your car engine. That premixed, controlled burn produces dramatically less particulate matter. There’s essentially no soot. The PM numbers are so low that EPA doesn’t require a particulate filter. And NOx can be managed through combustion tuning at the factory — lean-burn calibration that keeps flame temperatures below the threshold where nitrogen and oxygen combine into NOx.

The result: natural gas engines meet their emission standards at the tailpipe with nothing but good combustion engineering. No bolted-on aftertreatment. No consumables. No maintenance burden beyond normal engine service.

This isn’t a loophole. This is chemistry.

One clarification before we move on, because I’ll get emails if I don’t. Not all natural gas engines are created equal under Subpart JJJJ. The regulation distinguishes between lean-burn and rich-burn engines, and the distinction matters.

Lean-burn engines run with excess air in the combustion mixture — more air than the fuel needs for complete combustion. That excess air acts as a heat sink, keeping peak flame temperatures low, which is what holds NOx formation down. The engine meets its emission limits through combustion management alone; no aftertreatment required. This is the approach used in most residential and small commercial standby units, including the Cummins RS17A in my backyard.

Rich-burn engines run at or near stoichiometric — the chemically ideal air-to-fuel ratio for complete combustion. That produces higher exhaust temperatures and higher raw NOx, which means rich-burn engines above certain horsepower thresholds under JJJJ may require a three-way catalyst to meet their emission standards. That is aftertreatment — it’s simpler than the DPF/SCR/DEF stack on a Tier 4 diesel, but it’s hardware on the exhaust that needs maintenance and can degrade over time. If you’re specifying a larger commercial natural gas genset, ask whether the engine is lean-burn or rich-burn and what the catalyst requirements are. Don’t assume the aftertreatment-free story applies to everything running on natural gas.

The Standby Exemption — Again

In Part 1 we discussed the emergency standby exemption under §60.4211(f) that gives diesel generators some breathing room on emission standards. Subpart JJJJ has its own version of this for spark ignition engines, and it’s even more favorable.

Emergency stationary SI engines under Subpart JJJJ are subject to less stringent emission limits than their non-emergency counterparts, and critically, there is no aftertreatment expectation at all. The manufacturer certifies the engine to the applicable emission standard; the unit is designated and permitted as emergency standby, and the operational hour limits apply — but the physical hardware burden is the engine itself. That’s it.

The hour limitations are similar to the diesel side: 100 hours per year for maintenance and testing, unlimited hours for actual emergencies, and 50 hours per year for non-emergency situations under certain conditions. The engine has to be operated and maintained according to the manufacturer’s recommendations. But the critical difference is what happens between those hours.

A Tier 4 Final diesel sitting idle still has a DPF that needs periodic regen to burn off accumulated soot. Still has DEF that can freeze, crystallize, or degrade. Still has sensors monitoring the aftertreatment that can trigger an inducement response. The complexity doesn’t sleep.

A natural gas standby unit sitting idle is waiting for a call to action with exactly zero aftertreatment maintenance requirements. It starts. It runs. It carries load. It shuts down. That’s the operational profile, and there is nothing bolted to the exhaust stream that adds complexity, cost, or failure modes.

A Real Unit in a Real Backyard

I have a Cummins RS17A sitting on a concrete pad next to my garage in Licking County, Ohio. It’s a residential standby natural gas generator — a twin-cylinder engine connected to the local gas distribution system, with an automatic transfer switch (ATS) that puts it online when American Electric Power (AEP) loses the feed.

This unit shipped from the factory with no aftertreatment system. No DPF. No SCR. No DEF tank. The exhaust comes out of the engine, through a muffler, and into the atmosphere. It meets its EPA emission certification through combustion engineering alone — the same Subpart JJJJ framework I just described.

The extended warranty document from Cummins tells the story in a way that the regulation itself doesn’t make obvious. Nowhere in that warranty will you find the words “not for prime power use.” What you will find is the designation “Residential Standby” on every warranty tier, and an exclusion for “inappropriate use relative to designated power rating” and “inappropriate use relative to application guidelines.” The standby-only restriction is there, but it’s written in warranty language, not regulatory language.

Here’s what Cummins doesn’t put in writing but has stated verbally: the standby-only designation is directly tied to the EPA air permit classification that allows the engine to exist in its current form. The twin-cylinder engine in my RS17A ships without aftertreatment because it’s classified as emergency standby under Subpart JJJJ. If someone tried to operate it as prime power, they wouldn’t just void the warranty — they’d blow the air permit basis that allowed the engine to be manufactured without aftertreatment in the first place. Two consequences from one act, and the more serious one isn’t the warranty.

This is the regulatory architecture working exactly as designed. The engine is clean enough at the tailpipe that EPA doesn’t require aftertreatment for the limited-hour standby profile. The manufacturer builds the unit to that standard, certifies it, and prices it accordingly. The owner gets a backup power system that starts reliably, runs cleanly, and doesn’t require a second career in aftertreatment system maintenance to keep compliant.

Why Diesel Isn’t Going Anywhere

Before anyone accuses me of writing a natural gas sales brochure, let me be clear: diesel still wins in applications where it should win, and there are real engineering reasons why.

The equipment selection alone tells the story. The diesel standby generator market is enormous. Caterpillar, Cummins, Rehlko, and others offer diesel gensets from small commercial units up through multi-megawatt machines. The range of available sizes, configurations, and power ratings in diesel dwarfs what is currently available in natural gas for standby applications. If you need a 2 MW standby generator, your diesel options fill a catalog. Your natural gas options fill a short list.

Then there’s fuel security. A diesel generator with a full storage tank is fuel-independent. It doesn’t care what’s happening upstream. The pipeline can lose pressure; the gas utility can curtail service; the compressor station can go down — your diesel doesn’t know and doesn’t care. It has fuel on site, in a tank you can see and measure, and that fuel has an energy density that natural gas can’t match by volume. A 1,000-gallon diesel tank is a known quantity of runtime. A natural gas connection is a promise from someone else’s infrastructure.

That matters. During Winter Storm Fern in January 2026, pipeline pressure wasn’t the issue — but it could have been, and every facility manager who lived through that week was thinking about it. During Winter Storm Elliott and Winter Storm Uri it was. Fuel independence is not an abstraction when the grid is down, and the temperature is in single digits.

The NFPA Level 1 Problem

Here’s where natural gas gets complicated for critical facilities, and this is a detail that the “just switch to gas” crowd usually leaves out.

NFPA 110 — the Standard for Emergency and Standby Power Systems — defines Type 10 systems as those that must restore power within 10 seconds of a utility failure. Level 1 systems are the highest reliability classification, covering hospitals, certain data centers, and other facilities where loss of power is a life-safety issue. And Level 1 systems require on-site fuel capacity for 96 hours of operation at the generator’s nameplate rating.

Diesel handles both requirements without breaking a sweat. A diesel engine cranks, fires on compression ignition, and is at rated speed in seconds — 10-second start is standard across the product line. And 96 hours of fuel is a sizing calculation: figure out your fuel consumption rate, multiply by 96, and order the tank.

Natural gas has a harder time on both counts.

The 10-second start requirement is achievable with natural gas engines, but it’s not a given. Natural gas doesn’t compression-ignite — it needs a spark, and the air-fuel mixture needs to be right before the engine will fire cleanly. In cold weather especially, natural gas engines may need enhanced start aids — block heaters, battery warmers, intake air heaters, or pre-lube systems — to reliably hit that 10-second window. Every one of those additions is another system to maintain, another failure mode to manage, and another item on the commissioning checklist. The simplicity advantage I described in the aftertreatment comparison starts to erode when you bolt on the hardware needed to guarantee a cold-weather fast start.

Then there’s the 96-hour fuel question. If your natural gas comes from a utility distribution main, you’re relying on that pipeline staying pressurized for four straight days during the same emergency that knocked out the grid. That’s not fuel storage — that’s fuel faith. Some facilities address this with on-site liquefied natural gas (LNG) or compressed natural gas (CNG) storage, but now you’ve added cryogenic or high-pressure fuel storage infrastructure, and the cost and complexity calculation changes significantly.

This is why you still see diesel as the dominant choice for hospital emergency power, for NFPA 110 Level 1 / Type 10 installations, and for any application where the answer to “what if the gas goes out too?” can’t be a shrug. The regulatory burden of Tier 4 Final is real, but for some facilities, accepting that burden is the price of fuel certainty.

“But What If I Want to Run It as Prime Power?”

I get this question constantly. Someone reads about natural gas generators, sees the regulatory simplicity of the standby classification, and asks: what if I just run it all the time? What does it take to operate a natural gas engine as a prime power source?

The answer is: a completely different regulatory universe.

The moment you move a stationary spark ignition engine from emergency standby to prime power — or even to non-emergency use beyond the allowed hours — you leave the favorable corner of Subpart JJJJ and step into the full force of the regulation. The emission standards tighten significantly. For engines above certain horsepower thresholds, prime power and continuous duty applications require emission levels that may not be achievable through combustion tuning alone.

That means aftertreatment. The same catalytic converters, oxidation catalysts, and in some cases SCR systems that Subpart JJJJ doesn’t require for emergency standby engines may become mandatory for the same engine operating in a prime power role. You also pick up additional monitoring, recordkeeping, and reporting obligations. And depending on your state and local air district, you may need a formal air permit with emission caps, stack testing requirements, and annual compliance demonstrations.

The economics change too. A prime power natural gas engine runs thousands of hours per year, not dozens. The maintenance intervals are compressed. The engine wears at a fundamentally different rate. And the manufacturer’s warranty — if it covers prime power at all — will be structured very differently than the standby warranty we looked at earlier. The 2,000-hour cap in the Cummins residential warranty isn’t arbitrary: it’s calibrated to the standby duty cycle. A prime power application will blow through that ceiling in a fraction of the warranty period.

This is exactly what’s playing out right now in the data center world, where behind-the-meter natural gas generation is being deployed at scale — not as emergency standby, but as primary or supplemental power. Those installations are operating under different permits, different emission standards, different engine configurations, and different maintenance regimes than anything in the residential or commercial standby market. They are not running the same regulatory playbook, even though the fuel is the same.

One more variant I should mention, because a reader brought it up in the Part 2 comments and it’s a fair question: dual-fuel engines. These are machines — typically large-bore, utility-scale units from manufacturers like MAN and Wärtsilä — that run primarily on natural gas but use a small pilot injection of diesel (often around 1%) to initiate combustion, eliminating the spark plug ignition system entirely. The advantages are real: you get the emission profile of a gas engine, the compression-ignition reliability of a diesel, and reduced ignition system maintenance. The MAN 51/60DF and the Wärtsilä 34DF are proven platforms in municipal and utility generation.

But these are 5—20+ megawatt (MW) class machines operating in a completely different regulatory, procurement, and operational world than what we’re discussing in this series. They have their own permit requirements, their own emission standards based on size and duty cycle, and their own maintenance regimes. If you’re running a municipal power plant or an island grid, dual-fuel is a serious conversation. If you’re specifying backup power for a hospital or a commercial building, it’s a different catalog.

The standby exemption is a gift. It lets manufacturers build simpler, cleaner, more reliable engines for a specific duty cycle. But it only works if the engine stays in its lane. The moment the duty cycle changes, the regulatory framework changes with it — and the simplicity advantage disappears.

What This Means for Facility Owners

So where does this leave the facility engineer making a decision?

If you’re in an NFPA 110 Level 1 / Type 10 application — hospitals, certain data centers, life-safety systems — diesel is still likely your primary answer. The equipment selection is broader, the 10-second start is proven, the 96-hour fuel requirement is solved with a tank, and you own your fuel supply. The Tier 4 Final aftertreatment burden is real and it’s expensive, but it’s manageable if you budget for it and staff for it. Go in with your eyes open.

If you’re in a commercial or residential standby application with reliable natural gas service, Subpart JJJJ hands you a genuinely simpler machine. No aftertreatment hardware to maintain. No consumable fluids to stock. No sensor-driven derate risk during an emergency. No regen cycles eating into your test hour budget. And a warranty structure that confirms the manufacturer designed the product for exactly this: sit quietly, start fast, run clean, shut down, repeat.

And if someone asks you about running a natural gas engine as prime power — tell them to read the permit application first.

Chemistry does the work that the aftertreatment does on diesel. And chemistry doesn’t need a software update. But chemistry alone doesn’t answer every question, and the choice between diesel and natural gas is still an engineering decision, not a political one. As always, please drop me a comment with your thoughts or questions. Thank you for reading my column.

Next in the series: In part 4 we will look at the operational realities of reciprocating engine fleets versus gas turbines in standby and behind-the-meter generation, with some stories from the field that illustrate why iron selection matters more than the spec sheet suggests. We will look at who the engine manufacturers are playing in this market and the brand naming deals that often mask who the real manufacturer is.

Hello Readers, thank you for continuing to return to my column. In Part 1, we walked through the regulatory landscape—the EPA Tier system under 40 CFR Part 60 Subpart IIII, the standby exemption under §60.4211(f), and the patchwork of local air quality management districts that overlay the federal rules. Today we’re going to get into the operational weeds. Specifically, what happens when Tier 4 Final emission controls meet the real-world operating profile of a standby generator—and why it’s a nightmare for facility managers, service technicians, and the data center industry.

What Makes Up a Tier 4 Emission Control System

Before we get too deep into the operational issues, let’s look at what makes up a Tier 4 emission control system. It is extremely complex—a chain of five or six interdependent subsystems bolted onto the exhaust path, each one adding weight, cost, failure modes, and maintenance burden. Let’s walk through the components one by one.

1. Exhaust Gas Recirculation (EGR) Valve. The EGR valve recirculates a portion of exhaust gas back into the fresh air intake to slow the combustion process and reduce the formation of nitrogen oxides (NOx). EGR first found wide use of gasoline engines in the 1970s as an early smog-era emission control measure. Those valves have largely disappeared from modern gasoline engines thanks to electronic fuel injection and per-cylinder timing control, but in the diesel world they persist—and they’re known to get sticky, carbon up, and lock in one position or the other. A stuck-open EGR robs power and increases soot production. A stuck-closed EGR lets NOx climb. Either way, you’re creating a downstream problem for the rest of the aftertreatment chain.

2. Diesel Oxidation Catalyst (DOC). The DOC reduces carbon monoxide (CO) and unburned hydrocarbons (HC) by converting them to carbon dioxide (CO₂) and water (H₂O). It uses a precious-metal catalyst core—typically platinum and palladium—to oxidize these pollutants and, critically, to raise exhaust gas temperature. That temperature rise is what makes the next component in the chain work. Think of the DOC as the furnace preheater for the DPF.

3. Diesel Particulate Filter (DPF). The DPF is intended to capture diesel particulates—soot, the black smoke you see trailing behind older diesel equipment. Particulate matter (PM) has been recognized as one of the most significant contributors to respiratory disease in modern air quality science. You’ll hear the term PM 2.5 frequently—that refers to particles of 2.5 microns or smaller, small enough to penetrate deep into the lungs, and a key data point in human health studies. The DPF’s job is to prevent these particles from entering the air we breathe. The problem? The DPF needs sustained high exhaust temperatures to burn off accumulated soot—a process called regeneration—and the engine needs to carry significant load to generate that heat. If the DPF doesn’t get hot enough, it plugs up. This is the central problem for standby generators, and we’ll come back to it in detail.

4. Diesel Exhaust Fluid (DEF) Injection. DEF is a solution of 32.5% high-purity automotive-grade urea and 67.5% deionized water—essentially a stable, safe-to-handle form of ammonia. It’s injected into the hot exhaust stream downstream of the DPF, where it flash-vaporizes and mixes with the exhaust flow. In large utility and industrial power plants, that same NOx-reduction chemistry may use aqueous ammonia or even anhydrous ammonia, the latter of which is extremely dangerous to work with. The automotive-grade urea approach trades some efficiency for safety and simplicity, but it introduces its own set of logistics headaches—which we’ll get to shortly.

5. SCR/ASC Combination Device. The final stage in the exhaust path is a combination unit housing two catalysts:

   1. Selective Catalytic Reduction (SCR): The SCR catalyst reacts with the vaporized urea to convert NOx into harmless nitrogen and water. A NOx sensor in the exhaust stack monitors tailpipe emissions and adjusts the urea injection rate in a closed-loop control system to maintain compliance.

   2. Ammonia Slip Catalyst (ASC): The ASC sits downstream of the SCR to catch any excess ammonia that wasn’t consumed in the SCR reaction and convert it to nitrogen and water before it exits the stack. Without it, you’d smell ammonia at the exhaust—not something your neighbors or your air district want to deal with.

The DEF Quality Sensor: A Timely Change

I should note that the EPA has just made a significant change to this system. On March 27, 2026, EPA Administrator Lee Zeldin announced new guidance making it clear that the urea quality sensor (UQS) is no longer required and that NOx sensors can be used instead to meet regulatory requirements. This matters.

The UQS was a device that measured the concentration, temperature, and level of DEF, typically via ultrasonic technology, to ensure the urea solution was at the required 32.5% concentration for proper SCR function. According to the Truck and Engine Manufacturers Association (EMA), the UQS had one of the highest failure rates of any component in modern SCR systems. When a UQS failed or gave a false reading—and they did so frequently—the engine control unit would initiate an inducement sequence, progressively derating the engine’s power output until the operator addressed the fault. In the trucking world, a single bad sensor reading could strand a rig on the side of the road at 5 mph. For standby generators, a false UQS fault during an actual grid emergency could derate or shut down a unit that was desperately needed.

The EPA’s own analysis of warranty data from 11 of 14 manufacturers confirmed that UQS failures were a leading source of warranty claims and unnecessary inducement events. The new guidance allows manufacturers to monitor DEF quality by inference—if the NOx coming out of the stack is within limits, the DEF is doing its job. Compliance will now be measured by actual emissions performance rather than by the health of a notoriously unreliable sensor. Software updates to implement the change are explicitly authorized under the Clean Air Act and will not be treated as tampering.

The U.S. Small Business Administration estimated this change will save American diesel operators $13.79 billion annually, with $4.4 billion of that going to farmers alone. Those numbers reflect the cumulative cost of false inducement events, unnecessary service calls, towing, lost productivity, and warranty claims driven by a sensor that failed more often than the system it was monitoring.

Now, I want to be very clear about something, because there is a lot of bad information circulating on social media right now. People are conflating four different things: DEF (the fluid), the UQS (the sensor being eliminated), the SCR system (the catalytic reduction hardware, which remains), and the DPF (the particulate filter, which also remains). This ruling removes the requirement for one specific sensor. It does not eliminate DEF. It does not eliminate the DPF. It does not eliminate any emission standard. The SCR system, the DPF, and all Tier 4 aftertreatment hardware remain fully required. If you’ve read otherwise on the internet, you’ve been misinformed.

The Generator Problem: Two Operational Nightmares

For standby generation, the Tier 4 system creates a twofold operational problem that the regulations were never designed to contemplate. Tier 4 was developed for engines that work hard—over-the-road trucks pulling freight, construction equipment digging dirt, agricultural tractors under continuous PTO load. Standby generators are, by definition, equipment that sits idle the vast majority of its life and may run at reduced load when it does operate.

Problem 1: The DPF and the Load Problem

Generators often run at reduced loads—many times well under half load. They are also very often run with zero load for their weekly or monthly test runs, which typically last anywhere from 20 to 30 minutes. This type of low-load, short-duration operation is fundamentally incompatible with a DPF. The filter accumulates soot. Exhaust temperatures never reach the 1,000°F+ threshold needed for passive regeneration, and the DPF slowly chokes.

The Engine Control Unit (ECU) monitors DPF differential pressure continuously. When it senses rising backpressure from soot accumulation, it initiates a regeneration process. There are three escalating stages, and each one is worse than the last:

1. Active Regeneration: Triggered automatically by the ECU when sensors detect moderate soot levels. The engine injects small amounts of raw fuel into the exhaust stream ahead of the DOC to raise temperatures and burn off soot while the generator continues to operate. This burns excess fuel with zero benefit to the facility—you’re literally lighting diesel on fire to clean the filter.

2. Parked/Stationary Regeneration: Required when active regeneration can’t keep up. The generator must be taken offline and run at elevated idle for 30 to 40 minutes while a manually initiated or ECU-demanded cycle burns off accumulated soot at high temperature. This forces the generator out of service regardless of conditions—including in the middle of an actual outage if the ECU decides it’s had enough.

3. Recovery/Forced Regeneration: The most severe stage, triggered when previous regeneration requests have been ignored or have failed. At this point, the soot loading is critical, and the unit typically requires a service call—often involving physical removal and cleaning or replacement of the DPF element. This is downtime measured in days, not hours, while you wait for a qualified technician with the right diagnostic tools and a clean or replacement filter.

The first two stages burn excess fuel with zero benefit to the operator. The third is unplanned downtime at the worst possible moment. And here’s the critical point for reliability engineers: there is no “battle short” in these systems. In the Navy, a battle short is a switch that bypasses safety interlocks during combat—because sometimes the mission is more important than the protection. Tier 4 emission controls have no such provision. If the ECU senses a DPF problem, the inducement controls engage automatically. The generator derates. There is no override, no bypass, no emergency exception. Your standby generator—the unit whose entire purpose is to be there when the grid isn’t—can reduce its own output or refuse to run during the very emergency it was installed to serve.

Problem 2: DEF Logistics and the Shelf Life Trap

The second issue is the DEF itself. Diesel Exhaust Fluid is perishable. It has a shelf life of approximately 12 months when stored between 12°F and 86°F. Push storage temperatures above 86°F—as happens routinely in generator enclosures in the southern half of the country during summer—and that shelf life drops to as little as six months. Above 95°F sustained, you might get two months before the urea begins to decompose and the concentration drifts out of spec. And unlike diesel fuel, DEF is corrosive to aluminum, carbon steel, copper, and zinc, so it requires dedicated storage in stainless steel or high-density polyethylene containers.

Standby generators sit the vast majority of their lives. But having a sufficient quantity of fresh, in-spec DEF on hand to service emission controls during a major extended outage is a hard requirement. The SCR system consumes DEF at roughly a 2% ratio to diesel fuel—for every 50 gallons of diesel burned, you need a gallon of DEF. That ratio may not sound like much but consider a data center campus with thirty to a hundred multi-megawatt standby units, all running at full rated load during an extended grid outage. You are now managing a parallel consumable supply chain—DEF alongside fuel—under emergency conditions. The reality is to be truly ready for an emergency, your onsite DEF runtime storage needs to be equal to your onsite fuel runtime storage. Because if you run out of DEF, you might as well have run out of fuel.

Even a smaller facility like a hospital is required to keep its DEF supply current, continuously rotating stock to ensure freshness, and properly disposing of expired fluid. The expired DEF isn’t classified as hazardous waste under federal rules, but it’s not something you just pour down the drain either—it’s still ammonia-based, and your local jurisdiction may have specific disposal requirements. At scale, this becomes a major management headache and an ongoing expense that has nothing to do with the generator’s actual mission of providing reliable backup power.

The logistics challenge splits two ways. You either install a centralized DEF storage tank with dedicated piping routed to each unit—adding capital cost, maintenance burden, and another fluid system that can leak or freeze—or you maintain individual DEF tanks at every generator and manage inventory rotation unit by unit. Neither option is simple. Both add cost and complexity to a system whose fundamental value proposition is supposed to be “it sits there until you need it, and then it runs.” Regardless of where it sits, the climate control matters.

The Load Bank Solution

It becomes incumbent on the facility manager to make sure their units never get to the point that they need even a first-level active regen, because once that cascade starts, things get progressively worse. The answer is load—artificial load applied to the engine to drive exhaust temperatures up and keep the DPF clean.

The companies that manufacture load banks are making serious money on this issue, and they should be. For facilities buying containerized Tier 4 generators, the smart specification is a radiator-duct-mounted resistive load bank on every unit, sized to achieve at minimum 75% of rated load on its own. The idea is simple: during weekly or monthly test runs, you energize the load bank, bring the engine up to working temperature, drive exhaust heat through the DPF, and burn the soot off before it becomes a problem. Yes, you’re burning diesel fuel and blowing the energy out the radiator duct as waste heat. But the cost of that fuel is trivial compared to the cost of a DPF regen event, a service call, or—worst case—a unit that’s out of service when you need it.

However, there’s another catch, and it’s a regulatory one. Under 40 CFR §60.4211(f), air quality management districts typically limit standby generators to 100 hours of non-emergency operation per year. Many districts further restrict this to 20 non-emergency hours specifically for maintenance and testing. You need at least an hour over 75% load to properly exercise the DPF and burn it clean. Between monthly test runs of 20–30 minutes and the occasional load bank session, 20 hours doesn’t give you much margin to perform proper maintenance, conduct annual load testing per NFPA 110, and do your DPF housekeeping. The regulatory hour budget and the operational requirement are in direct tension with each other.

Not Just Generators: The Agricultural Parallel

These issues are not unique to standby generators. Agricultural equipment has been forced down the Tier 4 road as well, and the parallel is instructive. Like generators, farm tractors frequently operate at reduced loads—light-duty field tasks like mowing, spraying, or running augers don’t put anywhere near full load on the engine. The DPF accumulates soot even during the active working season because many common farm tasks simply don’t generate enough exhaust heat for passive regeneration. It’s not just an idle problem; it’s a light-load problem that persists even when the tractor is in productive use.

Having your tractor go down for 30 to 40 minutes for a parked regen in the middle of a time-critical planting window or during active pest control can be genuinely disastrous. You can’t tell the calendar to wait while your DPF cleans itself. This operational reality has spawned its own niche service industry—companies now offer portable PTO-driven dynamometers that load the tractor’s engine to full rated power through the Power Take Off shaft. The service tech shows up, couples the dyno to the PTO, and provides the precious hour at full load to clean the DPF before it becomes a problem. Avtron (now part of Legrand) is one of the companies building these units—trailer-mounted, air-cooled, capable of loading engines up to 680 horsepower. It’s a business that shouldn’t need to exist, but does, because the emission controls were designed for a duty cycle that doesn’t match how the equipment is actually used.

The agricultural sector’s frustration with DPF and DEF system reliability was a major driver behind the EPA’s March 27, 2026 UQS ruling. President Trump announced the guidance at a White House agriculture event, and Administrator Zeldin framed it explicitly as relief for farmers and truckers. The ag lobby’s political weight moved the needle. Standby generators don’t have that kind of constituency—but they face exactly the same thermal management problem.

The Hilliard Flashpoint

If you think this is all theoretical, it’s not. On March 19, 2026, residents of Hilliard, Ohio—a suburb on the northwest side of Columbus, about 40 miles south of where I’m writing this—packed a public hearing at The Well recreation center. The subject was a draft air emissions permit for 158 diesel backup generators at an Amazon Web Services data center on Scioto Darby Creek Road.

All 19 people who spoke at the hearing opposed the permit. The generators are Tier 2 units—not Tier 4—and residents are pushing the Ohio EPA to require Tier 4 as a condition of the permit. The facility sits within 1,000 feet of a public park, roughly 2,000 feet from a residential neighborhood, and approximately 4,000 feet from Beacon Elementary School. The draft permit would cap NOx emissions at 235 tons per year, though the Ohio EPA noted that comparable data centers in central Ohio have typically emitted less than 20 tons per year.

Amazon’s statement to the media was revealing: their Ohio generators run on average approximately 10 hours per year, primarily for routine testing and maintenance. That’s the profile of a standby fleet. But residents and the city of Hilliard are fighting on a second front as well—an adjacent proposal by AEP Ohio to install 228 Bloom Energy natural gas solid oxide fuel cells on a six-acre site to deliver 73 megawatts of behind-the-meter power to the same campus. The city tried to block that project through rezoning but was overruled by state power siting authority. They’re now appealing in court.

Here’s where it gets interesting for our discussion. The residents demanding Tier 4 generators are asking for exactly the system we’ve just spent this entire article explaining is an operational nightmare. They’re not wrong about the health concerns—diesel particulate is a real respiratory hazard, and nobody wants 158 uncontrolled diesel engines next to their kid’s school. But the Tier 4 “solution” brings its own enormous operational burden: 158 DPFs that need thermal management on engines that run 10 hours a year, 158 DEF systems with perishable fluid, and a maintenance regime that has to keep all of it functional on equipment that sits idle 99.9% of the time. The cost and complexity don’t scale linearly—they scale with the number of units, and 158 is a lot of units.

This is a tension that the regulatory framework hasn’t resolved. The public wants cleaner emissions. The operators know that Tier 4 on standby duty is an operational burden with real reliability consequences. And somewhere in the middle, facility managers are trying to keep the lights on.

And here is a connection my readers will recognize from my earlier piece on the 35 GW of backup generation sitting behind the meter. During Winter Storm Fern in January 2026, the Department of Energy issued 202(c) orders authorizing data centers—including Ohio data centers—to run their “emergency” generators to support the grid. So the generators Amazon calls emergency backup are the same generators DOE treats as dispatchable grid resources when the system gets tight. The residents’ concern that the permit does not clearly limit how often these generators can run is well-founded—“emergency use” is a flexible concept when the Secretary of Energy can redefine it with a letter.

It is also worth noting that under Ohio House Bill 15—the behind-the-meter generation law—much of this development bypasses local zoning review entirely. Residents find themselves fighting two separate regulatory battles through two different state agencies, and their own city government says its hands are tied. More than 50 data centers in Ohio have already received air pollution permits, mostly for diesel generators. The scale of what is being built in central Ohio communities, largely without meaningful local input, is extraordinary.

Looking Ahead

It goes without saying there is much to be resolved on the EPA front before the vision of Secretary Wright and the broader deregulatory agenda can come to fruition. The UQS ruling is a real and meaningful step—removing a failure-prone component that caused more problems than it solved. But the fundamental tension between Tier 4 aftertreatment and standby duty cycles remains unaddressed.

In Part 3, we’ll look at the natural gas alternative—standby generators running on spark-ignited natural gas engines under 40 CFR Part 60 Subpart JJJJ (Quad J). These engines have no DPF, no DEF, no aftertreatment system at all on standby duty. For those of us who’ve spent careers keeping the lights on, it’s worth understanding why that regulatory asymmetry exists and what it means for the future of backup generation.

Please share your thoughts, and thank you for reading.

Reference Links:

Hilliard coverage (Columbus Dispatch via Yahoo): Public opposed to 158 diesel generators

Hilliard coverage (ABC6/WSYX): Residents push back on generator plan

Signal Ohio (fuel cells + generators): Columbus suburb sues to block Amazon fuel cells

EPA UQS Guidance (official): EPA removes DEF sensor requirement

FreightWaves clarification: What the DEF guidance actually says and doesn’t

Ohio Capital Journal (broader context): Data centers and local government dysfunction

Hello readers, today we are heading back to those hidden 35 GW we discussed two weeks ago. I wanted to dig into what the regulations are, what compliance to the top tier really means, and who the players are in this market. To keep this digestible, I am breaking this into several short pieces.

The vast majority of standby generators in the size range that serve data centers are diesel, or as the EPA identifies them, compression ignition (CI) engines. Diesel engines do not have spark plugs. Instead, they have high compression ratios. This means the air in the cylinder is squeezed an extra amount; compressed air gets hot, and highly compressed air is hot enough to ignite the diesel fuel when it is sprayed into the cylinder. So, when you read the regulations, the dividing line is compression ignition and spark ignition (SI), or oil-based fuels like diesel and gas-based fuels like natural gas and gasoline. The EPA regulates these two categories under separate rules: 40 CFR Part 60, Subpart IIII (commonly called “Quad I”) for compression ignition engines, and 40 CFR Part 60, Subpart JJJJ (“Quad J”) for spark ignition engines. When you see those references in permitting documents, that’s what they’re pointing to.

There have been massive increases in the regulation of diesel engines over the last thirty to forty years. The requirement for low sulfur diesel was the first of these, rolled out in 1993, and it marked the increase in price for diesel from less than gasoline to more than gasoline. The regulation was tightened to Ultra Low Sulfur Diesel (ULSD) in 2006, further reducing the sulfur content to a maximum of 15 parts per million. That move also significantly increased the life of diesel engines because of the corrosive nature of high sulfur fuels. What many people don’t realize is that ULSD was also a prerequisite for the Tier 4 aftertreatment systems that came later. The diesel particulate filters (DPF) and selective catalytic reduction (SCR) catalysts that Tier 4 demands are poisoned by sulfur—EPA had to clean up the fuel before they could mandate the aftertreatment hardware. The two regulations were designed as a system.

Starting in 1996, the EPA rolled out their Tier system for compression ignition engine emission controls. This was intended to clean up the exhaust from diesel engines, a largely unregulated sector at that point.

As soon as the Tier system rolled out, it marked the end of the much-loved Detroit Diesel engine, an industrial staple since WWII. The Detroit was a two-stroke engine and was simply too dirty to meet the new regulations. Regardless, there are huge numbers of these Detroit engines still in service across the industry. They are grandfathered and will likely stay until parts disappear. Its big brother, the EMD (Electro-Motive Division) locomotive engine still continues in thousands of locomotives across the world as well as larger generation plants like Kauai. Unlike Detroit Diesel, EMD remains in business as a Caterpillar subsidiary through Cat’s acquisition of Progress Rail. Wärtsilä’s marine two-stroke business was spun off as WinGD (Winterthur Gas & Diesel), now owned by China State Shipbuilding Corporation, and MAN Energy Solutions (formerly MAN B&W) still manufactures extremely large two-stroke marine propulsion engines for use in cargo and tanker vessels. These marine engines typically burn Heavy Fuel Oil (HFO).

The EPA Tier program was rolled out over a number of years in a four-tier system, each tier tightening the regulations on diesel emissions. This was intended to reduce the industry’s impact and give manufacturers time to catch up. All emission values below are expressed in grams per brake horsepower-hour (g/bhp-hr), which is the unit you will typically see on American engine specification sheets. Here are the tiers and what each one implemented.

1. Tier 1 (1996–2003 phase-in): The first real federal standards for off-road diesels. Set baseline limits on NOx, hydrocarbons (HC), carbon monoxide (CO), and particulate matter (PM). Compared to unregulated pre-Tier engines, this was a modest but meaningful first cut—mainly achievable through combustion optimization (timing, injection pressure, air handling) without exhaust aftertreatment.

   1. Engines 175–300 hp: NOx 6.9 / HC 1.0 / CO 8.5 / PM 0.40

   2. Engines 300–600 hp: NOx 6.9 / HC 1.0 / CO 8.5 / PM 0.40

   3. Engines 600–750 hp: NOx 6.9 / HC 1.0 / CO 8.5 / PM 0.40

   4. Engines 750–1200 hp (generator sets): NOx 6.9 / HC 1.0 / CO 8.5 / PM 0.40

   5. Engines >1200 hp (generator sets): NOx 6.9 / HC 1.0 / CO 8.5 / PM 0.40

2. Tier 2 (2001–2006 phase-in): Tightened the NOx and PM limits further, still largely achievable through in-cylinder combustion improvements—better fuel injection systems (early common rail), turbocharging refinements, charge air cooling. Some engines started using cooled EGR (exhaust gas recirculation) at this stage to knock down NOx formation temperatures. Note that starting at Tier 2, EPA combined the HC and NOx limits into a single NMHC+NOx standard.

   1. Engines 175–300 hp: NMHC+NOx 4.9 / CO 2.6 / PM 0.15

   2. Engines 300–600 hp: NMHC+NOx 4.8 / CO 2.6 / PM 0.15

   3. Engines 600–750 hp: NMHC+NOx 4.8 / CO 2.6 / PM 0.15

   4. Engines 750–1200 hp (generator sets): NMHC+NOx 4.8 / CO 2.6 / PM 0.15

   5. Engines >1200 hp (generator sets): NMHC+NOx 4.8 / CO 2.6 / PM 0.15

3. Tier 3 (2006–2008 phase-in): Another incremental tightening, primarily targeting NOx and HC. Still mostly an in-cylinder strategy—advanced electronic fuel injection, refined EGR, improved combustion chamber geometry. The key thing about Tiers 1 through 3 is that manufacturers could generally meet them without bolting on exhaust aftertreatment hardware. Tier 3 did not adopt its own PM standard; engines continued to meet the Tier 2 PM limits. Critically, Tier 3 was never promulgated for engines above 750 hp. Those larger engine categories jumped directly from Tier 2 to Tier 4. This becomes important when we discuss the emergency standby exemption below.

   1. Engines 175–300 hp: NMHC+NOx 3.0 / CO 2.6 / PM 0.15 (carried from Tier 2)

   2. Engines 300–600 hp: NMHC+NOx 3.0 / CO 2.6 / PM 0.15 (carried from Tier 2)

   3. Engines 600–750 hp: NMHC+NOx 3.0 / CO 2.6 / PM 0.15 (carried from Tier 2)

   4. No Tier 3 standard exists for engines above 750 hp.

4. Tier 4 (Interim: 2008–2012, Final: 2012–2015 phase-in): This is the big one. Tier 4 Final demanded roughly a 90% reduction in NOx and PM compared to earlier tiers. This could not be met with combustion tricks alone—it forced the adoption of exhaust aftertreatment systems:

   1. DPF (Diesel Particulate Filter): Traps soot, periodically regenerates by burning it off at high temperature.

   2. SCR (Selective Catalytic Reduction): Injects DEF (diesel exhaust fluid, a urea solution) into the exhaust stream to chemically reduce NOx to nitrogen and water.

   3. DOC (Diesel Oxidation Catalyst): Oxidizes CO and unburned hydrocarbons.

      1. Tier 4 iterim emission standards:

         1. Generator sets 750–1200 hp, Tier 4 Transitional (2011–2014): NMHC 0.30 / CO 2.6 / PM 0.08

         2. Generator sets >1200 hp, Tier 4 Transitional (2011–2014): NMHC 0.30 / NOx 0.50 / PM 0.08

      2. Tier 4 final emission standards

         1. Engines 175–300 hp (2014+): NMHC 0.14 / NOx 0.30 / CO 2.6 / PM 0.01

         2. Engines 300–600 hp (2014+): NMHC 0.14 / NOx 0.30 / CO 2.6 / PM 0.01

         3. Engines 600–750 hp (2014+): NMHC 0.14 / NOx 0.30 / CO 2.6 / PM 0.01

         4. Generator sets 750–1200 hp, Tier 4 Final (2015+): NMHC 0.14 / NOx 0.50 / CO 2.6 / PM 0.02

         5. Generator sets >1200 hp, Tier 4 Final (2015+): NMHC 0.14 / NOx 0.50 / CO 2.6 / PM 0.02

Note that EPA gave the large generator set categories (above 750 hp) a slightly more relaxed NOx number—0.50 versus 0.30 for the smaller categories—and a PM number of 0.02 versus 0.01. Still requires aftertreatment, but a modest acknowledgment that the largest engines have a harder time meeting the absolute tightest limits.

To put the magnitude of these changes in perspective: going from Tier 1 NOx of 6.9 to Tier 4 Final of 0.30 g/bhp-hr is a 96% reduction. PM from 0.40 to 0.01 is a 97.5% reduction. Those numbers are physically impossible to achieve through combustion optimization alone, no matter how sophisticated the fuel injection, turbocharging, or EGR systems become. That is what forces the entire aftertreatment stack—DPF, SCR, DOC—into existence. Tiers 1 through 3 were evolutionary improvements to the engine itself. Tier 4 is a fundamentally different compliance architecture that bolts an entire chemical processing system onto the exhaust.

If you are someone who loves a diesel light truck, you have lived these transitions and probably used some colorful language as each tier was implemented. But it wasn’t without purpose. As we learn more, it has become increasingly apparent that exposure to PM 2.5—soot—is a real health problem, and particulate emissions from diesel engines were a big contributor. As someone who takes asthma medications and has children who also suffer from asthma, it is hard for me personally to kick against this particular rock. But Tier 4 brings a host of operational issues I will talk about in a later installment.

Fortunately, the EPA gives diesel generators an out. Because emergency standby generators only run limited hours, they are not considered a major source of air pollution by the federal EPA. So, the EPA allows emergency generators to be certified to lower tier standards: Tier 3 for engines from 50 to 750 hp, or Tier 2 for engines above 750 hp. The reason for the split is that Tier 3 was never promulgated above 750 hp, so there is no Tier 3 standard for those larger engines to meet—they certify to Tier 2 instead. Either way, these are emission levels achievable without exhaust aftertreatment.

To qualify for this exemption, the engine must follow strict operational rules defined in 40 CFR §60.4211(f). The rules break down into three categories:

Actual emergencies: There is no hour limit on operation during genuine emergencies—power outages, fires, floods, or grid failures. Run it as long as you need it. The regulation also counts voltage or frequency deviations of 5% or greater below standard as an emergency condition, which effectively covers any serious grid disturbance.

Maintenance and testing: A maximum of 100 hours per calendar year for all non-emergency operation, including monthly functional tests, annual load bank verification, ATS transfer testing, and any other readiness testing. The engine must be equipped with a non-resettable hour meter—that is a specific regulatory requirement, not optional.

Non-emergency situations: 50 hours per calendar year, but this is carved out from within the 100-hour total—it is not in addition to it. If you use 50 hours for non-emergency operation, you only have 50 hours left for maintenance and testing. And these 50 hours come with strict restrictions: they cannot be used for peak shaving or non-emergency demand response, and they cannot be used to generate income or supply power as part of a financial arrangement. There is one narrow exception: the hours can be used to supply power when dispatched by the local balancing authority or transmission operator specifically to avert voltage collapse or line overloads.

The bottom line is that the 100 hours is a hard number. If you exceed it, or use the engine for prohibited purposes, the unit is permanently reclassified as a non-emergency engine and must meet Tier 4 Final standards. That is not a fine—that is a potential engine replacement or an extremely expensive aftertreatment retrofit. It is up to the local authority to track the hours on these units.

The real fly in this ointment is the patchwork of local and regional air quality districts with their own more restrictive regulations. Most people think that this only applies to the entirety of California, but they would be incorrect. There are many rural air districts in California that have fairly reasonable air quality regulations for standby generators. California’s Airborne Toxic Control Measure (ATCM) for stationary compression ignition engines only applies to engines above 50 brake horsepower, and the state has 35 independent air districts, each with authority to set their own requirements. The most aggressive districts are South Coast AQMD (covering the Los Angeles basin) and Bay Area AQMD, which increasingly mandate Tier 4 as Best Available Control Technology (BACT) even for emergency standby engines. But there are plenty of California air districts that don’t go nearly that far.

California is not alone in having districts that make operating equipment a difficult task. The NESCAUM states—Connecticut, Maine, Massachusetts, New Hampshire, New Jersey, New York, Rhode Island, and Vermont—operate as a bloc with enhanced permitting requirements for stationary diesel engines. New York City requires air permits for engines 200 hp and above. New Jersey imposes operation restrictions on poor air quality days. The Houston and Atlanta metro areas are also notably restrictive as EPA ozone nonattainment areas. Some of these districts restrict non-emergency run time to as low as 20 hours a year. Many local districts implement Tier 4 requirements even though Tier 2 or Tier 3 meets federal standards.

These extremely restrictive air districts are scattered all across the country, and some are in states that would surprise you. Of all of them, the South Coast AQMD is probably the most famous, and is known to be extremely difficult to work with. But SCAQMD certainly isn’t the only one that is a challenge. This makes specifying equipment a nightmare because you have to be very careful to observe the specific regulations covering relatively small regional areas, or find yourself on the wrong end of the law. This is one of the things that will make Secretary Wright’s 202(c) proposal a nightmare to implement. When regulations can change by the city block, it becomes almost impossible.

Next we will talk about some of the realities of a Tier 4 system when applied to a standby generator. As always, please share your thoughts about this piece.

Hello readers, thanks for coming back. This piece is a promised piece in response to a request that I describe how outage management systems work, what makes them tick. Let’s get into it

First for the most part Distribution Outage Management, I will just call it OMS (Outage Management System) after this, often gets confused with Transmission OMS. The two systems have two completely different functions. On Distribution it’s the system that tells the utility your power is out, sends you (the customer) text messages, and drives the map you can look at to see how many people are out with you. Transmission OMS is all about planning and managing who gets to take what out when, so we don’t get in trouble. From here on we are only talking about Distribution OMS.

OMS is usually going to be part of the Supervisory Control And Data Acquisition (SCADA) system, or a more complex Energy Management System (EMS). However there are vendors that make stand alone OMS programs that often run outside the secure SCADA world on the business environment. Milsoft owns a major slice of the standalone OMS market for small Co-ops that often run small legacy SCADA systems with limited capability. It’s packaged to run on the business environment, collect outage data via firewall connections to SCADA, and Smart Meter data from the Customer Information System (CIS). If you want to look at where OMS started, the Milsoft product is a good example, it started as something stand alone.

The Milsoft style systems are a legacy that are rapidly disappearing as SCADA systems grow more complex. Now OMS is often just a part of a larger package known as Advanced Distribution Management Systems (ADMS) that has grown well beyond the boundaries of OMS, but let’s not get drawn down that rabbit hole. ADMS is a topic for another day (see Smart Grid - Where We Are Now). Let’s get into the OMS system itself and what makes it work.

The foundation of any modern OMS system is a Geographic Information System (GIS) map where all the customer addresses and equipment are geographically located with GIS addresses. So everything on the map ties to the global positioning system that drives maps like Google Maps. Every customer address is tied to a meter number, and that data lives on the map. On the GIS map the tech builds a connectivity model. The connectivity is built on a per phase basis of a three phase system. So every meter is connected to a transformer, every transformer is connected to a junction box or overhead line. Every line or junction box will eventually connect to a protective device. Those protective devices connect upstream to larger protective devices that serve several downstream devices until it connects to the breaker at the substation. Think about it as a river system in reverse with power flowing upstream to tributaries and finally to a small spring. This connectivity model is an electronic version of how things are actually connected in the real world. The connectivity model includes all the switches and open cross connections that allow power to be rerouted. These electronic switches can be operated by the operator, so the model reflects what is actually going on with the distribution system. This connectivity model is the basis for how the OMS resolves outages.

When a customer reports an outage to an OMS, it is usually done via an Interactive Voice Response (IVR) system, you know, that hated computer that answers the phone. But in this case it helps because an IVR can answer and process thousands of calls in the same time it takes a human Customer Service Representative (CSR) to answer a few dozen. The IVR flags the customer address as out in the OMS and the OMS generates an outage ticket. If calls continue to come in, and more addresses get flagged, it moves the outage disconnect point upstream to the next protective device, it flags all of the downstream addresses as out of power. Within just a few minutes a pretty clear picture of the size of the outage emerges. It can also be driven by a breaker operating that SCADA can detect and provide that input into the OMS. Most OMS systems will lock an outage after a preset time period of no new activity, ten to twenty minutes. After that, any new calls outside the outage window are flagged as a new outage.

If the system has smart meters or Automated Meter Reading (AMR) equipment deployed, it changes the game a bit. As soon as the first smart meter reports an outage or the first customer calls, the OMS will “ping” or request a status report from all the AMR meters in that general area. As AMR meters report back if the power is on or off the outage boundaries are determined within a minute or two, often before the operator can even dispatch a troubleshooter. AMR meters also help on restoration because once power is restored, the meters can be pinged to verify all the power is back on.

Because the connectivity model reflects actual field conditions, if switching is performed to restore all or some of the customers, it can be accurately reflected in the OMS.

Beyond simply reporting outages, the system also records outage counts and outage lengths so accurate reports can be generated for oversight agencies. These are the numbers behind the reliability indices you may have heard of: SAIDI (System Average Interruption Duration Index), SAIFI (System Average Interruption Frequency Index), and CAIDI (Customer Average Interruption Duration Index). Those indices are what regulators use to grade your utility’s performance, and they drive where the utility spends money on things like pole replacements, tree trimming, and equipment upgrades. The maps and customer counts are also ready sources of reporting for media during storm events.

So next time your power is out and you pull up that map on your utility app to see how big the outage is, you have an idea of what is going on in the background. The OMS is not just telling you the lights are off. It’s the system that tells your utility where the problem is, how many people are affected, where to send the crews, and ultimately where to spend money to keep it from happening again. I hope you enjoyed the short piece, as always leave me a comment.