Curtailment: Economic, Congestion and Manual — What Causes Each, and Who Pays
In 2024, grid congestion forced ERCOT to curtail more than 8 TWh of wind and solar energy. The West zone alone accounted for 3.1 TWh of wind and 2.2 TWh of solar. Across the year ERCOT has been curtailing an average of 1.2 GW every hour — the equivalent of switching off thirty-seven 125 MW solar farms, continuously.
California's numbers are smaller and differently shaped. CAISO curtailed 3.4 TWh in 2024, up 29% on 2023, and 93% of it was solar. The first five months of 2025 saw 2,742 GWh curtailed, with March 2025 setting a single-month record of 919 GWh.
These are two different diseases with the same symptom. ERCOT's curtailment is a transmission problem — energy that cannot get out of a pocket. California's is an oversupply problem — energy nobody needs at midday. The remedies differ, the contractual treatment differs, and a model that treats curtailment as one undifferentiated haircut will get both wrong.
This post closes Series C with the market side of curtailment: what causes each type, how large each has become, what actually reduces it, and the double-counting error that shows up in almost every merchant revenue model.
ℹ️ Note: The contractual allocation of curtailment between buyer and seller — deemed generation, free allowances, PTC gross-ups — was covered in the Series D post on that subject. This post is about the market cause and the magnitude.
What Are the Types?
Four distinct causes, which happen to produce identical meter readings.
| Type | Cause | Typical market |
|---|---|---|
| Oversupply | System-wide generation exceeds demand — the duck curve | CAISO |
| Congestion | Transmission cannot move energy out of a pocket | ERCOT |
| Economic | Prices are negative or near zero; generating is irrational | Both |
| Minimum generation / inertia | Must-run conventional plant held online for stability | Both |
The distinction between the first three matters because they respond to completely different interventions. Oversupply is solved by storage, flexible load, or export. Congestion is solved by transmission, or by moving the generation. Economic curtailment is a price response and is, in one sense, the market working correctly.
Economic or Congestion? The Line Is Blurrier Than It Looks
The categories overlap, and the overlap is where most of ERCOT's curtailment actually sits.
A node-level study found that 74.3% of curtailment events were driven by unpriced congestion signals, forcing economic curtailment when system loads dipped below 40 GW. Read that carefully: the mechanism was economic — prices fell and generators stopped — but the cause was congestion.
That is the same identity the first post in this series established from the price side. Congestion depresses the nodal price; below a threshold the generator stops; the price effect becomes a volume effect. Whether an hour is recorded as "economic" or "congestion" curtailment is partly a classification convention and partly a question of whether the constraint was priced into the dispatch or acted on directly.
For a developer the practical consequence is that arguing about which category an hour belongs in is usually less useful than looking at where it happened. Curtailment is overwhelmingly local, and the location tells you which disease you have.
Two Markets, Two Diseases
Worth separating properly, because the diagnosis determines the remedy.
ERCOT is a transmission problem. Installed wind and solar reached 65 GW in 2025, supplying 36% of ERCOT demand between January and September, with utility-scale solar output rising 50% year on year to 45 TWh over that period. The West zone and the Panhandle hold the best resource in the market and the worst export constraints — the same pockets the basis post identified. Energy is curtailed there not because Texas does not want it but because it cannot leave.
CAISO is an oversupply problem. Solar at 93% of curtailment, concentrated at midday, with curtailment in 2026 jumping by nearly 2 GW over the midday solar peak — enough that, adding curtailed output back, average hourly solar generation would have pushed well over 20 GW from late morning through the afternoon. This is the duck curve in its purest form: a system that has more solar than it can absorb in the hours solar produces, regardless of transmission.
The remedies follow directly. More transmission helps ERCOT and does comparatively little for CAISO's midday problem. Storage helps both but for different reasons — in ERCOT it moves energy past a constraint in time rather than space; in CAISO it moves it from a surplus hour to a deficit one. Flexible load helps CAISO most, because the problem is that demand is not there when supply is.
How Does This Relate to Basis?
Same underlying cause, two expressions, and the modelling error is to count them twice.
This deserves to be stated precisely, because it is the most common defect in the merchant revenue models this series has been describing.
Congestion in a pocket first depresses the nodal price — that is basis, a price effect, covered in the first post. As the constraint deepens, the price goes negative and generation stops — that is curtailment, a volume effect. They are sequential stages of one mechanism, not two independent risks.
A model that applies a generation-weighted basis deduction to all available megawatt-hours, and then separately deducts curtailed megawatt-hours, has charged the worst hours twice: once as a price haircut on energy that was never delivered, and again as lost volume.
The correct sequence is:
1. Available generation (the P50)
2. Less curtailed volume → delivered generation
3. Apply basis to DELIVERED volume only
Or, more simply, use a capture rate that embeds both — but then do not apply a separate basis deduction on top of it.
What Actually Reduces It?
Four things, and one of them is now quantified well enough to model directly.
Co-located storage. This is the finding worth carrying. In the first half of 2025, co-located solar sites in ERCOT captured 72% of their average locational price, against 57% for standalone sites — fifteen percentage points of capture rate, from putting a battery at the same point of interconnection. The battery absorbs energy in the hours the node is saturated and releases it when the constraint clears, converting curtailment and negative pricing into arbitrage.
It is no coincidence that Texas nearly doubled its battery fleet in 2025, adding 6 GW to reach 13.9 GW and 22.9 GWh of operational grid-scale storage by early 2026, concentrated in exactly the pockets with the worst curtailment.
Transmission. The direct remedy for the ERCOT disease, and the slowest. Everything the previous two posts said about the interconnection process applies: cost allocation improved, build rates did not.
Load growth in the pocket. Data centres, electrolysers and industrial load sited behind the constraint absorb generation locally. This is the fastest-moving of the four and the least controllable by any individual project.
Flexible demand. The CAISO remedy: demand that can move into the midday surplus. Vehicle charging, thermal storage, and industrial processes with tolerance for timing.
Note what is absent from that list: nothing a solar project can do to itself reduces curtailment. Tracking, module choice and inverter sizing change the production profile marginally; they do not change whether the network can take the output. Curtailment is a property of the location, not of the plant.
There is one partial exception worth noting, because it cuts against the usual instinct. A higher DC-to-AC ratio — oversizing the array relative to the inverters — is normally a good trade, buying more energy in the shoulder hours at the cost of clipping at midday. In a pocket where midday is precisely when the node saturates, that trade improves: the clipped hours were going to be curtailed or priced near zero anyway, and the shoulder hours the extra DC capacity fills are the ones that still clear. The optimal DC/AC ratio is therefore not a purely technical number; it depends on the shape of the local price and curtailment profile, and a project designed against a flat price assumption will have picked the wrong one.
That is a design decision responding to a locational problem, which is about as close as plant configuration gets to addressing curtailment — and it is worth running explicitly rather than inheriting a standard ratio from a portfolio template.
Minimum Generation and Inertia: The Type Nobody Models
The fourth cause, and it is the one most likely to be absent from a model entirely.
A power system needs more than energy. It needs inertia, voltage support, and enough synchronous plant online to ride through a fault. Where those services can only be provided by conventional generation, some of that plant is held on at its minimum stable output regardless of price — and in an hour when renewables could serve the whole load, the must-run floor displaces them.
This is why curtailment can occur at moments when, on an energy balance alone, it should not. The system is not short of megawatt-hours; it is short of the physical characteristics that come attached to a spinning machine.
Two features make it awkward to model.
It is technology-dependent and changing. Grid-forming inverters, synchronous condensers and evolving reliability standards all reduce the must-run floor over time. A curtailment assumption calibrated on today's minimum generation requirement is calibrated on a constraint that is actively being engineered away — which for once is a trend running in a project's favour.
It is invisible in the classification. An hour curtailed because thermal plant was held on for inertia typically appears in the data as economic or oversupply curtailment, because the visible mechanism was the price. The underlying cause is a reliability constraint, and it responds to completely different interventions than either storage or transmission.
The honest treatment is to acknowledge it as a component of the observed curtailment history without trying to separate it out, and to note that the direction of travel on this particular cause is downward.
Is It Getting Worse?
Yes, and the reason is a race with a predictable outcome in the short term.
The trend lines are unambiguous. CAISO curtailment rose 29% from 2023 to 2024. The first five months of 2025 saw 2,742 GWh curtailed, with March setting a single-month record of 919 GWh. In 2026 midday curtailment jumped by nearly 2 GW. ERCOT installed wind and solar reached 65 GW in 2025, with utility-scale solar output alone up 50% year on year.
The mechanism is straightforward arithmetic. Generation is being added in months. Transmission is added in years. Load takes years to site and build. Storage is the fastest of the remedies — Texas doubled its fleet in a single year — but even that is chasing a build-out of generation that is larger and faster.
Two things follow that are worth separating, because they point in opposite directions over different horizons.
In the near term, curtailment rises. Each new project in a constrained pocket adds to the volume that cannot leave, and the remedies do not arrive on the same timescale. A model assuming today's curtailment rate holds for a project entering service in three years is, on current trends, optimistic.
Over the longer term, it is a self-limiting problem. Curtailment is an economic signal, and a sufficiently bad one stops new development in the pocket, redirects it elsewhere, and justifies the transmission and storage that relieve it. The question for any individual project is not whether the pocket eventually clears but whether it clears within the project's debt tenor.
That is the shape of the assumption a model should carry: elevated and rising for some years, improving thereafter, with the timing of the improvement tied to identifiable transmission and load additions rather than to a general expectation that things get better.
Who Bears It Contractually?
Briefly, since the Series D post covered this in full: it depends entirely on cause, and the categories in that contract are not the same as the categories in this post.
The general pattern is that buyer-instructed curtailment is compensated — the offtaker told the plant to stop, and pays deemed generation for what would have been produced — while congestion and economic curtailment are allocated to the seller, on the reasoning that the seller chose the location.
That allocation is the reason this post matters commercially. The 8 TWh ERCOT curtailed in 2024 was overwhelmingly congestion-driven, which means it was overwhelmingly uncompensated. A free allowance in the PPA can absorb the first slice; beyond that the generator carries it.
How Do You Model Curtailment in Excel?
As a volume deduction that precedes the price calculation, with the co-location case run explicitly.
The inputs
Assumptions, labelled as such:
Capacity 200 MW
Available (P50) generation 438,000 MWh
Average locational price $40.00/MWh
Curtailment rate, standalone 8%
Price capture rate, standalone 57%
Curtailment rate, co-located 4%
Price capture rate, co-located 72%
The standalone case
Available generation 438,000 MWh
Less curtailment 8% (35,040) MWh
Delivered generation 402,960 MWh
Realised price $40.00 × 57% $22.80/MWh
Revenue 402,960 × 22.80 = $9,187,488
The co-located case
Available generation 438,000 MWh
Less curtailment 4% (17,520) MWh
Delivered generation 420,480 MWh
Realised price $40.00 × 72% $28.80/MWh
Revenue 420,480 × 28.80 = $12,109,824
The value of the battery, on the solar revenue alone
Co-located revenue $12,109,824
Standalone revenue $9,187,488
-----------
Uplift $2,922,336 (+31.8%)
A thirty-two percent uplift on the solar revenue line, before the battery earns a single dollar of its own. That is the number that explains the ERCOT build-out, and it is routinely missing from co-location business cases — which tend to model the battery's own revenue stack and treat the host's improvement as a qualitative benefit.
The decomposition, so the model says which effect did the work
Volume effect (420,480 − 402,960) × $22.80 = $399,456 13.7%
Price effect 420,480 × ($28.80 − $22.80) = $2,522,880 86.3%
-----------
$2,922,336
The battery earns most of that by improving when the host sells, not by preventing curtailment. Two thirds of a project's exposure in a congested pocket is price, not volume — which is consistent with everything the first post in this series said and is the opposite of how curtailment is usually discussed.
ℹ️ Note: Do not apply a separate basis deduction on top of a capture rate. The 57% and 72% figures already embed the price effect of congestion. Charging basis again is the double-count described above, and on these numbers it would overstate the loss by roughly $2.4m a year.
To build the curtailment and capture-rate model, the co-location comparison and the volume/price decomposition, prompt Dezzmond with your node and generation profile.
Why a P90 Case Does Not Cover This
A point worth making explicitly, because it is a genuine and common confusion about what conservatism means.
Energy yield assessments produce P50, P90 and P99 cases, and lenders routinely size debt on P90. That is real conservatism — about the resource. The P90 is a statement that in nine years out of ten, the wind will blow or the sun will shine at least this much.
It says nothing whatever about whether the network will accept the output.
Resource variability and curtailment are independent quantities with different distributions and different drivers. A P90 resource year in a congested pocket can be curtailed more heavily than a P50 year, because curtailment depends on what else is generating and what the network can carry, not on how good the resource was. In some circumstances the relationship is actively perverse: a strong regional wind year is one in which every wind farm in the pocket is producing hard simultaneously, which is exactly when the export constraint binds.
So a lender sizing on P90 generation with a flat curtailment assumption has applied conservatism to one variable and a point estimate to another, and believes it has been conservative overall.
The correct treatment is to carry curtailment as its own distribution, node-specific, and to test the combination — a P90 resource year with an adverse curtailment year — rather than assuming the two haircuts are additive or that one covers the other. Series F returns to what P50, P90 and P99 do and do not mean; the point here is narrower, which is that none of them are about the grid.
What Do Developers and Lenders Actually Check?
- Is curtailment deducted before basis, or are both applied to the same megawatt-hours?
- Is the assumption node-specific? A system-wide curtailment average is close to meaningless.
- Which disease is it — congestion or oversupply? The remedies do not transfer.
- Is curtailment compensated under the PPA, and is the free allowance exhausted?
- Has the co-location case been run on the host's revenue, not just the battery's?
- What is in the queue behind this project in the same pocket?
- Is there a funded transmission upgrade that would relieve the constraint, and when?
Frequently Asked Questions
How much renewable energy is being curtailed?
ERCOT curtailed more than 8 TWh of wind and solar in 2024, averaging about 1.2 GW every hour across the year. CAISO curtailed 3.4 TWh in 2024, 29% more than 2023, with 93% of it solar.
What is the difference between economic and congestion curtailment?
Economic curtailment is a price response — prices go negative and generating stops. Congestion curtailment is a physical constraint. In practice they overlap heavily: one node-level study attributed 74.3% of events to unpriced congestion signals producing economic curtailment.
Why is California's curtailment different from Texas's?
California's is oversupply — too much solar at midday relative to demand, which transmission does not fix. Texas's is congestion — energy that cannot leave a pocket, which transmission does fix. Storage helps both, for different reasons.
Does co-locating storage reduce curtailment?
Yes, and it does more for price capture than for volume. In the first half of 2025, co-located ERCOT solar sites captured 72% of their average locational price against 57% for standalone — with most of the benefit coming from selling at better hours rather than from avoiding curtailment.
Does sizing debt on a P90 case cover curtailment risk?
No. P90 describes resource variability, not network availability. The two are independent, and a strong regional resource year can produce worse curtailment because every plant in the pocket generates hard at once. Curtailment needs its own node-specific distribution.
Is curtailment going to keep rising?
In the near term, on current trends, yes — generation is added in months while transmission, load and storage take years. Over a longer horizon it is self-limiting, because the signal redirects development. The question for a project is whether the pocket clears within its debt tenor.
Who pays for curtailed energy?
Under most PPAs, buyer-instructed curtailment is compensated through deemed generation while congestion and economic curtailment sit with the seller, subject to any free allowance. Since the great majority of ERCOT's curtailment is congestion-driven, most of it is uncompensated.
Closing: Series C, and the One Variable Underneath It
Eight posts on market structure, and they converge on a single point that is easy to state and routinely ignored: location is a revenue variable, and it is usually the largest one nobody modelled.
Basis is location expressed as price. Curtailment is location expressed as volume. Capacity accreditation is location and technology expressed as a reliability payment. Network upgrades are location expressed as capital. The battery revenue stack is location expressed as volatility. Even the interconnection queue is a mechanism for rationing access to specific points on a specific network.
None of that appears in an energy yield report, and very little of it appears in a power price curve. Both of those documents are careful, well-governed and usually close to correct — and a project can be built on both of them and still miss its case by twenty percent, because the questions they answer are not the questions that determine revenue.
The practical discipline that follows is narrow. Model at the node. Deduct volume before price. Never apply basis and a capture rate to the same megawatt-hour. Run the co-location case on the host as well as the battery. And publish the breakeven — the basis, the accreditation, the curtailment rate at which the structure stops working — because that number survives a forecast being wrong and a central case does not.
Series E begins next, on project finance mechanics: what actually sizes the debt, why the constraint changes by market, and the circularity that every sculpting model runs into.
Sources: Modo Energy — The Curtailment Crisis: Saving Wind and Solar Investments in ERCOT · EIA — Solar and Wind Power Curtailments Are Increasing in California · CAISO — Wind and Solar Real-Time Dispatch Curtailment Reports · Grid Status — Curtailment: When We Throw Away Clean Energy