Methodology

Standard carbon accounting compares a captured plant to the same plant uncaptured. That is the right comparison only if the plant would have operated either way. Where capture revenue is what keeps a plant open, the correct counterfactual is the plant closing and its output being served from elsewhere.

λ is the fraction of each captured tonne that represents genuine abatement. For firms that would operate regardless, λ = 1 by construction. For propped firms it can fall well below 1, and below ρk = 1 − η it turns negative — capture becomes worse than closure, because it leaves the uncaptured residual where closure would have removed everything.

Public cost per real tonne is then simply the credit divided by λ.

The framework is closely related to the carbon-leakage literature, which measures the same quantity from the opposite direction. OECD work finds leakage offsets ~13% of domestic emission reductions in cement and steel; note that those estimates measure the marginal response to a carbon price rather than a plant closure, so they bundle ρ and k together and include demand destruction. Useful as a check that ρ < 1 in reality, not as a direct estimate of k.

The derivation assumes all of a closed plant's output is re-served elsewhere. Four reasons for holding it there:

• It is the assumption most favourable to the capture project, so every propping cost figure is a floor, not a central estimate.
• It isolates the mechanism under test — the intensity of replacement, not its quantity.
• It is close to correct where most contested tonnage sits: power (the grid must balance) and ethanol (RFS volume mandates fix demand by statute).
• Any other value would be invented. No estimate exists.

At ρ = 0, λ = (η−1)/η = −0.111 for any k. But the sign flip requires ρk < 0.10, so replacement would have to fall below ~18% for market pulp — implausible, which bounds the downside.

Holding ρ = 1 understates propping cost by roughly 30–60% at the pessimistic end.

These began as judgement. Grounding them in the literature changed the argument materially, and in two cases reversed it.

Each project is scored on a Counterfactual Viability Index — the estimated probability its host is still operating in 2035 with no capture revenue — from three weighted components: sector demand trajectory (40%), host operating margin excluding capture revenue (40%), and new-build versus retrofit (20%). Below 0.45 is propping, above 0.62 greening, between them contested.

This is a prior, not an estimate, and it is the weakest construct in the analysis. It is why the article leads with the cost table, which requires no classification at all.

The obvious fix — fitting a facility-survival model on the full universe of US plants in these sectors and predicting onto the capture hosts — was considered and rejected on three grounds:

1. Selection into capture is endogenous to survival. A plant about to close does not sign a twenty-year CO₂ offtake. Capture hosts are selected on viability, including on unobservables. A model fitted on the general population would under-predict their counterfactual survival and inflate the propping share — bias running in favour of this analysis's own argument.
2. The control group is contaminated. If the credit keeps marginal plants supplying their markets, it depresses prices for competitors, so the hazard facing non-capture plants already embeds the effect of the policy existing. This is the mechanism Caballero, Hosoi and Kashyap identify in the zombie-lending literature.
3. It is extrapolation. The quantity wanted is survival with no credit at all — off the support of the data since 2018 — in sectors undergoing structural transition.

The result would be standard errors resting on a contaminated control group. A labelled prior is more honest than false precision.

Levelised cost of CO₂ captured splits into an annualised capital charge and cash operating cost:

At 9% WACC over 20 years with an 0.85 capacity factor, CRF = 0.1095. Cash operating cost runs 35–55% of levelised depending on configuration: high-purity streams are capital-light because capture is compression and dehydration on a separation that already exists, while dilute flue gas and direct air capture are capital-dominated.

An important caveat on the capital figures. The sourced quantity is levelised cost. Cash cost was set as an assumed share of it and capital intensity derived by inversion — so the capital numbers carry no independent information and should not be cited as measured project capex. They do land near the two publicly observable figures (cement ~$559/tpa against Heidelberg Mitchell's roughly $500; DAC ~$2,774/tpa against Stratos's roughly $2,600), but those were checks, not inputs.

The two tests that follow from the split: financing requires clearing full levelised cost; continuing to operate requires clearing cash cost only, because capital is sunk. That gap produces the 30% / 69% repeal asymmetry in the article.

40 project or aggregate line items, 85.2 Mtpa of live announced US capture capacity. Compiled from company disclosures, SEC filings, DOE programme documentation, the IEA CCUS projects database and the Global CCS Institute's 2025 status report. Aggregate rows are used where a sector has many small or unnamed proposals.

Roughly 18 free parameters against 41 data rows. At that ratio the model is not estimating anything — it is a structured way of expressing priors, and the article leads with the cost table for that reason.

Three acknowledged omissions all bias the same way, toward overstating abatement: the energy penalty of capture, realised throughput versus nameplate, and storage permanence.

Five substantive corrections were made. Four moved the conclusion against the original thesis, which is the relevant fact when judging whether the remaining findings are motivated.

• Cost concept. The first version shut plants when revenue fell below levelised cost — wrong test for a built asset, whose capital is sunk. Correcting it collapsed modelled stranding and relocated the damage to the investment decision.
• Collinearity. The classification used five components, two of which were pure functions of the same new-build binary — 30% of weight on one variable counted twice.
• Market pulp k. 0.55 → 0.85. Largely dissolved the propping case in the sector the analysis was built around.
• Cement k. 1.05 → 0.88. Imports are cleaner than US production, removing the leakage-prevention defence.
• Cost-of-capital claim. Tested and found null — financing terms change no participation decision.

The clearest test: if k > 1 in pulp and paper — if closures are absorbed by higher-intensity imports rather than cleaner domestic survivors — then propping is carbon-leakage prevention rather than distortion, and the critique loses its economic foundation.

Four measurements would move the central estimates from assumption to evidence, in order of value per unit of effort:

1. Realised versus nameplate volumes (hours) — EPA GHGRP publishes annual injected tonnes per facility.
2. Capex/opex splits by configuration (hours) — NETL and IEAGHG reports; de-circularises the capital figures.
3. Abandonment rate after loss of public support (one day) — the 24 DOE awards cancelled in May 2025.
4. Replacement rate and intensity from closures (weeks) — event study on the 2024–26 pulp closure wave, ~6 Mt with documented dates; recovers ρ and k jointly.

k — pulp & paper

• Springer, Int. J. Life Cycle Assessment — GHG emissions from tissue paper: virgin pulp vs recycled waste paper — Recycled vs virgin kraft intensity ratio ~0.80 (kraft 508 vs recycled 408 kgCO2e/t); virgin ~30% higher over lifecycle. Basis for revising market-pulp k from 0.55 to 0.85
• DOE Bandwidth Study on Energy Use in Pulp and Paper — Mill energy intensity ranges; used with the BioResources review finding that older mills are NOT systematically less efficient, which undercut the original assumption that closures are absorbed by cleaner survivors
• BioResources — Energy efficiency challenges in pulp and paper manufacturing: a tutorial review — Finding that mill age is a poor predictor of energy efficiency

k — cement

• EPA — Cement carbon intensities fact sheet — US clinker 0.787-0.934 tCO2/t (25th-75th percentile). US intensity ~20% ABOVE rest-of-world average, so imports are on average cleaner. Basis for revising cement k DOWN from 1.05 to 0.88
• Carbon Brief — Why cement emissions matter for climate change — Cross-check on international cement intensity comparison

k — steel

• IEEFA steel fact sheet — BF-BOF ~2.2 tCO2/t crude steel; basis for revising steel k UP from 1.20 to 1.57
• SteelWatch — How much GHG emissions can be cut by new EAFs — Gas-based DRI-EAF ~1.4 tCO2/t, scrap EAF ~0.3. Ratio to BF-BOF gives k ~1.57

k — coal power

• NBER w28949 — Coal-Fired Power Plant Retirements in the U.S. — Coal retirements displaced primarily by increased dispatch from existing gas plants. With coal ~0.95-1.0 and CCGT ~0.40 tCO2/MWh this supports k ~0.42
• WattTime — Marginal Emissions Factors for the U.S. Electricity System — Regional variation in marginal emissions factors; MRO ~70% higher avoided CO2 than WECC

Leakage rates (context for rho*k)

• OECD — Carbon prices, emissions and international trade in sectors at risk of carbon leakage — Leakage offsets ~13% of domestic emission reductions in cement and steel. NOTE: measures marginal response to a carbon PRICE, not plant closure, so it bundles rho and k and includes demand destruction. Useful as a reality check that rho < 1, not as a direct estimate of k
• Springer, Env Econ Policy Studies — Why is carbon leakage for energy-intensive industry hard to find? — Meta-level view: average leakage ~14%, up to ~40% for basic metals and non-metallic minerals (Aichele & Felbermayr)

45Q after OBBBA

• Global CCS Institute — US Preserves and Increases 45Q Credit in OBBBA — $85/t point source, $180/t DAC, EOR/utilisation parity from 4 Jul 2025
• Clifford Chance — OBBBA: Updates to Section 45Q — Legal detail on parity and placed-in-service dates
• Payne Institute — Key Changes for 45Q Under OBBBA — Impact assessment

DOE cancellations

• Utility Dive — DOE cancels $3.7B in carbon capture, decarbonization awards — 24 OCED awards cancelled May 2025; Calpine Sutter and Baytown $540m
• FAS — One Year into Trump II: DOE Programs Stalled, Awards Cancelled — 321 further awards, >$8bn, Oct 2025; 55 carbon-management relevant

Global CCS status

• Global CCS Institute — Global Status of CCS 2025 — 64 Mtpa operating capacity Jul 2025, +25% y/y; 416 Mtpa pipeline

Project pipeline

• IEA CCUS Projects Database — Cross-check on project status and capacity

US outlook

• Petroleum Economist — Outlook 2026: Carbon capture in the US — Air Products Louisiana >5 Mtpa; Nucor-ExxonMobil ~0.8 Mtpa; Gulf Coast hubs 2026-30

US projects

• Akin Gump — Notable US Carbon Capture and Storage Projects — Project-level cross-check

Gas + CCS for data centres

• ExxonMobil / Calpine — CO2 transportation and storage agreement, Baytown — Up to 2 Mtpa; 95% design capture; ~500 MW low-carbon power
• Wood Mackenzie — 100-200 GW of new data centre capacity needed by 2030 — ~$115/MWh including capture; 3-4 year deployment; 450 TWh 2025 demand
• NextEra / ExxonMobil 1.2 GW gas + CCS — 1.2 GW, 90%+ capture, marketed to hyperscalers from Q1 2026

Behind-the-meter gas

• Kroll — Gas Turbines Today: Powering Hyperscaler Data Center Surge — ~101 GW announced behind-the-meter gas; 57 GW with equipment orders; 7 GW under construction

Pulp and paper closures

• ResourceWise — 2025 Sees Continued Pulp and Paper Mill Closures — ~6 Mt North American capacity lost in 2025; ~10% US containerboard capacity cut
• ResourceWise — International Paper Georgetown Mill Closure — ~300,000 t fluff pulp capacity withdrawn
• Domtar Corp Form 10-Q FY2026 (SEC EDGAR) — Coosa Pines indefinite idle May 2026, ~270,000 ADMT market pulp
• Packaging Corp of America Form 8-K FY2025 (SEC EDGAR) — Wallula No.2 machine and kraft pulping shut by Q1 2026, 250,000 t

Pulp and paper CCS

• Clean Air Task Force — Mapping opportunities to decarbonize the U.S. pulp and paper industry — 78 Mt fossil CO2/yr; ~62 Mt/yr biogenic capture technical potential; Microsoft-CO280 3.7 Mt purchase
• RSC Energy & Environmental Science — BECCS in the US pulp and paper industry — Techno-economic basis for mill capture cost range
• Springer — CO2 capture from the Kraft mill limekiln: techno-economic analysis — $91/t for 90% capture from limekiln flue gas

Ethanol CCS

• Inside Climate News — Summit sold its Midwest pipeline as a carbon solution; now it will be used for fossil fuels — Pivot from sequestration to EOR
• Iowa Capital Dispatch — Proposed new route for Summit pipeline — Route restructuring; 27 Iowa plants retained; four dropped
• Carbon Herald — Summit faces $15M lawsuit as pipeline delays continue — Permitting and counterparty risk

Ethanol economics

• farmdoc daily — Trends in the Operational Efficiency of the US Ethanol Industry — Sector operating economics
• farmdoc daily — The Clean Fuel Production Tax Credit (45Z) — Up to $1.00/gal; 50 kg CO2e/mmBTU benchmark; CI-linked value

Gasoline demand

• EIA — Increasing fuel efficiency leads to decreasing gasoline consumption — 143 Bgal (2018) to 135 (2026) to 111 (2035); basis for ethanol demand CAGR

Cement CCS

• Carbon Herald — Heidelberg Advances Carbon Capture Plans At Mitchell Cement Plant — 2 Mtpa design; $500m DOE award cancelled May 2025; test well continues

Power plant rules

• Federal Register — Repeal of GHG Emissions Standards for Fossil Fuel-Fired EGUs — Removal of the CCS-based BSER standard: the regulatory pull for power CCS is gone, leaving only the subsidy push
• Harvard EELP — Regulating GHGs for New and Existing Fossil Fuel-Fired Power Plants — Regulatory timeline

Zombie firm theory

• Caballero, Hosoi and Kashyap — Zombie Lending and Depressed Restructuring in Japan — Foundational treatment of subsidised survival and impaired creative destruction
• Banerjee and Hofmann — The rise of zombie firms: causes and consequences (BIS) — Prevalence and productivity consequences across 14 advanced economies
• Acharya et al. — Zombie Lending and Policy Traps (NBER w29606) — Policy-trap dynamics: why support programmes are hard to withdraw once dependence forms
• Gao — Zombie firms, state subsidies and aggregate productivity (Economica 2025) — Direct treatment of state subsidies rather than bank credit as the survival channel

Policy durability

• Meckling, Kelsey, Biber and Zysman — Winning coalitions for climate policy (Science) — Green industrial policy builds the constituency that later sustains carbon regulation
• Political economy of green industrial policy in liberal states (Energy Policy 2025) — Conditions under which green constituencies outweigh incumbent fossil coalitions
• Nature Climate Change — Modelling the impacts of policy sequencing on energy decarbonization — Sequencing and feedback effects on policy survival

Not advice. Independent analysis published for discussion. Not investment, legal, tax or policy advice, and not a recommendation regarding any security, project or company named.

No affiliation. No commercial relationship with any project, developer, operator or agency named. Companies are discussed solely as public examples.

Model, not measurement. The central quantities are outputs of a model built on stated assumptions, several of which are judgement rather than estimates. Figures should not be cited as measured values.

Point-in-time. Project statuses, capacities and policy parameters reflect information available as of July 2026. Capture pipelines change rapidly and announced capacity routinely fails to materialise.