Carbon Capture at Scale

Two different things share the name. Point-source capture takes CO₂ from a flue gas where it is already concentrated, which is thermodynamically easy and avoids emissions rather than removing them. Direct air capture takes CO₂ from ambient air at roughly 420 parts per million, which is thermodynamically expensive and genuinely removes it. This brief is about removal — carbon dioxide removal in the round, engineered and land-based — but it carries the point-source record anyway, because point-source projects are the only large-scale empirical test of the geological storage half that every removal pathway assumes.

The framing under test is that carbon removal can reach the scale the mitigation scenarios assume. Testing it needs both numbers, and both are available. Established Direct air capture with storage removed approximately 1,500 tonnes of CO₂ worldwide in 2025. The median assumption for it in 1.5 °C pathways is 20 million tonnes a year by 2050 — a factor of roughly 13,000, and against the upper bound of the assessed range a factor above a million. That is the single number this brief exists to state, and it is the least interesting thing in it.

The interesting finding is where the framing actually fails. It is not that direct air capture is expensive. It is that the scenarios' removal is overwhelmingly land, the land is already largely committed, and the engineered methods that could substitute for land deliver one part in a million of what would be required. And beneath all of it sits a constraint that is neither chemical nor financial: removal is a service whose only product is a claim, so measurement, reporting and verification decide whether any of it counts. Solar radiation modification is a different intervention with a different risk profile, treated in Climate Engineering; the two meet only where the moral-hazard argument links them, and it links them asymmetrically.

Established Start with what the scenarios contain, because the gap is only meaningful against a specific assumption. IPCC AR6 Working Group III, Chapter 12, gives median annual deployment by 2050 in 1.5 °C pathways as BECCS 2.75 [0.52–9.45] GtCO₂ a year, land-sector removals, mainly afforestation and reforestation, 2.98 [0.23–6.38] GtCO₂ a year, and direct air capture with storage 0.02 [0–1.74] GtCO₂ a year. Cumulatively over 2020–2100: 328 [168–763] GtCO₂ from BECCS, 252 from the land sector and 29 from direct air capture. Engineered removal is a rounding error in the pathways — which is the first thing most discussion of this subject gets backwards.

Established Now the deployed side, from the field's independent scientific assessment. The third edition of The State of Carbon Dioxide Removal reports conventional removal from afforestation and reforestation averaging 2,200 MtCO₂ a year (range 1,800–2,600) across 2014–2023, and novel removal at approximately 2.0 MtCO₂ a year in 2025, of which biochar is 1.46, BECCS 0.51, biomass burial 0.05, enhanced weathering 0.0038, mineral products 0.0022 and direct air capture with storage 0.0015. Read the last figure again in tonnes: 1,500. Established The assessment's own framing of the policy gap is narrower and more useful than a technology-against-scenario comparison. National pledges reach 2.5 GtCO₂ a year in 2030, 2.7 in 2035 and 3.6 in 2050; the highest-ambition Paris-consistent scenarios need a median of 2.9, 3.9 and 8.8. The removal gap is therefore 0.3 GtCO₂ in 2030, 1.2 in 2035 and 5.2 in 2050 — widening by a factor of seventeen over twenty years.

Established Growth is fast and the discount the field does not apply to itself is large. Novel removal grew at 36% a year over 2020–2025, led by biochar; the assessment's accompanying release states that closing the gap requires removal to scale faster than solar photovoltaics or electric vehicles did. Against that: only about 20% of projected novel removal capacity has historically been delivered, and on projects actually under construction 2030 novel removal reaches roughly 8.4 MtCO₂ a year. The pipeline is not the capacity. Sustained perfectly, 36% compounding from 2 Mt reaches roughly 1 GtCO₂ a year around 2045, which is conceivable; apply either the 20% delivery rate or the market's own delivery record and the curve does not arrive.

Established Direct air capture is where the framing is most often tested and most often mis-stated, in both directions. Climeworks' Orca plant in Iceland, design capacity 4,000 t a year, has captured just over 2,400 tonnes in total since 2021, with a maximum of roughly 1,000 tonnes in a single year — never reaching design. Mammoth, design 36,000 t a year, captured 105 tonnes in its first ten months; on-site reporting in June 2025 put 2024 capture at approximately 100 tonnes, with 12 of 72 collector containers installed and a filter fault identified in second-generation prototypes in 2023 still unresolved and addressed with a temporary fix rather than a delay. The company's own corporate emissions were 1,079 tCO₂ in 2022 and 1,700 in 2023 — a 57% year-on-year increase — against 876 tonnes captured between December 2023 and October 2024, and it estimates that Mammoth could take up to ten years to offset its own lifecycle emissions. Climeworks has raised approximately $800 million and cut roughly a quarter of its workforce in May 2025. Frontier Its plant-level cost is around $1,000 per tonne and its own stated 2030 targets are $250–350 per tonne captured and $400–600 per tonne removed. The leading operator's own 2030 target for a fully removed tonne is four to six times the US Department of Energy's $100 viability threshold — an interested party publishing a number against its own interest, which is the reason to weight it heavily.

Established The affirmative case is under construction and it is real. 1PointFive's Stratos in Ector County, Texas — liquid-solvent capture, design up to 500,000 t a year on a 65-acre site — secured a Class VI injection permit in April 2025, storing in a deep saline formation about four miles away at over 5,000 feet under an EPA-approved monitoring plan, with Microsoft, Amazon, All Nippon Airways and Bain among the buyers. At design capacity this one plant would be roughly 330 times the entire world's 2025 output, which is the correct way to read both its significance and how low the base is. Established The US Department of Energy's regional hubs — Project Cypress in Louisiana at up to $600 million targeting 1 Mt a year by 2030, and the South Texas hub at up to $500 million with potential expansion funding — went through an instructive political cycle: at least 10 of 21 original hub projects terminated in October 2025 with both flagships on a leaked cancellation list, restoration confirmed to the House Appropriations Committee in April 2026, and the $1.2 billion preserved in the FY2027 request. Frontier A technology whose unit economics depend on a subsidy and a tax credit has a deployment curve that is a function of an election, and the 2025 restructuring at the leading operator was attributed partly to exactly that uncertainty.

Frontier Independently assessed costs do not support the marketing, and the projection is circular. The State of CDR assessment puts levelised gross removal by direct air capture at US$400–2,500 per tCO₂, falling to US$100–600 only once deployment reaches gigatonne scale, and notes that gigatonne-scale capture would raise global energy demand by approximately 5–6%. Constrained sustainable potential is 0.5–5 GtCO₂ a year by 2050. Direct air capture becomes affordable at gigatonne scale, and reaching gigatonne scale requires it to be affordable — a projection that assumes its own conclusion, and the field states it without noticing.

Established The storage half is where the best-documented failures are, and they are not the failures people expect. Boundary Dam Unit 3 in Saskatchewan, the world's only coal-fired power CCS retrofit, cost over CAD$1 billion and was designed for 1 Mt a year at a 90% capture rate. From January 2015 to December 2023 it captured 5.8 Mt against 9.2 Mt at design — 57% — and has never made the annual target in nine years, with the captured CO₂ going largely to enhanced oil recovery. Frontier Gorgon in Western Australia is the single most important datum in the storage file, and not because it underperformed. Chevron's LNG project was approved on condition it capture 80% of reservoir CO₂ on a five-year rolling average from July 2016. It achieved 33%, 34% and 30% in the three years to FY2023-24 — the last the lowest on record — and 44% cumulatively, never meeting the target in any year, with cost per tonne rising from an initial $70 to $222 against $3.2 billion of capital. The binding constraint was reservoir behaviour, not capture chemistry, not finance, not politics — under an operator with Exxon and Shell as partners, three of the most capable subsurface engineering organisations that have ever existed. Every gigatonne-scale removal pathway assumes storage is the easy part; Gorgon is the best-instrumented test of that assumption and it failed on the geology.

Frontier The aggregate record, and the counterweight. IEEFA's review of thirteen flagship CCS cases — roughly 55% of total nominal capture capacity operating worldwide, and by its estimate more than two-thirds of all anthropogenic CO₂ ever captured — found ten in operation, two failed and one suspended, and concluded that failed and underperforming projects considerably outnumbered successful ones, with the successes concentrated in natural gas processing serving the fossil fuel industry. Interested party: IEEFA campaigns against fossil finance and the finding runs with its interest; the underlying project-level data are public and the specific claims are checkable, which is why they are carried. Established Against it: Northern Lights, the Equinor, Shell and TotalEnergies joint venture, began operations in 2025 as open-access CO₂ transport and storage under Norway's Longship programme, with EU co-funding for its second phase, and has demonstrated cross-border CO₂ shipping in practice. Dedicated saline storage, sold as a service, working. It is also one project.

Established BECCS is the scenarios' workhorse and nobody is building it. It carries 2.75 GtCO₂ a year in 2050 and 328 GtCO₂ cumulative in the median pathway; actual 2025 deployment is 0.51 MtCO₂ a year, about 0.02% of the 2050 median. Frontier The constraint is land and biomass, and the assessment is blunt: constrained sustainable potential from dedicated biomass plantations is 0.1–0.9 GtCO₂ a year by mid-century, because feedstock competes with bioenergy, chemical-industry demand and agriculture, and AR6 notes that bioenergy is the most land-intensive renewable option. Costs are assessed at US$15–400 per tCO₂. The assessed sustainable potential of the scenarios' largest removal pillar is roughly a third to a tenth of what the scenarios draw from it. Established AR6 flags the problem itself, noting that reaching the higher end of removal volumes raises feasibility issues — rapid upscaling from a small base, underdeveloped verification, and models that under-represent variable renewables and demand-side options and thereby inflate the required removal. That last clause is an assessment body reporting that its own scenario ensemble may overstate the need for the thing being assessed, and the interest runs against the finding.

Established Enhanced rock weathering and ocean alkalinity enhancement are the two methods whose physics is fine and whose measurement is not. Crushed silicate rock spread on farmland weathers with rainwater and CO₂ to bicarbonate, ultimately stored in the ocean; deployed 2025 volume is 0.0038 MtCO₂ a year — 3,800 tonnes globally — at an assessed US$50 to over US$300 per tCO₂, with some lifecycle assessments exceeding US$1,000, and a constrained potential of roughly 0.7 GtCO₂ a year. Frontier Agronomic co-benefits are reported and are reported by interested parties: a Newcastle University trial co-authored with an enhanced-weathering developer found spring oat yields 15% higher on average — 9.3% ploughed, 20.5% direct-drill — soil pH up 0.2–0.29 units, and no negative environmental impacts detected. A developer co-authoring its own favourable yield result is the textbook case for discounting.

Frontier The binding problem is quantification, and the field says so about itself. Clarkson and colleagues' review of enhanced-weathering measurement in Frontiers in Climate — lead-authored from an enhanced-weathering company, with Yale, Heriot-Watt and RMI co-authors — finds that no single method is adequate. Solid-phase mass balance suffers low signal-to-noise against high background soil cations; released cations may bind to the exchangeable phase rather than export, producing false positives; supposedly immobile tracer elements can move under some pH, redox or colloid conditions; gas-phase flux measurement can be dominated by short-term organic carbon cycling; you often cannot tell whether carbonic acid, which captures CO₂, or a strong acid, which does not, drove the weathering; and fertiliser inputs confound the cation budget. Carbon is captured in soil and stored in the ocean, so the system is open and end-to-end verification is structurally hard. The recommendation is a surplus measurement strategy combining approaches — which is an admission that no method closes the loop. A negative verifiability review from a company selling the method is evidence weighted upward, not down. Established The market has not waited: the Frontier buyer coalition committed $27 million for 90,000 tonnes from Terradot, implying about $300 a tonne for 2025–2029 delivery, with Google separately committing 200,000 tonnes into the early 2030s, and Terradot spread 48,000 tonnes of basalt across 1,800 hectares in Brazil in its first year.

Established Ocean alkalinity enhancement has had its first permitted open-water trial and the result is genuinely instructive in both directions. In August 2025 the WHOI-led LOC-NESS project released highly purified sodium hydroxide over a six-hour engineered release in Wilkinson Basin in the Gulf of Maine — the first EPA-permitted trial of its kind, authorised after a year-long review including 75 days of public comment — monitored with shipboard sensors, four autonomous underwater vehicles and gliders, satellite imagery, modelled particle tracks, CTD profiles, plankton tows and microbial, zooplankton, fish-larvae and lobster-larvae sampling. Preliminary results: the experiment successfully and safely produced conditions allowing the surface ocean to take up carbon, a portion of the uptake could be quantified from pH, pCO₂ and tracer data, and no significant biological impacts were detected inside versus outside the patch. Frontier Two things deserve emphasis and they point opposite ways. The environmental-safety finding is a real and under-reported positive. And “a portion could be quantified” is the honest limit: this is the most heavily instrumented marine removal experiment ever run and it could not close the carbon budget. Assessed cost is US$100–150 per tCO₂, possibly above US$300, and US$600 first-of-a-kind; global technical potential is described as poorly constrained and estimated at under 1 to 100 GtCO₂ a year, a two-order-of-magnitude range which is a statement that nobody knows.

Established Ocean iron fertilisation is the field's one worked example of a method being retired, and the reason it was retired is the reason to cover it. Adding iron to iron-limited surface waters reliably produces a phytoplankton bloom; across roughly a dozen mesoscale experiments from the 1990s onward the export flux to depth — the part that constitutes removal — was small, highly variable and hard to measure, with a persistent gap between bloom carbon and sequestered carbon. An unsanctioned commercial release off British Columbia in 2012 converted a scientific dispute into a governance incident, and the regulators moved. Established London Convention and London Protocol parties adopted Resolution LC-LP.1 (2008), establishing that ocean fertilisation other than legitimate scientific research should not be allowed, and LC-LP.2 (2010), an assessment framework for judging what counts as legitimate research; in October 2013 they adopted Resolution LP.4(8), adding Article 6bis, an Annex 4 listing marine geoengineering activities — currently ocean fertilisation only — and an Annex 5 assessment framework. The amendment is still not in force: as of 2025 nine parties have accepted it. Thirteen years, nine ratifications. Frontier The lesson generalises poorly in one direction and precisely in another. It does not generalise as “removal gets banned” — research remains permitted under an assessment framework. It generalises exactly as this: the method whose removal could not be verified was the one that lost its licence. Which is the same constraint now binding enhanced weathering and ocean alkalinity enhancement. The institutional side belongs to Geoengineering Governance; the fact belongs here.

Established And the market has already demonstrated what happens when verification lags the transaction. CDR.fyi, the field's transaction tracker, records 49.5 million tonnes of durable removal sold against 1.6 million tonnes delivered — 3.2% — across 1,222 purchasers and 854 suppliers, with total spend of $12.46 billion, implying roughly $251 a tonne; the largest supplier by volume is a biochar producer with 2.4 Mt sold and 383,446 t delivered. Frontier Two readings and both are correct: an industry with a $12 billion order book that did not exist eight years ago, and an industry where 97% of what has been purchased has not been delivered, with the delivered fraction dominated by the method with the lowest cost, the shortest verification chain and the most contested permanence. Established The precedent for verification failure is the voluntary market's REDD+ episode, and it should be stated with numbers. West and colleagues, in Science in 2023, analysed 26 project sites across three continents, 18 with sufficient public baseline data, and found that approximately 94% of the credits examined did not represent real emission reductions, that only 6.1% could be tied to genuine additional reductions, and that 68% — over 60 million credits — came from projects that did not significantly reduce deforestation at all, against roughly 89 million credits claimed. Frontier Verra, the certifier, disputes the method, arguing that synthetic controls cannot represent pre-project conditions because project areas are selected for acute deforestation risk, making placement non-random, and announcing a shortening of baseline reassessment from ten years to six and consolidation onto a single methodology with jurisdictionally allocated baselines. Interested party: the certifier defending its own product — and the objection is not frivolous, because non-random treatment assignment is a real problem for synthetic controls. The structural lesson survives the objection: a market ran for a decade at scale issuing credits whose central quantity, the counterfactual, was never independently verifiable, and it took outside academic work to establish the size of the error.

Frontier Biochar is the method the discourse least discusses relative to its actual share, and that is the most interesting live fact in the industry. It is roughly three-quarters of 2025 novel removal, the largest delivering supplier is a biochar producer, and it works now and cheaply at an assessed US$70–360 per tCO₂. Its contested quantity is permanence: centuries rather than millennia, and highly dependent on production conditions and the soil environment it ends up in. Frontier Whether permanence tiers are a coherent basis for a market — the claim that a tonne stored for a thousand years is a different product from a tonne in a forest, and that pricing them separately fixes the REDD+ failure mode — is the market's central design bet. The distinction has been adopted; whether it survives an audit is untested.

Frontier Is geological storage in saline formations a solved problem? Essentially all scenario work assumes it is, and the flag should not be established, which is how it is usually treated. For: Sleipner's long record, Northern Lights operating from 2025. Against: Gorgon, where reservoir pressure management constrained the injection system under three of the world's most capable subsurface operators. The question is not whether CO₂ stays put but whether a permitted injection rate can be sustained, and those are different engineering claims that the literature routinely merges.

Speculative Several methods sit in the option space with essentially no evidence, and enumerating them is more useful than pretending the field is only DAC and trees. Artificial upwelling and downwelling is contemplated in the London Convention's fertilisation scenarios as deep-water upwelling using physical devices; the sign is not obvious, because upwelling brings nutrients and dissolved carbon to the surface, and it is under consideration for listing. Ocean biomass sinking — kelp cultivation and sinking, crop-residue baling and sinking — has a handful of commercial attempts and is unresolved on both permanence and benthic impact, and is likewise under consideration. Frontier Mineralisation into building materials, at 0.0022 MtCO₂ a year, is real chemistry at small volume with generally good permanence. Biomass burial and bio-oil injection, at 0.05 MtCO₂ a year, is simple and cheap and its permanence depends entirely on the burial environment remaining anoxic. Handwave Soil organic carbon sequestration has a large claimed potential, notorious reversibility and measurement difficulty, and is not counted as novel removal at all — the flux is frontier, the durable crediting of it is not.

Frontier Is the scenarios' removal requirement itself an artefact? This is the most interesting live hypothesis in the slot and it comes from the assessment body. AR6 WG3's own authors note that insufficient representation of variable renewables and limited demand-side options in integrated assessment models inflate removal requirements. If that is right, the gap closes not because removal scales but because less of it turns out to be needed. It is an assessment body reporting a bias in its own instrument, which is the shape of finding that deserves weighting up and almost never gets cited.

Frontier Is fossil-industry capital disqualifying, or the only route to scale? Both positions are held by serious people. Occidental owns the largest direct air capture project; enhanced oil recovery is the destination for Boundary Dam's CO₂; and the subsurface skills that storage requires sit inside the same firms. This is reportable as a dispute rather than resolvable, and it is a fact readers should weigh rather than a disqualification. Frontier The removal version of moral hazard is structurally the same argument as the one in Climate Engineering, with one asymmetry that matters: unlike solar radiation modification, where the individual-attitudes evidence mostly comes back null, the removal version has a documented case — offset markets that demonstrably supported continued emission against credits which were about 94% non-additional.

Speculative And one pointer this brief records without leaning on it. A 2025 paper in Environmental Science & Technology titled Three Years of Field Trials Indicate a Sustained Enhanced Rock Weathering Signal with Limited CO₂ Removal exists. Its full text was not obtained and no claim about its contents is made here. The title alone is recorded because a result of that shape from a three-year field programme is the kind of finding that moves a method's flag, and a reader tracking enhanced weathering should know it is there.

Established The first bottleneck is thermodynamic and no engineering removes it. Separating a gas present at 420 parts per million costs work, and that floor is a property of the atmosphere rather than of the technology. It is why gigatonne-scale direct air capture would raise global energy demand by approximately 5–6%, and why the cost curve for capture from air will never converge on the cost curve for capture from a flue gas.

Frontier The second is measurement, and it is the binding one for the industry as a whole. Removal is a service whose only product is a claim, so a tonne that cannot be verified is not a tonne. Enhanced weathering's own measurement review says no single method is adequate and the carbonic-versus-strong-acid attribution is often unresolvable; the best-instrumented ocean alkalinity trial ever run quantified “a portion”; ocean iron fertilisation lost its licence on exactly this. The physical bottleneck for direct air capture is thermodynamics; the system bottleneck for carbon removal is verification.

Frontier The third is storage performance under a permit rather than storage security in principle. Gorgon achieved 44% against an 80% licence condition on reservoir grounds, and the record on point-source capture generally — ten operating, two failed, one suspended across thirteen flagship cases — is a worse base rate than any removal pathway assumes for its own storage leg.

Established The fourth is land, and it binds the largest pathway hardest. BECCS's constrained sustainable potential from dedicated plantations is 0.1–0.9 GtCO₂ a year against a scenario median of 2.75, because feedstock competes with bioenergy, chemicals and food. Conventional land-based removal is already running at 2.2 GtCO₂ a year, near the top of what land can be asked to do while also feeding people and holding biodiversity — a constraint Future Agriculture and Biodiversity Restoration approach from the other side.

Frontier The fifth is capital and policy risk, which behave here as an engineering constraint rather than a financial one. Ten of twenty-one DOE hub projects were terminated in October 2025 and the flagships restored in April 2026; the leading operator cut a quarter of its workforce in the interval. A plant designed on a fifteen-year subsidy horizon inside a four-year political cycle is a plant with a bottleneck in its balance sheet, and none of the scenario work models it.

Established And the sixth is delivery itself, which nobody discounts for. The announced-to-delivered ratio for novel removal is about 20%; the purchased-to-delivered ratio in the durable market is 3.2%; Boundary Dam runs at 57% of design over nine years; Gorgon at 44% against its obligation; Mammoth at roughly 0.3% of design in its first ten months. The ratio of announced to delivered runs between 3% and 57% depending on where you cut it, and no scenario applies such a discount.

Established Nothing on this map produces the result this brief waits on. Atmospheric Management owns methane, nitrous oxide and the ozone regime — the boundary is the molecule — and its finding that ozone proved tractable because the substances had identifiable producers, drop-in substitutes and legible harm is invoked here by reference and not restated. Ocean Engineering owns the ocean as an industrial surface and hands the ocean-as-climate-instrument questions here. Planetary Stewardship owns the boundaries accounting. None of them produces a number that changes this verdict.

Frontier What this slot waits on is three facts about the world, recorded as typed requirements below, and they are not equally hard. A verification standard a buyer and a regulator both accept is the binding one and is currently absent for the two fastest-growing contracted pathways. A price that clears is a market fact: the implied average across the durable market is about $251 a tonne against independently assessed direct air capture costs of $400–2,500, so the transactions clearing today are clearing on the cheap methods and the expensive ones are being bought as demonstration. Capacity is the third and the most conventional — plants, spreaders, ships, wells — and it is the one thing on the list that money reliably buys.

Established One dependency runs the other way and deserves stating. The scenarios this brief measures against may themselves be the thing that changes: AR6's own authors record that their models inflate the removal requirement by under-representing renewables and demand-side measures. The most credible route to closing the gap in the published literature is not that removal scales; it is that less removal turns out to be needed. That is a dependency on a modelling result, and it is being produced by the same community that produced the requirement.

Established First, a direct air capture plant that runs a full year at design capacity. None ever has. Orca has never reached 4,000 t a year in five years; Mammoth managed roughly 105 tonnes against 36,000 in ten months. Stratos in 2026–27 is the test, and the number to watch is delivered tonnes in a calendar year rather than nameplate.

Frontier Second, a closed carbon budget for an enhanced-weathering field deployment — one site where solid-phase, aqueous-phase and downstream measurements agree within a stated uncertainty. The field's own measurement review says no method does this alone and recommends combining them; nobody has published the triangulation at commercial scale. Until somebody does, the tonnes being contracted at roughly $300 each are a modelled quantity.

Frontier Third, a multi-year ocean alkalinity trial with a quantified rather than partial uptake fraction. LOC-NESS is the state of the art, it was safe, and it quantified a portion. The follow-on question is whether a longer release with a larger tracer budget can close the fraction, and whether the safety result holds at a scale where the alkalinity patch is not swiftly diluted.

Frontier Fourth, a geological storage complex operating at its permitted injection rate for five consecutive years. Gorgon is the counter-case and Northern Lights is the test. This is the single experiment that would most change the confidence interval on every removal pathway, because every one of them ends in a well.

Established Fifth, an independent audit of delivered against contracted tonnes across the durable market with a standard permanence definition. CDR.fyi tracks the transactions; nobody audits the tonnes. The REDD+ finding was produced by outside academics rather than by the market's own machinery, and there is no reason to expect the durable market to behave differently. Frontier Sixth, a BECCS lifecycle assessment at commercial scale with an explicit counterfactual land use. The land constraint is the scenarios' weakest joint and the empirical base is thin enough that the assessed potential range spans an order of magnitude.

Established Direct air capture splits into two engineering families with different failure modes. Solid sorbent, as at Orca and Mammoth, cycles a filter material and is limited by sorbent durability and cycle time — the second-generation filter fault identified in 2023 and addressed with a temporary fix is the concrete version of that risk. Liquid solvent, as at Stratos, uses a potassium hydroxide loop with a calcination step and is limited by high-temperature heat and by plant scale-up. Both terminate in the same place: compression, transport, and a Class VI injection well with an approved monitoring plan.

Frontier The energy requirement is the design constraint that scales worst. Gigatonne-scale direct air capture would raise global energy demand by 5–6%, which makes the siting problem a power-system problem — Iceland's geothermal in one case, Texan grid and gas in another — and links this brief to Hydrogen Economies and Industrial Ecology, where the same low-carbon energy is being claimed for other uses.

Established The storage engineering is the mature part on paper and the weakest part in the record. Stratos stores in a deep saline formation about four miles away at over 5,000 feet under an EPA-approved plan; Northern Lights runs open-access shipping and injection as a service; and Gorgon, with the most capable operators available, was constrained by reservoir pressure management. Permitted injection rate, not containment, is the engineering variable that has actually failed.

Frontier The land-based methods are logistics rather than chemistry. Enhanced weathering is a crushing, haulage and spreading operation — Terradot moved 48,000 tonnes of basalt across 1,800 hectares in a year — whose energy and transport footprint is a first-order term in the lifecycle assessment, which is why some assessments exceed $1,000 a tonne. Ocean alkalinity is a chemical production, shipping and controlled-release problem with an instrumentation package attached, and the instrumentation is the expensive half.

Frontier And the engineering requirement that binds hardest is not a plant at all. A verification system a regulator and a buyer both accept implies instrumented field sites, agreed uncertainty budgets, tracer chemistry, and a registry that can retire a claim. That is an engineering programme with no owner and no capital line, and it is the thing standing between a $12 billion order book and 1.6 million delivered tonnes.

Within this map: Climate Engineering, the other intervention — a different risk profile, a different timescale, and the one place the two arguments genuinely join is moral hazard, where the removal version has a documented case and the solar version does not; Geoengineering Governance, which owns the London Protocol machinery this brief cites as a fact rather than as an institution; Atmospheric Management, whose boundary with this slot is the molecule and whose methane-removal roadmap is structurally the same argument at an earlier maturity; Ocean Engineering, which keeps the ocean-as-industrial-surface topics; Planetary Stewardship, whose 2025 accounting makes ocean acidification the seventh transgressed boundary — relevant here for exactly one reason, that carbon removal is the only intervention in this category that acts on acidification, because reflecting sunlight does not; Future Agriculture and Biodiversity Restoration, which own the land this brief's largest pathway is competing for; and Industrial Ecology and Hydrogen Economies, which compete for the same clean energy.

One framing error is worth borrowing rather than restating: Planetary Cooling Concepts names the pattern of a building technology mis-sold as a planetary one, and it has an exact analogue here in direct air capture being priced as a climate solution while removing 1,500 tonnes a year. This brief supersedes the carbon programme's treatment as this slot's coverage; that page stays as the programme's shorter entry point. Outside the map: separation thermodynamics; reservoir engineering and subsurface geomechanics; soil geochemistry and weathering kinetics; marine carbonate chemistry; and the economics of environmental markets, which supplies the REDD+ literature.

Frontier The institutional requirement is a verification standard, and it does not exist for the methods growing fastest. A removal claim has to satisfy two different audiences with different tolerances: a buyer, who needs a defensible corporate claim, and a regulator, who needs an enforceable one. Direct air capture with geological storage comes closest, because the EPA's Class VI framework supplies a monitoring plan and a physically metered input. Enhanced weathering and ocean alkalinity have neither — open systems, contested attribution, and no agreed uncertainty budget — and they are the two pathways attracting the largest advance purchases.

Established The one institution that has actually acted on a removal method acted on verifiability. The London Convention and Protocol did not ban ocean fertilisation because blooms are dangerous; it required that activity other than legitimate scientific research not be allowed and built an assessment framework for deciding what qualifies. Thirteen years later the treaty amendment carrying that machinery has nine acceptances and is not in force, which is a measurement of institutional speed rather than of intent. The forum question belongs to Geoengineering Governance.

Established Many of this brief's sources are interested and the direction matters in each case. IEEFA campaigns against fossil finance and its underperformance findings run with that interest, though the project-level data are public. 1PointFive describes its own asset. Northern Lights is the operator. Frontier is a buyer coalition announcing its own purchase. The Newcastle yield trial is co-authored with a developer. Frontier Three sources run against their own interest and are weighted up accordingly: AR6 WG3 reporting that its models may inflate the removal requirement; the enhanced-weathering measurement review, lead-authored from an enhanced-weathering company, concluding that no method is adequate; and Climeworks publishing a 2030 removed-tonne target four to six times the policy threshold its own sector cites.

Frontier And the political-risk institution is the tax credit. The economics of every US project in this brief rest on a subsidy and a credit that a change of administration can suspend and restore inside eighteen months, as happened between October 2025 and April 2026. An industry whose unit economics are set by an appropriations line is an industry with a governance dependency it does not describe as one.

Frontier The mitigation-deterrence objection is serious here in a way it is not elsewhere, because it has a case study. Offset markets operated at scale for a decade against credits that outside analysis found approximately 94% non-additional, and those credits were used to justify continued emission. That is not an analogy for the durable removal market; it is the same market's immediately preceding generation, with many of the same buyers. Frontier The counter-argument is that engineered removal has a physically measurable input and so cannot fail the same way. That is true of direct air capture and false of the two pathways growing fastest, whose own literature says the quantity cannot yet be closed.

Established The land question is a distributional question wearing a technical one's clothes. The scenarios draw 2.98 GtCO₂ a year from afforestation and reforestation and 2.75 from BECCS, both land-based, against 0.02 from direct air capture. Land at that scale is somebody's land. The assessed sustainable potential of dedicated biomass plantations is 0.1–0.9 GtCO₂ a year precisely because feedstock competes with food and with biodiversity, and the shortfall between assumption and potential is the space in which displacement happens.

Established The ocean cases show what a consent process looks like when it is actually run. The LOC-NESS trial went through a year-long EPA review including 75 days of public comment before a six-hour release of sodium hydroxide, and reported no significant biological impact. That is a workable model, and it is worth noting how much slower and more expensive it is than the alternative of not asking. Frontier The unsanctioned 2012 iron release off British Columbia is the counter-example and it cost the method its licence.

Frontier Fossil capital is the ethical dispute this brief declines to settle. The largest direct air capture project is owned by an oil company; the destination for the best-known point-source project's CO₂ is enhanced oil recovery; the subsurface competence storage requires lives inside the industry whose product created the problem. Both positions — that this is disqualifying, and that it is the only route to scale — are held by people who understand the technology. It is a fact readers should weigh for themselves rather than a disqualification.

Handwave And the claim that no serious source in this brief supports: that removal can substitute for emissions reduction. Every source here says otherwise, including the sellers — the LOC-NESS researchers state explicitly that marine removal does not replace the need for reduced emissions. At 2 MtCO₂ a year of novel removal against roughly 37 GtCO₂ a year of emissions, the arithmetic does not need an argument.

Established The framing fails, and it fails in an unexpected place. The scale problem is not that direct air capture is too expensive. It is that the scenarios' removal is overwhelmingly land — 2.98 GtCO₂ a year from afforestation and reforestation and 2.75 from BECCS against 0.02 from direct air capture — that the assessed sustainable potential of BECCS from dedicated plantations is three to thirty times below what the scenarios draw from it, and that conventional land-based removal at 2.2 GtCO₂ a year is already near the top of what land can be asked to do. The industry the framing invites us to assess is, at present, not a material contributor to the quantity in question, and its own assessment body says so.

Frontier The cut is completed, unusually, by the assessment community itself. AR6 WG3 records that its models likely inflate the removal requirement by under-representing variable renewables and demand-side measures. So the most credible route to closing the gap in the published literature is not that removal scales — it is that less removal turns out to be needed. A civilisation that mis-specified how much of a technology it required, and then measured its progress against the mis-specification, is making a legible and correctable error, and this is the rare case where the correction is already in the record.

Established The durable finding, independent of any scenario, is about the ratio of announcement to delivery. Novel removal delivers about 20% of projected capacity. The durable market has delivered 3.2% of what it sold. Boundary Dam runs at 57% of design over nine years; Gorgon at 44% against a licence condition; Mammoth at a fraction of a per cent of nameplate in its first ten months. Between 3% and 57%, depending where you cut it, and no scenario applies such a discount to anything. That is a general lesson about how this civilisation forecasts its own industrial capacity, and it is not confined to carbon.

Speculative And the structural reason this is hard is worth stating in the general form. The two completed planetary interventions in the record worked through identifiable producers with substitutes. Carbon removal has no producer to regulate — only a buyer to verify. That inverts the institutional problem: instead of finding the small number of firms making a harmful substance and giving them an alternative, it requires establishing that an absence happened, everywhere, at a stated uncertainty, for a thousand years. Handwave Whether an economy can be built on verifying absences at planetary scale is not a question the historical record answers, and asserting either way is assertion.

These horizons track plant commissioning, contract delivery dates and assessment cycles rather than technology readiness:

• 10 yr: Established The near-term test is delivery, not announcement: whether Stratos runs a full year at anything approaching 500,000 tonnes, whether Northern Lights sustains its permitted injection rate, and whether the durable market's 3.2% delivered fraction moves. Frontier Expect novel removal to grow fast in percentage terms and remain small in absolute ones — 8.4 MtCO₂ a year by 2030 from projects actually under construction, against a 2030 pledge level of 2.5 GtCO₂, almost all of it land. Frontier Expect the verification question for enhanced weathering to be settled or to become a scandal inside this window, because contracted volumes are already running ahead of the measurement literature.
• 25 yr: Speculative If the 36% growth rate held perfectly it would put novel removal near 1 GtCO₂ a year around 2045, and no removal technology has yet held a growth rate through the transition from demonstration to commodity. Speculative The more likely division is that biochar and mineralisation become ordinary industries at moderate volume while direct air capture stays a subsidised demonstration sector unless a verification-plus-price combination arrives. Frontier The 2050 gap of 5.2 GtCO₂ a year either shrinks because the scenarios are re-specified or it does not close at all.
• 50 yr: Speculative At this horizon storage integrity becomes an empirical rather than modelled question for the first time, because the earliest large injections will have decades of monitoring behind them. Speculative Land-based removal's ceiling is likely to be set by food and biodiversity rather than by carbon accounting, and the negotiation over that ceiling is a political process with no current forum. Handwave Anything more specific about which pathway dominates is a guess dressed as a projection.
• 100 / 250+ yr: Handwave Beyond useful forecasting for the technologies, but not for the physics: a tonne stored geologically is intended to stay for this horizon and beyond, and the permanence claims being sold today are claims about it. Handwave No institution in the record has verified anything for a thousand years, and the market that prices thousand-year permanence has existed for under a decade.

• Depends on Nothing on this map. This brief waits on no result another brief produces. Atmospheric Management holds the other molecules, Ocean Engineering holds the ocean as an industrial surface, and Planetary Stewardship holds the boundaries accounting; none produces a number that changes this verdict. The one dependency that runs the other way is a modelling result rather than a brief: AR6's own authors record that their scenarios may inflate the removal requirement. No typed depends-on edge is claimed.
• Requires (not on this map) Three things, none of them a brief on this map. First, a verification standard a buyer and a regulator both accept. This is the binding one. Enhanced weathering's own measurement review, lead-authored from a company selling the method, finds no single technique adequate and reports that the carbonic-versus-strong-acid attribution is often unresolvable; the best-instrumented ocean alkalinity trial ever run quantified only “a portion” of its uptake; and the one method the international system has actually restricted, ocean iron fertilisation, lost its licence because its removal could not be verified. The precedent for what happens without such a standard is not hypothetical: outside analysis of 26 REDD+ project sites found approximately 94% of the credits examined did not represent real emission reductions, and it took academics rather than the market to establish it. Second, a price that clears for a verified durable tonne. The durable market has transacted $12.46 billion for 49.5 million tonnes sold, implying roughly $251 a tonne, while independent assessment puts direct air capture at $400–2,500 and the leading operator's own 2030 target at $400–600 removed — so today's clearing price clears only the cheap methods, and 3.2% of what has been sold has been delivered. Third, industrial capacity three orders of magnitude above today's: 2.0 MtCO₂ a year of novel removal against a 2050 gap of 5.2 GtCO₂ a year, with historically only about 20% of projected capacity delivered and 8.4 MtCO₂ a year in 2030 from projects actually under construction. Capacity is the only one of the three that money reliably buys, and it is the one the discourse spends most of its attention on.
• Enables Every overshoot pathway and every net-zero commitment that assumes residual emissions are neutralised rests on this. No typed enabling edge is claimed, and the reason is the arithmetic: at 2 MtCO₂ a year of novel removal, nothing on this map can honestly rest a result on it yet. The one place removal is genuinely load-bearing rather than aspirational is ocean acidification, where it is the only acting intervention in this category — reflecting sunlight does nothing for it.
• Adjacent Separation thermodynamics; reservoir engineering and subsurface geomechanics; soil geochemistry and weathering kinetics; marine carbonate chemistry; the economics of environmental markets, which supplies the offset-integrity literature; and within this map Climate Engineering, Geoengineering Governance, Atmospheric Management, Future Agriculture and Industrial Ecology.

Handwave “Direct air capture is at $100 a tonne, or has a demonstrated path to it.” Independent assessment puts it at US$400–2,500 per tonne today and $100–600 only at gigatonne scale — a projection that assumes the scale it is meant to justify. Frontier The strongest evidence against the $100 figure is published by the party with the most to gain from it: the leading operator's own 2030 target is $250–350 captured and $400–600 removed. When an interested party's own number is worse than the policy target, that number is the one to use.

Established “Design capacity is capacity.” No direct air capture plant has ever operated at design capacity. Orca has captured just over 2,400 tonnes in total since 2021 against a 4,000 t a year design; Mammoth captured about 105 tonnes in ten months against 36,000, with 12 of 72 collector containers installed and an unresolved second-generation filter fault. Reporting nameplate as output is the single commonest error in this subject, and it is made by supporters and critics alike — the critics because a large nameplate makes the scale gap look smaller than it is.

Frontier “Geological storage is a solved engineering problem.” Gorgon says otherwise: 44% cumulatively against an 80% licence condition, on reservoir pressure grounds, under Chevron with Exxon and Shell as partners. Sleipner and Northern Lights are the affirmative cases and Northern Lights began operating only in 2025. Containment and sustained permitted injection rate are different claims and the literature merges them.

Frontier “Enhanced weathering's removal can be verified today.” Its own review says no single method is adequate: low signal-to-noise against background soil cations, false positives from exchangeable-phase retention, mobile “immobile” tracers, gas-flux measurements dominated by short-term organic carbon cycling, fertiliser confounding, and frequently no way to tell whether carbonic acid or a strong acid drove the reaction. Meanwhile 90,000 tonnes have been contracted at about $300 each. Established “Ocean alkalinity enhancement's uptake has been quantified in the field.” The best trial ever run quantified a portion, and said so plainly. Its safety result — no significant biological impact detected — is the finding that deserves more attention than it gets.

Handwave “Ocean iron fertilisation removes carbon at a verifiable, creditable rate.” It does not, and that is why it is the one method the international system has moved against. The blooms are real; the export flux to depth is small, variable and hard to measure. Established “It was banned.” It was not: LC-LP.1 (2008) provides that activity other than legitimate scientific research should not be allowed, an assessment framework decides what qualifies, and the 2013 treaty amendment carrying the binding machinery has nine acceptances and is not in force. A resolution that is universally described as a ban and a treaty amendment that binds nine countries are two different things.

Frontier “Contracted removal is delivered removal.” 49.5 million tonnes sold, 1.6 million delivered — 3.2%. Established “Announced capacity is a forecast.” Historically about 20% of projected novel removal capacity has been delivered. Frontier “Biochar permanence is equivalent to geological storage.” Biochar is the largest delivering pathway and its permanence is condition-dependent, on the scale of centuries rather than millennia. It is also the method the discourse discusses least relative to its share, which is an odd distribution of attention for the thing actually doing the work.

Frontier “The REDD+ finding transfers directly to engineered removal.” It does not. The failure there was counterfactual baselines — establishing what deforestation would have happened — and engineered removal has a physically measurable input. What transfers is the structural lesson: a market can run for a decade at scale on a quantity nobody independently verified. Verra's methodological rebuttal, that project areas are selected for acute deforestation risk and so synthetic controls mis-specify the baseline, is a real objection from an interested party, and it does not rescue the structural point.

Handwave And the framing claim itself: “removal can substitute for emissions reduction.” No credible pathway supports it and every source in this brief says so, including the sellers. Established The related and more defensible error is to read the scenarios as a technology forecast. They are not: they are what a model needed to balance, and their own authors record that under-representing renewables and demand-side options inflates the removal they call for. Reading a scenario's removal line as a target to be built rather than as an artefact to be examined is how a rounding error became an industry's business plan.