One tonne is a system boundary
The numerator is total cost. The denominator is net atmospheric CO₂-equivalent durably stored, not gas leaving a contactor.
A credible ledger includes plant construction, sorbent manufacture and replacement, electricity, heat, water, compression, transport, injection, monitoring, and eventual closure. A fossil-powered plant can still capture more than it emits, but its gross-to-net ratio worsens and clean-energy demand grows with every tonne.
The target has four simultaneous tests
Cost, net negativity, scale, and durability must coexist. Meeting any one alone does not produce the promised commodity.
The feedstock is everywhere and almost empty
At roughly 420 ppm, a cubic metre of air contains less than a gram of CO₂. The plant must expose an enormous air volume to chemistry without spending the product's value on pressure drop.
For an idealized separation at ambient temperature, the concentration term alone is on the order of 120 kWh of reversible work per tonne. That is a floor, not a machine specification: it excludes pressure drop, imperfect selectivity, heat transfer, sorbent regeneration, compression, and entropy created by real processes.
A contactor is a coupled chemical factory
A better sorbent can lower regeneration energy but bind more slowly, degrade faster, dislike humidity, or require a larger contactor. The optimum belongs to the whole plant.
Contact the air
Move or expose enough ambient air to a reactive surface while limiting fan pressure, land, fouling, and weather sensitivity.
- Layer
- Volume
- System metric
- Net tonnes durably stored per dollar and per year
Bind selectively
Capture CO₂ at roughly 420 parts per million while water, oxygen, and contaminants compete for sites and shorten sorbent life.
- Layer
- Material
- System metric
- Net tonnes durably stored per dollar and per year
Regenerate
Reverse the binding with heat, vacuum, humidity, electricity, or chemical work without destroying the capture medium.
- Layer
- Energy
- System metric
- Net tonnes durably stored per dollar and per year
Condition
Dry and compress a diffuse product into a specification that pipelines, ships, wells, or mineral processes can accept.
- Layer
- Compression
- System metric
- Net tonnes durably stored per dollar and per year
Transport and inject
Match a capture plant to pore space, wells, pressure management, permits, liability, and a route between them.
- Layer
- Infrastructure
- System metric
- Net tonnes durably stored per dollar and per year
Measure and verify
Prove net removal after lifecycle emissions and monitor storage strongly enough that a buyer can own a durable claim.
- Layer
- Product
- System metric
- Net tonnes durably stored per dollar and per year
Liquid-solvent systems can exploit large process equipment and high-temperature regeneration. Solid-sorbent systems can use lower-temperature heat and modular contactors. Electrochemical and moisture-swing routes rearrange the energy bill. None escapes the need to contact dilute air, release concentrated CO₂, and keep the capture medium productive across many cycles.
The capture plant ends where infrastructure begins
CO₂ is not removed when it leaves the regeneration vessel. It must reach a reservoir or durable mineral product with a chain of custody.
Compression is continuous load
Pipeline and injection specifications require conditioning and high pressure, adding equipment and electricity after capture.
Pore space is local
Capacity on a geological map is not a permitted, characterized well with pressure management and an operator.
Networks need anchor volume
A small capture plant cannot cheaply carry a dedicated pipeline, while shared hubs need coordinated schedules and liability.
MRV creates the commodity
Sampling, metering, lifecycle models, subsurface monitoring, and registries make a durable tonne distinguishable from a claim.
Utilization is not automatically storage. Fuel returns carbon quickly; some concrete and mineral routes retain it much longer. The accounting question is not whether CO₂ became a product but how much stayed out of the atmosphere, for how long, compared with the counterfactual.
$100 is a procurement specification
The target only matters if it means a net, durable, verified tonne at useful scale. Quoted capture cost, credit price, subsidy value, and full delivery cost are not interchangeable.
| Number | What it can prove | What it cannot |
|---|---|---|
| Modeled $/t | Design sensitivity and improvement priorities | Bankable delivered cost |
| Credit sale price | What one buyer accepted under one contract | Plant cost or broad demand |
| Nameplate t/y | Designed throughput | Uptime, net removal, or storage |
| Captured tonnes | Contactor output | Lifecycle emissions and durability |
| Verified net stored | The product the climate ledger needs | Whether it can scale cheaply |
The observable cost curve runs through contracts, not lab benchmarks. Early Frontier advanced-market commitments settled at $990/t for Heirloom and an average of $440/t for CarbonCapture, stepping from about $650/t toward $350/t by 2028 (Measured contract prices; the step is a Projected schedule). The largest Frontier and Stripe offtake of 2026 cleared at $315/t (Measured). Stratos, the 500 kt/yr 1PointFive plant, is estimated at $400–500/t with long-term credit pricing of $300–450/t, and Climeworks quotes $400–600/t credits (Projected). In short: $650 in early contracts, $440 on the CarbonCapture commitment average, $315 on the largest 2026 offtake, against an institutional floor of roughly $200–250/t by the early 2030s (Projected). The $100/t headline remains a Target that analyses call highly aspirational with current chemistries.
Peer-reviewed models of a 1 Gt/yr industry put liquid-solvent DAC at $341/t (range $226–544) and solid sorbent near $374/t (Derived/Projected). IPCC AR6 finds $100–300/t likely (Projected), and Global CCS Institute scenarios span $137–412/t at a 5% discount rate (Projected). Energy is the physical term beneath those models: first-of-a-kind designs consume 1.5–2.5 MWh/t against an ideal concentration-only minimum near 0.12 MWh/t (or about 0.17 MWh/t when idealized compression and other boundary terms are included), about an order-of-magnitude engineering multiplier (Derived). Closing that multiplier, not a new chemistry alone, is what moves delivered cost toward the modelled range.
Contracted tonnes are not delivered tonnes
Climeworks reports capacity factors of 40–50% in upgraded Mammoth units and a 50% opex reduction after its September 2026 sorbent and mechanical upgrades, with more than double capture in the upgraded containers (Company-reported). Delivery, not contracting, is the curve that has to move: at 0.05% of the contracted DAC volume delivered by mid-2025, the market had priced far more tonnes than it had received. Mammoth’s reported 675 net tonnes in H1 2026 are 3.75% of half its 36,000 t/yr design capacity; this is a six-month output-to-nameplate comparison, not a lifetime capacity factor.
Who is building what
Direct air capture, biomass burial, ocean alkalinity enhancement, and rock weathering operate under fundamentally different energy and permanence budgets. Search the record, or filter by removal mechanism.
ClimeworksMammoth & OrcaSolid amine sorbent filter collectors desorbed at ~100°C with geothermal heat, coupled to Carbfix basalt mineralization
- Reported evidence
- Orca operates at ~4,000 t/yr nominal capacity; Mammoth initiated operations in 2024 with up to 36,000 t/yr design capacity in Hellisheiði, Iceland.
- Announced next step
- Megatonne-scale regional capture hubs and driving capture cost down toward $300/t.
- Unresolved risk
- Desorption thermal energy demand (~1.5–2.5 MWh/t), sorbent degradation rates, and basalt reservoir permeability.
Heirloom CarbonLimestone looping & kilnCyclic carbonation of slaked lime (Ca(OH)₂) on vertical stacking contactors, coupled to electric calcination and concrete mineralization
- Reported evidence
- Operational commercial facility in Tracy, California capturing ~1,000 t/yr with verified third-party credit issuance to corporate buyers.
- Announced next step
- Deploying commercial facilities through the DOE Project Cypress regional DAC hub in Louisiana.
- Unresolved risk
- Contacting kinetics requiring several days per carbonation cycle, large surface land footprints, and electric calcination capital intensity.
1PointFive / Carbon EngineeringStratos plantLiquid potassium hydroxide air contactors coupled to calcium carbonate pellet reactors and gas-fired oxy-calcination with CCS
- Reported evidence
- Construction advanced on Stratos in Ector County, Texas, engineered for 500,000 t/yr gross CO₂ capture capacity.
- Announced next step
- Commercial startup of Stratos followed by multiple standardized 1 Mt/yr industrial DAC hubs.
- Unresolved risk
- Complex chemical plant balance-of-plant, water loss in arid environments, and parasitic power and natural gas supply requirements.
Charm IndustrialFast pyrolysis & bio-oil injectionFast pyrolysis of agricultural residues into liquid bio-oil, pumped down EPA Class I/V industrial disposal wells into deep geologic formations
- Reported evidence
- Delivered and audited tens of thousands of tonnes of permanent carbon removal to Frontier and voluntary buyers since 2021.
- Announced next step
- Expanding mobile and modular pyrolyzer fleets to reach multi-hundred-thousand-tonne annual processing rates.
- Unresolved risk
- Bio-oil acidity (pH 2–3) and corrosiveness during transit and injection, biomass collection logistics, and feedstock seasonality.
GraphyteCarbon CastingLow-temperature drying and mechanical compression of biomass waste into dense blocks, sealed in impermeable barriers and buried in monitored landfills
- Reported evidence
- Inaugurated the Loblolly production facility in Arkansas with commercial offtake contracts from American Airlines and Frontier.
- Announced next step
- Sub-$100/t removal cost through ultra-low process energy (~10% of energy required by DAC or pyrolysis).
- Unresolved risk
- Long-term barrier integrity against subterranean microbial degradation and moisture ingress over thousand-year horizons.
UndoEnhanced Rock WeatheringSpreading finely crushed volcanic basalt on agricultural cropland to accelerate chemical weathering and bicarbonate runoff to oceans
- Reported evidence
- Over 200,000 tonnes of basalt spread across UK, US, and Australian farms with empirical soil pore-water and cation flux tracking.
- Announced next step
- Million-tonne annual carbon removal with agricultural co-benefits (soil pH balancing and nutrient release).
- Unresolved risk
- Quantifying field dissolution rates in heterogeneous soils, trace heavy metal accumulation (nickel/chromium), and transport emissions.
EquaticSeawater electrolysis & mineralizationElectrolysis of seawater passing an electric current to split water, trap dissolved CO₂ as solid calcium and magnesium carbonates, and co-produce green H₂
- Reported evidence
- Pilots operated in Los Angeles and Singapore; broke ground on Equatic-1 in Tuas, Singapore, targeting ~3,650 t/yr CO₂ removal.
- Announced next step
- Co-production economics delivering sub-$100/t net CDR subsidized by green hydrogen sales.
- Unresolved risk
- Electrode fouling from mineral scaling, ocean water discharge permitting, and high electricity demand per net tonne.
Ebb CarbonElectrochemical alkalinity enhancementBipolar membrane electrodialysis removing hydrochloric acid from seawater to increase ocean alkalinity and enhance atmospheric CO₂ drawdown
- Reported evidence
- Commissioned marine pilot at Pacific Northwest National Laboratory (PNNL) Sequim Marine Laboratory processing coastal seawater.
- Announced next step
- Commercial deployment alongside coastal desalination outfalls and industrial seawater-cooled power facilities.
- Unresolved risk
- Membrane lifespan in raw seawater, ecological impacts of local alkaline plumes, and verifying marine air-sea CO₂ equilibration kinetics.
Frontier ClimateAdvance Market CommitmentAggregated pre-purchase commitments ($1B+ fund backed by Stripe, Alphabet, Shopify, Meta, and McKinsey) to guarantee demand for permanent removals
- Reported evidence
- Contracted hundreds of millions of dollars in ex-ante and delivery-based purchases across DAC, biomass, mineralization, and geochemical pathways.
- Announced next step
- Catalyzing an industrial ecosystem capable of delivering gigatonne-scale removal by 2050.
- Unresolved risk
- Counterparty delivery delays, lack of harmonized global MRV standards, and additionality questions under Article 6 of the Paris Agreement.
Announced removal purchases describe advance contracts, not delivered physical carbon in the geosphere. Durability requires verified monitoring across centuries.
An optimistic view, with conditions
The curve can move if capture and storage work together
Contactors, sorbents, heat integration, modular manufacturing, and shared storage can all improve far above the physical minimum. The proof is sustained net tonnes, not increasingly large announcements.
Measure the whole plant
Publish energy, sorbent loss, uptime, gross capture, lifecycle emissions, and injected mass on consistent boundaries.
Share infrastructure
Pair multiple capture approaches with characterized storage, compression, transport, and common MRV.
Buy net tonnes
Procurement should pay after verified delivery and make durability and reversal liability explicit.
Removal is already broader than direct air capture
The 2026 State of Carbon Dioxide Removal assessment estimates 2.0 MtCO₂ of novel removal in 2025, up from 1.4 Mt in 2023. Biochar accounts for about 1.46 Mt of the 2025 estimate, nearly three quarters. In the disclosed durable-credit market, CDR.fyi reports biochar supplied 89.4% of delivered tonnes in Q2 2025, while large BECCS commitments dominated purchases. These are distinct accounting systems: estimated physical removal is not the same as credits sold or delivered.
Scaling 2 Mt/yr to 1 Gt/yr requires a 500-fold increase. The same assessment’s deployment chapter counts 42 Mt/yr of 2030 company capacity ambitions, but says reaching those plans from 2025 deployment would require about 84% compound growth each year. Ambitions are not permitted, financed or operating plants. DAC deserves close attention because it measures atmospheric carbon directly and can pair with durable geological storage, yet its delivered share is currently small.
The storage bottleneck has a public audit trail. EPA’s Class VI tracker lists US geological injection permits under review; states with their own permitting authority are outside that tracker. An application, a final permit, and demonstrated annual injection are three different stages. For STRATOS, 1PointFive describes 500,000 t/yr as design capacity once fully operational; the article should not treat it as captured or stored volume until an operating report publishes that result.
Clean electricity also has an opportunity cost. At 1.5–2.5 MWh of plant energy per tonne, one MWh supports only about 0.4–0.67 gross DAC tonnes before lifecycle emissions, transport and storage. Where that MWh can immediately displace fossil generation, the avoided emissions may be more valuable than DAC; where a grid is already very clean or curtails supply, removal becomes more attractive. The comparison needs hourly marginal power emissions and verified net removal, not an annual average carbon label.
Count permits, contracted tonnes and delivered tonnes separately
The DOE regional DAC hubs page describes a phased funding programme, and the Project Cypress award sheet says later funding depended on go/no-go decisions. An award, an operating plant and an audited net removal are different milestones. EPA's Class VI tracker covers applications under federal review but excludes wells in states with permitting primacy; its count cannot be used as a national time series without state data. Published DAC offtake prices reflect buyer contracts; no consistent audited operating-cost series exists across Mammoth, STRATOS and the other named plants.
Sources, method, and boundaries
The thermodynamic figure is a transparent ideal-gas concentration estimate, not observed plant consumption. Government goals and awarded capacity are labeled as targets and programmes, never operating performance.
