Industries / Energy, water and waste / Carbon capture
Capture from air
Taking carbon dioxide out of ordinary outside air, where it is far more dilute than in a smokestack, with machines that blow air over a material that holds CO2 and then release it, by heat, for storage or use.
6
Milestones achieved, from 1999 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1999
Air capture proposed
Klaus Lackner, Hans-Joachim Ziock and Patrick Grimes of Los Alamos National Laboratory argued in February 1999 that CO2 could be taken directly from the air. Capture at smokestacks, they wrote, cannot deal with diffuse sources or with old plants that are too costly to fit with capture, and air capture would also need no network of pipelines from each source. They concluded that "there are no fundamental obstacles to this approach."
Credit
Klaus Lackner, Hans-Joachim Ziock and Patrick Grimes, Los Alamos National Laboratory
2017
Air CO2 sold to growers
Climeworks, a spin-off of ETH Zurich founded in 2009 by Christoph Gebald and Jan Wurzbacher, opened a plant at Hinwil near Zurich that Carbon Brief, which attended the opening, called the world's first "commercial" direct air capture plant. It sits on the roof of a municipal waste incinerator that supplies the heat it needs, can capture 900 tonnes of CO2 a year, and sells the CO2 to the greenhouses of Gebrüder Meier Primanatura nearby. Gebald put the cost at about $600 a tonne; Carbon Brief noted that the sales did not cover the plant's full costs.
Credit
Climeworks, with Gebrüder Meier Primanatura, the waste company KEZO and the Swiss Federal Office of Energy
2018
A pilot sets a cost
David Keith, Geoffrey Holmes, David St. Angelo and Kenton Heidel of Carbon Engineering published in the journal Joule the design of a plant to capture about 1 million tonnes of CO2 a year from air, using a potassium hydroxide solution and a calcium loop to recover it. They reported results from a pilot plant for the main process steps and a capital cost breakdown developed with an independent engineering firm, and put the cost at $94 to $232 a tonne, depending on financial and energy assumptions. 1PointFive's Stratos plant in Texas uses Carbon Engineering's technology.
Credit
Carbon Engineering
2021
Air capture meets storage
Climeworks opened Orca at Hellisheiði in Iceland in September 2021, designed to remove 4,000 tonnes of CO2 a year. It sends the captured CO2 to Carbfix, a subsidiary of Reykjavík Energy, which pumps it into the ground. Climeworks calls Orca the world's first large-scale plant for direct air capture and storage.
Credit
Climeworks and Carbfix
2025
Output far below design
On 15 May 2025 the Icelandic paper Heimildin reported, from Climeworks' own figures, that the company had captured just over 2,400 tonnes of CO2 in Iceland since 2021, about 1,000 tonnes in its best year. Mammoth, its second plant there, designed for 36,000 tonnes a year and partly commissioned in 2024, captured 105 tonnes in its first ten months. Poor capture led Climeworks to write down the value of Orca by $1.4 million in 2023, and by its own count the company's footprint was 1,700 tonnes of CO2 in 2023. Climeworks did not answer Heimildin's questions before publication.
LessonAn air capture plant's design capacity says little about what it delivers; the figure that matters is tonnes stored per year, checked by someone outside.
Credit
Heimildin reporters Bjartmar Oddur Þeyr Alexandersson and Valur Grettisson
2026
Air capture hubs kept
The US DOE had awarded two direct air capture hubs, Project Cypress in Louisiana and the South Texas hub, a 1PointFive project, $550 million and $500 million, of which each had received a first $50 million. In October 2025 both were on a list of awards the department was considering cancelling. In April 2026 DOE told Congress that both would keep their funding; Energy Secretary Chris Wright said the department backed projects that "had a credible way to be helpful."
LessonAir capture at this size depends on public money, and the terms of that money can change with each government.
Credit
US Department of Energy, as reported by Pipeline & Gas Journal from Reuters
Watch for
Mammoth reaches design rate
Climeworks says changes to its sorbent, the filter material that holds CO2, and to the plant's machinery have more than doubled capture in upgraded collector containers at Mammoth. Two upgraded containers, run for more than six months, reach the design peak of 1.37 tonnes of CO2 a day each, and Mammoth delivered 675 tonnes of net removal (CO2 stored after subtracting the plant's own emissions and downtime) in the first half of 2026, against 119 tonnes in the first half of 2025. It says operating cost per tonne fell by more than half and that all 12 containers of one module will be upgraded by the end of 2026.
Achieved whenan independent verifier or carbon registry publishes a full year of Mammoth's net removals with the upgraded module, and the total matches the claimed rates.
Claimed by
Climeworks and its co-chief executive Jan Wurzbacher, September 2026
Stratos comes online
1PointFive, a subsidiary of Occidental, is building Stratos in Ector County, Texas, designed to capture up to 500,000 tonnes of CO2 a year from air once fully running, with storage permits (Class VI well permits from the US Environmental Protection Agency) secured in April 2025. In September 2025 Occidental said Stratos was on track to begin capturing CO2 later that year. In March 2026 1PointFive said Phase 1 was "in the final stage of startup" and expected online in the second quarter of 2026, with the remaining pellet reactors, the calciner (the kiln that releases the CO2) and CO2 injection still to start. An open data set built from state records in October 2026 found that Brown Pelican, the plant's storage site, had its permit and wells but was not yet injecting, with the plant then due at the end of 2026.
Achieved whenCO2 injection has begun and EPA monitoring reports or an independent verifier publish the tonnes captured and stored.
Claimed by
1PointFive and Occidental, September 2025 and March 2026
Open
01
Test the AI-picked sorbents
Make and measure, in a university lab, a small set of metal-organic frameworks (MOFs, crystals with pores built from metal atoms and organic links) on which two computer screens for air capture disagree, to find out which screen predicts real CO2 uptake from humid air.
The briefCredit · unclaimed02
Sorbent aging in outdoor air
A small rig that cycles a commercial amine sorbent between taking CO2 from outdoor air and releasing it by heat for months, beside a twin column fed the same air through a filter, logging CO2 taken up on every cycle along with ozone, temperature and humidity, and publishing all the raw data.
The briefCredit · unclaimed
Open 01 · the brief
Test the AI-picked sorbents
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Make and measure, in a university lab, a small set of metal-organic frameworks (MOFs, crystals with pores built from metal atoms and organic links) on which two computer screens for air capture disagree, to find out which screen predicts real CO2 uptake from humid air.
Why it is still open
The Open DAC 2023 project of Meta's Fundamental AI Research lab and Georgia Tech (Anuroop Sriram, David Sholl and colleagues) ran more than 38 million quantum chemistry calculations on more than 8,400 MOFs and named 135 as promising for air capture. In May 2025 Xin Jin, Susana Garcia and Berend Smit of EPFL reported that when they redid the calculation they found only one MOF that qualified, ASEJOZ01, which was not on the list, and that about 40% of the structures used were flagged as wrongly charged by the checking tool MOFChecker. The follow-up data set, ODAC25 (about 60 million calculations on about 15,000 MOFs), tests its models against other calculations, not lab measurements. A search of the papers that cite Open DAC 2023 found none that made and tested its picks.
Built with what exists
The ODAC23 and ODAC25 data and the machine-learned models released with them in Meta's fairchem library; MOFChecker; the CoRE MOF 2025 database; the Cambridge Structural Database for published recipes; a volumetric gas sorption analyzer and a humid breakthrough column, standard equipment in university porous-materials labs.
First step
Run MOFChecker over the 135 recommended structures and list which have a published recipe that a lab could repeat.
Done looks like
At least ten MOFs, including ASEJOZ01 and at least five of the 135 that pass MOFChecker, measured for CO2 uptake at 400 ppm and 25 °C, dry and at 50% relative humidity, with the raw data published and a table scoring each screen's predictions against the measurements.
Open 02 · the brief
Sorbent aging in outdoor air
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A small rig that cycles a commercial amine sorbent between taking CO2 from outdoor air and releasing it by heat for months, beside a twin column fed the same air through a filter, logging CO2 taken up on every cycle along with ozone, temperature and humidity, and publishing all the raw data.
Why it is still open
Climeworks says the sorbent decides how much CO2 a plant can capture and how often the material must be replaced, and its reported tenfold gain in sorbent life is a lab result. The US National Energy Technology Laboratory (NETL) wrote in May 2024 that sorbents based on polymeric amines "degrade when exposed to air, humidity, and heat" and that checking a sorbent's lifetime "requires an expensive long-term test", so it is developing accelerated aging tests instead. Shubham Jamdade, Xuqing Cai and David Sholl wrote in 2025 that wear from trace ozone in outdoor air "appears to have not been considered previously," and estimated it with calculations only. Searches of the web, Crossref and GitHub found isotherm data and lab aging studies for common sorbents, but no open record of long cycling in real outdoor air.
Built with what exists
Lewatit VP OC 1065, an amine resin sold by LANXESS and widely used in air capture research; two small packed tubes with fans and heating tape; Sensirion SCD30 or SCD41 NDIR CO2 sensors, checked against a reference gas; ozone readings from the nearest government air monitor (EPA AirNow in the US, the European Environment Agency in Europe) or a calibrated ozone sensor; a Raspberry Pi to log; Python.
First step
Buy a kilogram of the resin and two sensors, and run 50 cycles indoors to check that the CO2 taken up matches the CO2 released within 10%.
Done looks like
At least 1,000 cycles over at least three months, with CO2 uptake per cycle for the outdoor and filtered columns, the loss in uptake plotted against ozone dose with its uncertainty, and the logs and code published.