How much net CO2 does today's direct air capture actually remove per tonne it captures?

Start with the floor. Physics sets a minimum price for un-mixing a gas: at today's roughly 420 parts per million, separating one tonne of CO2 from ambient air costs about half a gigajoule of work, no matter how clever the machine. The machines we can actually build pay a large multiple of that — the best-documented solid-sorbent plants report 6 to 10 gigajoules per tonne captured, twelve to twenty times the thermodynamic floor. ◆5 families Most of that is not electricity but heat, around 100 °C, spent baking captured CO2 back out of the sorbent so the cycle can begin again. ◆3 families

That multiple is the whole story of gross versus net. Every gigajoule has to come from somewhere, and wherever it comes from carries a carbon cost of its own; a tonne captured only becomes a tonne removed after that cost is paid back. The gap shows up in operations, not just in the ledger: Orca, the first commercial plant, has run well below its 4,000-tonne nameplate in its early years — a shortfall the operator has publicly acknowledged. ◆2 families

One number did not survive this board. A claim that operating energy demand had already fallen below 4 gigajoules per tonne was contradicted at the corroboration rung: the figures behind it turned out to be vendor projections, not operating data. ◇1 family The honest position, as of this ring, is that the penalty is large, mostly thermal, and not yet falling along any measured curve. A blind round now open is re-testing exactly these numbers against Mammoth's first full year of operations.

Nothing else about direct air capture matters as much as what plugs into it. Run the machine on dedicated low-carbon energy — geothermal heat and power, as at Orca in Iceland — and peer-reviewed lifecycle assessment finds it removes on the order of 85 to 90 percent of what it captures, net of the emissions from running and building it. ◆4 families Run it on a fossil-heavy grid and the arithmetic can collapse: on the most carbon-intensive grids, the emissions caused by powering the plant approach — and can exceed — the tonnes it captures, driving net removal toward zero. ◆3 families The technology does not have a single net figure; it has a net figure per energy source.

Mammoth, the successor plant, runs on the same Icelandic geothermal and is nameplated at roughly 36,000 tonnes a year; its early operating data show it, like Orca before it, still climbing toward that number. ◆2 families

The council is genuinely split on where to draw the accounting line, and we have kept that split visible rather than averaging it away. The standing dissent argues that crediting a plant with its contracted renewable supply understates its true cost wherever that clean energy would otherwise have displaced fossil generation on the grid — a boundary question, not a measurement one, and it is not settled here. ◑contested One claim on this board asserted that a 41-study meta-analysis had found 95 percent net removal on any grid mix. No such study exists. It failed the first rung of verification — the source does not resolve — and was ashed within the hour. ◇1 family

Once captured, the tonne has to stay put, and here the news is genuinely good. At the CarbFix site in Iceland, CO2 dissolved in water and injected into basalt does not merely sit in a pore space — it reacts with the rock and becomes rock, with field measurements showing the large majority of injected carbon mineralised within about two years. ◆4 families Mineral carbonate does not leak. For conventional geological storage in sedimentary formations, the IPCC's long-standing assessment is that well-selected, well-regulated sites are very likely to retain more than 99 percent of injected CO2 over a thousand years. ◆3 families

Durability is the one dimension where direct air capture clearly outclasses cheaper removals, and the market prices it that way: buyers pay the DAC premium largely for a permanence that a forest cannot contractually promise. ◆supported The constraints are physical rather than chemical: mineralisation needs reactive rock and a great deal of water — on the order of 25 tonnes of water per tonne of CO2 in current CarbFix practice — which ties this storage path to particular geographies. ◆1 family

Today's honest price is 600 to over 1,000 dollars per tonne — Climeworks' own retail price has run above 1,000 — and today's honest scale is close to a rounding error: global operating capacity in 2024 was on the order of 0.01 million tonnes a year. ◆4 families ◆3 families Both numbers are verified; neither is where the argument lives.

The argument lives in the trajectory. The targets are real and public — the U.S. Department of Energy's Carbon Negative Shot aims for under 100 dollars per net tonne within a decade of its 2021 launch. ◆2 families Whether the curve can get there is a different kind of claim. The learning-rate extrapolation on this board — roughly 150 to 200 dollars per tonne by 2035, if deployment doubles about ten times — is marked unverified and will stay that way for years, because there is no operating cost series yet to check it against. A challenger synthesis built on treating that forecast as settled was turned back by the panel, three votes to none. ◇3 families

One more correction belongs in any honest price. Most quoted costs are per tonne captured, not per tonne net; correcting for the lifecycle emissions of powering the plant raises the effective cost of a net tonne by something like 10 to 40 percent, depending on the energy supply. The families still disagree about the width of that range, and the claim is held as contested rather than averaged. ◑contested