Direct Air Capture Machines That Think Ahead Could Nearly Double Their Real Climate Payoff

Direct air capture has an energy problem. Machines that pull carbon dioxide straight out of the atmosphere typically consume enormous amounts of heat and electricity per tonne captured, which is the main reason the going price for DAC-based carbon removal still sits above $500 per tonne, according to weighted averages compiled by CDR.fyi in October 2026. A new modeling study suggests the cheapest fix may not be a better sorbent or a bigger machine. It may be better timing.

The research, led by Keju An and colleagues at the National Laboratory of the Rockies in Golden, Colorado, and published in Frontiers in Environmental Engineering, asks a deceptively simple question: what if DAC machines stopped running on fixed settings and instead adapted hour by hour to the weather and the power grid? The answer, in the team's simulations, is a near doubling of the real climate payoff.

Why running on autopilot wastes energy

A solid-sorbent DAC unit cycles air through materials that grab CO2, then releases the gas under heat and vacuum in a six-stage temperature-vacuum swing adsorption (TVSA) cycle: vacuum extraction, heating, desorption, cooling, pressurization, and adsorption. But the air never stays the same. Temperature swings change how eagerly sorbents bind CO2. Humidity matters too: water co-adsorbs alongside CO2 and must be boiled off during regeneration, costing extra energy. A machine tuned to fixed setpoints burns energy whether the hour is favorable or not.

Until now, engineers have lacked the tools to study this properly. Detailed adsorption models are so computationally expensive that simulating a full year of hour-by-hour operation was impractical, so plants are designed and operated from cycle-averaged assumptions. That made it impossible to explore how machines should behave as weather and grid carbon intensity change around them, which is precisely where a large share of performance may be hiding.

Building a model that remembers

The team's breakthrough is a physics-informed reduced-order model: a stripped-down but physically faithful mathematical description of the TVSA cycle that runs at a tiny fraction of the cost of a full simulation. It tracks six coupled states, including fast and slow CO2 loading pools on the sorbent, free and bound water, gas-phase CO2 concentration, and bed temperature.

Crucially, it captures cycle-to-cycle memory. Incomplete regeneration means each cycle never fully clears the sorbent, leaving residual loading that carries into the next cycle. The sorbent's starting state, and therefore its working capacity and regeneration energy demand, depends on everything that happened before. Performance cannot be predicted from a single steady-state snapshot, which is why simpler models miss the opportunity adaptive control would exploit.

The researchers generated roughly 120,000 cycle-resolved simulations using Latin hypercube sampling across weather conditions, operating strategies, system designs, and initial sorbent states, then trained a compact neural network, a single hidden layer of 39 neurons, to reproduce the physics model's state transitions at minimal computational cost. That surrogate was fast enough to embed inside a model predictive control framework that plans each cycle adaptively. Before trusting it, the team benchmarked the underlying physics model against published experimental data from Sendi and colleagues, reproducing CO2 productivity within about 10 percent and electricity demand within about 11 percent error across twelve cases.

The biggest untapped efficiency reserve in direct air capture may not be new materials or bigger machines, but smarter timing.

The two-site test

The controller was tested against real hourly 2025 weather, electricity price, and grid carbon-intensity data for two contrasting American sites: humid Atoka, Texas and dry, cooler Madison, Wyoming. The baseline machine ran fixed settings all year, completing about 1,094 cycles. The adaptive controller ran roughly 1,449 cycles in Atoka and 1,379 in Madison, by shifting energy-hungry regeneration steps into hours when electricity was cheap and clean and exploiting favorable humidity windows for capture.

The results were striking. In Atoka, electricity demand fell 18.8 percent while annual CO2 productivity rose 25.5 percent. In Madison, energy use dropped 8.9 percent while productivity climbed 22.4 percent. And on the metric that actually matters, carbon-adjusted net productivity (capture minus the emissions embedded in the electricity consumed), the gains nearly doubled at both sites: up 95.2 percent in Texas and 94.4 percent in Wyoming.

Why timing beats bigger machines

Sensitivity analysis revealed how much the control knobs matter. Raising regeneration temperature from 70 to 130 degrees Celsius cut annual productivity by roughly 10 tonnes of CO2 per cubic meter of sorbent bed while adding 0.7 to 0.8 megawatt-hours of electricity demand per tonne captured, a punishing double penalty from hotter, slower cycles. Humidity climbing from 0.1 to 0.9 relative humidity added roughly 0.5 MWh per tonne, the direct cost of evaporating co-adsorbed water.

This is the strategic implication for an industry racing to scale. Mega-projects dominate the headlines, but the per-tonne economics of every plant hinge on energy. A software layer that squeezes nearly a fifth off energy use and a quarter more throughput from identical hardware is the kind of lever that changes project economics without a construction permit. As the recent BCG-XPRIZE analysis noted, operational control and process optimization are among the most important levers for bringing down the cost of atmospheric carbon removal.

There is a wrinkle. The economics did not always follow the climate math: electricity costs fell slightly in Atoka but rose 3.7 percent in Madison, because carbon-aware control sometimes pays more to harvest cleaner power. Plant operators chasing the cheapest kilowatt-hour may not automatically chase the cleanest one, and that tension between cost optimization and carbon optimization is now quantified for the first time.

The caveats

The authors are careful about what this is. The model is a control-oriented screening tool, not a plant-specific predictor, and because the surrogate learns from the physics model, its uncertainties propagate into the optimization results. The reported gains are relative, model-based estimates rather than guarantees for any real installation, and validation on operating hardware is the necessary next step.

Still, the framework delivers what the field has lacked: a scalable path from adsorption physics through fast dynamic simulation to real-world control. If DAC is to close the gap between $500-plus tonnes and the sub-$100 aspirations of the industry, it will need every lever available. This study makes a persuasive case that one of the biggest ones has been hiding in plain sight: not the machine, but the schedule it runs on.