Context & scale

Projecting the future cost of new technologies is a key challenge for research and policymaking related to the low-carbon energy transition. While experience curve-based extrapolations are widely used, they cannot be directly applied to novel technologies where past deployment data are lacking. Here, we introduce a new methodology for technology cost forecasting based on an empirical analysis of technology-inherent characteristics that determine the cost reduction potential. Our results reduce uncertainty about the future cost of novel energy technologies and can inform integrated assessment modeling and climate policymaking. The methodology can be applied to any novel technology with a defined initial component scale, expanding the toolbox for technology cost forecasting.

Highlights

•

Proposes new method to project future costs of novel technologies

•

Combines technology-inherent characteristics with component-based experience curves

•

Method applied to direct air capture (DAC)

•

Net removed costs projected at $226–$835/tCO2 for 1 Gt-CO2/year cumulative capacity

Summary

Several low-carbon technologies, such as solar photovoltaics or batteries, have experienced massive cost reductions in the recent past. However, non-mature technologies will also be required to meet the Paris climate targets. The cost of novel technologies, like direct air capture (DAC) technologies, remains highly uncertain. Here, we introduce a new method to project future costs of novel technologies by assigning empirically grounded experience rates to technology components based on their similarity to mature technologies in terms of design complexity and customization needs. After an ex-post validation of this method, we apply it to three DAC technologies combined with CO2 transport and storage (DACCS) to provide probabilistic estimates of the cost of CO2 net removed. At 1 Gt-CO2/year cumulative capacity, we project DACCS costs at $341/tCO2 ($226–$544 at 90% confidence) for liquid solvent DACCS, $374/tCO2 ($281–$579) for solid sorbent DACCS, and $371/tCO2 ($230–$835) for CaO ambient weathering DACCS.

Graphical abstract

Keywords

1. technology innovation
2. experience curves
3. cost projections
4. direct air capture
5. net removed cost
6. carbon dioxide removal
7. negative emissions
8. technological learning

Introduction

Rapid emission reductions are needed so that the Paris Agreement’s target to limit global warming to well below 2°C remains attainable.1 Pathways in line with this target presume a swift transition to low-carbon energy sources and—on top—the deployment of carbon dioxide removal (CDR) technologies2,3,4,5 to remove historic emissions and compensate for emissions that cannot be completely eliminated. Several renewable energy technologies, like solar photovoltaic (PV) and wind, have already experienced increased deployment and rapid cost reductions.6,7,8 Similar momentum is exhibited in the electrification of energy services, such as electric vehicles9 and heat pumps.10 However, we are still in the early development phase of additional low-carbon solutions, including many CDR technologies.11,12 As of 2022, total global CDR capacity was 2 Gt-CO2/year, of which only 0.1% was from novel CDR methods, a fraction of which is from engineered solutions such as direct air carbon capture (DAC) combined with storage (DACCS).13 The remaining capacity originates from conventional nature-based CDR methods that use land reservoirs for carbon storage.13 For context, the US alone is targeting approximately 1 Gt-CO2/year in CDR by 2050, combining both land sinks and engineered solutions.14 However, as conventional CDR methods are expected to saturate by mid-century,15 there is a need to scale up novel CDR methods including DACCS.

DACCS offers a scalable, permanent, and relatively easily measurable, reportable, and verifiable CDR method.16,17 It is therefore often regarded as a backstop technology.18,19,20 However, DAC technologies are still in their infancy; e.g., the first working prototype using solid sorbent DAC was only built in 2013.21 To date, high costs have hindered large-scale deployment of DACCS,22,23 and while decision-makers in the public and private sectors anticipate massive cost reductions with future DACCS deployment, the extent of these reductions remains unclear. While future technology costs are inherently uncertain,24,25 experience curves present a well-established approach for projecting costs.8,26,27,28,29,30 The few extant DACCS cost projections,18,21,31,32,33,34,35 however, feature highly inconsistent results, with results varying by a factor of 10 ($37–$386/tCO2). Recent studies use historical analogies for projecting DACCS cost34 and deployment.12,36,37 The challenge, however, lies in identifying which analogies are appropriate. Therefore, we focus on understanding and analyzing the inherent characteristics of these technologies and their components.

To this end, we leverage a recently developed framework that relates a technology’s cost reduction potential to two of its inherent characteristics. Specifically, we analyze how three DAC technologies and their individual components compare with mature technologies in terms of design complexity and the need for customization, thereby estimating their cost reduction potential as expressed in experience rates. After an ex-post validation of this method, we then use multi-component experience curves38,39,40 to perform probabilistic extrapolations of the cost of CO2 net removed in a mature technology state. The three technologies comprise (1) a high-temperature liquid solvent process, (2) a low-temperature solid sorbent process, and (3) a calcium oxide (CaO) ambient weathering process (see experimental procedures). Our findings show cost ranges at 1 Gt-CO2/year cumulative capacity of $226–$544/tCO2 (1), $281–$579/tCO2 (2), and $230–$835/tCO2 (3) (5th–95th percentile given in US$2022 terms). These cost estimates exceed those previously reported in the literature. While none of the assessed technologies reach the US long-term policy target for CDR of $100/tCO2 at Gt-scale, cost projections do align with those of alternative mitigation strategies such as sustainable aviation fuels. Hence, DACCS could indeed play an important role as a backstop technology contingent on accelerated deployment.

More broadly, this study provides researchers with a methodology to derive component-level experience rates for emerging low-carbon technologies, enabling cost projections without requiring historical deployment data.

Results

Estimating three DAC technologies’ experience rates

Methods commonly used for cost projections include expert elicitations41,42,43,44 and model-based approaches.8,28,30,45,46,47 Previous literature highlights that while experts are fully capable of describing current situations (“matters of fact”48), they struggle to estimate costs in the longer term,27 especially for emerging technologies.49 As a consequence, model-based projections relying on historical trends, and more specifically experience-curve-based methods, have outperformed expert elicitations in projecting the future costs of energy technologies.27 Experience curves posit that costs decrease with deployment due to learning.38 Specifically, Wright’s law50 indicates that every doubling of deployment reduces costs by a fixed, technology-specific percentage known as the experience rate.51

Given the nascent state of DAC technologies and their minimal installed capacity, historical deployment data are limited. Hence, extant experience curve analyses of DAC and DACCS18,21,31,32,33,34,35 assume wildly different experience rates, often referring to rates for other low-carbon technologies such as solar or wind. Our review reveals a wide range of previously applied experience rates: 10%–20% for capital expenditure (CAPEX) and 2%–10% for operational expenditure (OPEX) (Figure 1A). For solid sorbent DACCS, for example, this results in substantial variations in cost from $170 to $270/tCO2 at 0.5 Gt-CO2/year35 to $110/tCO2 at 362 Gt-CO2/year32 cumulative capacities (see Table S1).

Figure 1 Approach for determining experience rates for probabilistic cost projections

(A) Experience rates assumed in selected previous cost projections for liquid solvent and solid sorbent DACCS, ranging from 10% to 20% (CAPEX) and 2% to 2.5% (variable OPEX). These rates were taken from other low-carbon technologies without systematic assessment, leading to large inconsistencies in projected costs.

(B) Our approach for determining experience rates for nascent technologies. We break down technologies into components, rank them via expert interviews considering design complexity and customization needs, and derive component-level experience rates (including distributions) from technologies with analog complexities. When these rates are combined with technology-specific cost estimates for the initial capacity (equivalent to the first plant), probabilistic cost projections are obtained.

More recently, the innovation literature has demonstrated that the inherent characteristics of design complexity52,53 and customization need54,55 largely determine a technology’s experience rate.56 Design complexity describes the number of technical parts in a technology and their level of interaction, while the customization need is determined by the degree to which a technology must be adapted to its environment56 (details in Note S2). Simpler and more standardized technologies tend to move along the experience curve more rapidly than those with greater complexity and customization needs.56 A recently proposed technology typology56 distinguishes three technology types based on the combination of these two characteristics. Each technology type is associated with different average experience rates and standard deviations (type 1: 0.22, σ: 0.05; type 2: 0.13, σ: 0.04; type 3: 0.05, σ: 0.04) (Figure 1B).

Leveraging this link between experience rates and design complexity and customization need, we empirically evaluate both factors for novel DAC components and systems through expert interviews (Figure 1B). We first decompose DAC technologies into components, following established cost estimation guidelines.57,58,59,60 On the component level, we differentiate between novel and off-the-shelf components (Table S3). A component is considered novel if it is specifically developed for the DAC industry and not commercially available from any other industry. Conversely, components that are commercially available and do not require adaptation for DAC are categorized as off-the-shelf components. We then classify each novel component and the overall DAC technology (the system level61) according to its design complexity and need for customization (Figures 2A–2C), subsequently allocating experience rates in line with the identified technology types (Tables S4–S8). While in the main text, we focus on DACCS, we also apply the method to three established low-carbon technologies, namely concentrated solar thermal power plants, onshore wind, and solar PV, to validate our method ex-post (see Note S4).

Figure 2 Expert evaluation of design complexity and customization needs of DAC system levels and novel components

Expert interviewees (n = 22) rated system levels and novel components for liquid solvent (A), solid sorbent (B), and CaO ambient weathering DAC (C). Large circles represent average expert ratings while small icons represent individual ratings (see Tables S4–S6).

We analyze three DAC technologies with distinct CO2 capture processes but identical transport and storage pathways. Liquid solvent DAC utilizes continuous absorption and regeneration stages for CO2 capture,21 while solid sorbent DAC utilizes cyclic adsorption-desorption processes in CO2 collectors.62 CaO ambient weathering DAC binds CO2 to CaO in a carbonation phase and regenerates it in a calcination phase63 (for more details, see experimental procedures and Figure S1). The three technologies were selected for their comparatively high maturity,34 as demonstrated by existing pre-pilot, pilot, and commercial-scale plants.21,64,65

The high-temperature liquid solvent DAC technology uses potassium hydroxide (KOH) absorption paired with regeneration via calcium looping.66 At the system level, it is categorized as a type 3 technology with an estimated mean experience rate of 8% (Figure 2A). Being more complex than other DAC technologies, liquid solvent DAC can benefit from cost reductions attributed to larger component sizes (economies of scale per plant), though this comes at the cost of lacking modularity. The system comprises several novel components—including the air contactor and the interconnected regeneration unit, which consists of the pellet reactor, slaker, and calciner—as well as commercially available (“off-the-shelf”) components including air separation units, fines filter, compressors, steam turbines, and buildings. All novel components are design-intensive and can be mass-customized (type 2), with an estimated mean experience rate of 13%. The air contactor, similar to cooling towers, requires mass customization for meteorological variations. The interconnected pellet reactor, slaker, and calciner necessitate system-specific mass customization. Changes in inputs, e.g., pellets, require adjustments to all regeneration unit components. The three components operate on established principles. The pellet reactor, based on water treatment technology, is the simplest in design, followed by the slaker and calciner. The calciner, similar to rotary kilns in kraft pulp processes, is characterized by oxygen-fired operation in a high-purity oxygen environment and the material must be able to withstand high temperatures (900°C/1,652°F), making it the most complex component.

The low-temperature solid sorbent DAC technology uses an amine-functionalized sorbent material for separation of CO2 from air through a temperature-vacuum swing adsorption/desorption cycle.67,68 At the system level, it is design-intensive and can be mass-customized, primarily based on the integration of the sorbent material with the air contactor. It is classified as a type 2 technology and has an estimated mean experience rate of 12% (Figure 2B). The system comprises novel components—the air contactor with fans and the sorbent material—and commercially available (off-the-shelf) components such as switching valves, compressors, buildings, condensers, vacuum pumps, and gas storage balloons. Among the novel components, the air contactor is also classified as type 2, with an estimated mean experience rate of 14%. It is more complex than the liquid solvent DAC air contactor, due to its cyclic nature of adsorption and desorption. However, it holds the potential for standardization across environments. The sorbent material, a bulk polymer based on known cellulosic materials, was developed specifically for DAC and low CO2 concentrations. Despite its relatively simple design, it presents R&D challenges and customization needs for site-specific conditions, classifying it between types 2 and 3 with an estimated mean experience rate of 10%. It is likely that different sorbents will become established in different environments. Despite the sorbent material’s high customization needs, its chemical substance sets it apart from the mechanical components. The experience rate is estimated in line with that of bulk polymers, ranging from 18% to 37%69 (see Table S7).

The CaO ambient weathering DAC technology uses enhanced weathering with high-temperature calcination for repeated use of mineral feedstocks to remove CO2 from the air.63 At the system level, it falls under type 2, with an estimated mean experience rate of 14% (Figure 2C). The system-level design is characterized by a combination of simple and complex robotic components with many moving parts. Although most components can be standardized, customization is needed for the raw material layer to adapt to local conditions, such as temperature and humidity. The CaO ambient weathering DAC technology is composed of novel components like the large air contactor structures with tray movement robotics and an electric kiln, as well as commercially available (off-the-shelf) components such as compressors, conveyors, buildings, and raw materials. The novel components, the air contactor structure, and the electric kiln are also categorized as type 2 with an estimated mean experience of 13%. The air contactor is similar to that of the solid sorbent air contactor in terms of customization needs, although with slightly less complexity. Comparable designs are found in the mining and food industries, allowing for widespread use with minimal customization. The electric kiln, capable of withstanding temperatures up to 870°C/1,598°F, is more complex than the liquid solvent calciner. It is indirectly heated through custom-designed steel tubes and requires a reliable supply of renewable electricity. It can be mass-customized to accommodate regional temperature variations.

Probabilistic DACCS cost projections

Based on the characterization of the three DAC technologies and the derived component-level and system-level experience rates (Table S7), we perform DACCS net removed cost projections based on single- and multi-component experience curves. Figure 3 shows the projected cost trajectories, considering the use of solar PV electricity (tracking, with storage) at $39/MWh and natural gas for heat at $5.60/GJ (average energy cost 2022–205070,71). Cost estimates are given for the 50th (5th–95th) percentile.

Figure 3 Comparison of cost projections based on single-component and multi-component experience curves

Comparison of single-component experience curve approach (left) with multiple-component experience curve approach (center), alongside their respective distributions at 1 Gt-CO2/year cumulative design capacity (right). Each technology’s cost trajectory starts at the same initial plant scale for both approaches; deviations in costs become apparent as cumulative capacity increases. Thick lines represent the 50th percentile and colored areas the range between the 5th and 95th percentiles of cost of CO2 net removed projections. Costs are calculated based on energy cost projections from 2022 to 2050 using tracking solar PV and storage and natural gas. The curve curvatures are discussed in Note S5.

The cost estimates for the initial capacity (equivalent to the first commercial plant size) of liquid solvent, solid sorbent, and CaO ambient weathering DACCS are $670/tCO2 ($521–$894/tCO2), $1,282/tCO2 ($1,180–$1,392/tCO2), and $2,481/tCO2 ($2,153–$2,874/tCO2), respectively. At 1 Gt-CO2/year cumulative capacity, single-component experience curves (Figure 3, left) project low costs of $165/tCO2 ($43–$459/tCO2) for liquid solvent at an 8% experience rate, $36/tCO2 ($7–$236/tCO2) for solid sorbent at a 12% experience rate, and $16/tCO2 ($2–$123/tCO2) for CaO ambient weathering DACCS at a 14% experience rate. However, these single-component experience curves have two limitations: (1) they assume learning across all components, thereby overlooking the fact that some components do not profit from additional deployment as they are already widely used in other applications, e.g., compressors in gas pipelines. (2) Even for novel components, they cannot capture the wide variation in experience rates (here, 11%–27%).

In contrast, multi-component experience curves (Figure 3, center) can reflect those realities. At 1 Gt-CO2/year cumulative capacity, multi-component experience curves, present a more conservative picture, with lower average cost reductions. The costs fall to $341/tCO2 ($226–$544) for liquid solvent, $374/tCO2 ($281–$579) for solid sorbent, and $371/tCO2 ($230–$835) for CaO ambient weathering DACCS. As cumulative capacity increases, the multi-component experience curves flatten, reflecting a diminished proportion of costs for components with higher experience rates.72 For liquid solvent, solid sorbent, and CaO ambient weathering DACCS, from the initial capacity up to 10 Mt-CO2/year cumulative capacity, the experience rates are 7%, 8%, and 10% respectively for the 50th percentile. As cumulative capacity increases from 10 Mt to 1 Gt-CO2/year, the experience rates decline to 5%, 5%, and 7% respectively.

To compare the DACCS net removed cost trajectories based on multi-component experience curves, two key differences emerge: the first commercial plant sizes and their respective operational start dates, and the technological readiness levels (TRLs) of these plants. A solid sorbent DAC plant with a capacity of 4,000 tCO2/year has been operational in Iceland since 2021,73 a CaO ambient weathering DAC plant with a capacity of 1,000 tCO2/year has been operational in California, US, since 2023,74 and a liquid solvent DAC plant with a capacity of 500,000 tCO2/year is under construction in Texas, US, and is expected to be operational in 2025.75 The TRLs correspond to the commissioning year, with solid sorbent DAC being the most advanced and therefore having the highest TRL of 7–8. Conversely, no liquid solvent DAC plant has yet been commissioned. It therefore has the lowest TRL of 4 (see Table S16). The first commercial plant sizes are reflected in the different starting points of the experience curves and the variation in TRLs results in different initial cost ranges (5th–95th percentile). Figure 5 shows the years in which the plants started to operate.

For liquid solvent DACCS, at an initial capacity of 500,000 tCO2/year, the initial 5th–95th percentile cost range is highest, reflective of a comparatively low TRL of 4.58 Despite the large cost range, liquid solvent DACCS demonstrates the lowest initial costs, because components already profit from cost reductions attributed to large component sizes.66 Net removed costs for liquid solvent DACCS are projected to decrease by 49% from initial to 1 Gt-CO2/year (50th percentile) after 11 doublings in cumulative capacity. While novel components (11% of initial costs) benefit from experience rates between 11% and 15%, the majority of the initial costs (89%) is tied to components with low experience rates (2%–8%) (detailed in Note S5).

For solid sorbent DACCS, at an initial capacity of 4,000 tCO2/year, the initial 5th–95th percentile cost range is lowest, reflective of a comparatively high TRL of 7–8.58 Net removed costs of solid sorbent DACCS are projected to decrease by 71% from initial to 1 Gt-CO2/year after 18 doublings in cumulative capacity. Novel components (9% of initial costs) benefit from high experience rates (14%–27%). However, the remaining cost components (91% of initial costs) are associated with lower and medium experience rates of 13.5% or less (detailed in Note S5).

For CaO ambient weathering DACCS, with a 1,000 tCO2/year initial capacity, the initial 5th–95th percentile cost range is smaller than for liquid solvent but larger than for solid sorbent DACCS, reflective of a TRL of 5–6.58 Compared with the other two processes, CaO ambient weathering DACCS has the highest initial cost. However, at the 1 Gt-CO2/year cumulative capacity, shows the lowest cost (50th percentile), though it exhibits a broader 5th–95th percentile cost range compared with the other processes. Net removed costs of CaO ambient weathering DACCS are projected to decrease by 85% from initial to 1 Gt-CO2/year after 20 doublings in cumulative capacity. With high estimated experience rates of 11%–15%, cost reductions are driven by novel components, contingency, and start-up costs (90% of initial costs) (detailed in Note S5).

Beyond the uncertainties included in the Monte Carlo simulations (e.g., the capacity factor; see Table S16), the sensitivity analysis (detailed in supplemental experimental procedures) shows that that the discount rate, plant lifetime and electricity and/or heat requirements are among the top 20 most sensitive parameters for the net removed cost of the three DAC approaches.

Influence of energy costs on DACCS net removed cost

Given the high energy requirements of DAC technologies, future energy costs will substantially influence the net removed cost.33,76

Figure 4A shows DACCS net removed cost at initial capacities with 2022 energy costs (left-hand bars), DACCS net removed cost at 1 Gt-CO2/year cumulative capacity and still with 2022 energy costs (center bars), and DACCS net removed cost at 1 Gt-CO2/year cumulative capacity but at projected 2050 energy costs (right-hand bars). Our base case considers tracking solar PV with storage (12 h battery) for electricity and natural gas for heat, with 2022–2050 energy costs ranging from $79/MWh to $29/MWh for solar PV and $5.50/GJ to $5.80/GJ for natural gas.70,71 Initial net removed costs of all three DAC technologies are determined by CAPEX and fixed OPEX, contributing to 75%–91% of total net removed costs.

Figure 4 Breakdown and comparison of cost shares for different DAC technologies and energy sources

(A) Comparison of cost shares of liquid solvent, solid sorbent, and CaO ambient weathering DACCS, using tracking solar PV and storage (battery for 12 h), as well as natural gas as energy sources. The bars represent cost of CO2 net removed, including the cost share of capital and fixed operating expenses (CAPEX + fixed OPEX) and variable operating expenses (OPEX), and their combined absolute values shown atop each bar. The bars differentiate costs across increasing cumulative capacity and between energy cost projections for 2022 and 2050.

(B) Absolute net removed cost in $/tCO2 for these technologies at 1 Gt-CO2/year cumulative design capacity and different electricity inputs at 2022 prices. For additional analyses considering wind electricity, see Figure S5.

For all three technologies and scenarios, an increase to 1 Gt-CO2/year cumulative capacity lowers the CAPEX and fixed OPEX shares in cost due to experience effects, subsequently raising the variable OPEX share. In the projections with 2050 energy costs, net removed cost shares at 1 Gt-CO2/year cumulative capacity shift with falling electricity costs and rising natural gas prices. The fully electrified solid sorbent and CaO ambient weathering DACCS benefit significantly from declining electricity costs (with cost reductions of $51/tCO2 and $77/tCO2, respectively). Liquid solvent DACCS, which relies on both electricity and natural gas, is less likely to decline in variable OPEX as the assumed rise in natural gas costs counterbalances electricity cost reductions. Small differences in net removed costs at 1 Gt-CO2/year cumulative capacity are observed when analyzing different electricity sources at both initial generation levels and in 2050 (Figure 4B).

Discussion

In this paper, we project the future costs of DACCS considering technology characteristics. More generally, we develop—and based on historic cases—validate a novel cost projection approach, particularly valuable for technologies with limited historical deployment. We thereby leverage the strengths of experience curves to expand the toolkit for technology forecasting. Rather than asking experts for notoriously difficult cost predictions,77 we use expert interviews to empirically assess two inherent characteristics of existing technology components—factors that experts can judge more easily. We then leverage recent advancements in understanding inherent technology characteristics to derive component-level experience rates to perform probabilistic cost extrapolations using multi-component experience curves. This offers a more realistic estimate than single-component experience curves, which risk overestimating cost reductions by failing to separate experience effects.78,79

In principle, our method can be applied to any novel technology. The approach is particularly useful for technologies comprising both novel and commercially available components for which experience rates are unavailable or cannot be calculated based on empirical observations, such as advanced geothermal, fusion reactors or small modular fission reactors. The ex-post validation of the method (see Note S4) shows that some components, such as solar PV modules, have exceptionally high experience rates that warrant individual re-evaluation (as has been done for DACCS solid sorbent material, see Table S7).

Contextualizing DACCS cost projections within net-zero targets, Figure 5 shows multi-component experience curves at 2050 energy prices. This allows for a direct comparison of alternative DAC technologies at 1 Gt-CO2/year cumulative capacity and places them alongside 2050 forecasts for the US CDR policy target,80 sustainable aviation fuel abatement cost projections, and EU Emissions Trading System (EU ETS) price projections.

Figure 5 Net removed cost trajectories for liquid solvent DACCS (blue), solid sorbent DACCS (orange), and CaO ambient weathering DACCS (green) at 2050 energy prices, on a log-log scale

Each technology’s cost trajectory starts at the initial plant scale and increases to 1 Gt-CO2/year cumulative design capacity. Each DACCS system starts its cost trajectory when the first commercial scale plant begins operation, with the start year varying and highlighted for each process. Thick lines represent the 50th percentile and colored areas the range between the 5th and 95th percentiles of cost projections. The dark gray lines represent the long-term US CDR policy target of $100/tCO280 (1), and the projected 2040 and 2050 EU ETS prices at $2022282/tCO2 and $2022464/tCO281 (2), and the light gray lines highlight the abatement cost estimates projected for sustainable aviation fuel in 205082 (3) (see Table S23).

Comparing the three DACCS systems, although clear differences emerge in initial net removed costs, overlapping 5th–95th interpercentile ranges at 1 Gt-CO2/year cumulative capacity between all three technologies ($272–$542/tCO2) underline the preliminary nature of the ranking. Given the uncertainty involved, all three DAC technologies including transport and storage should be rapidly deployed to see what cost improvements will be achieved; our projections give an estimate of potential cost reductions, complementing recent literature that focuses on deployment projections.12,36,37

Importantly, our results cast doubt on the $100/tCO2 target established by Frontier’s Advance Market Commitment (AMC)83 and the US Department of Energy.80 This target was originally intended to reflect an average CDR cost across various nature-based and engineered solutions, yet it has been widely regarded as the benchmark for economic viability for DACCS.34,65,84 The widespread acceptance of this target for DACCS has led to overly optimistic cost estimates from academics (e.g., in integrated assessment modeling studies) and start-ups,85 which in turn has reinforced the target as a critical goal. While DACCS will very likely be part of the portfolio of CDR measures needed to limit global warming to well below 2°C, it is crucial to avoid unrealistic cost projections.

Although no DACCS system meets the desired cost target at 1 Gt-scale in our results, they all show potential to be competitive when compared with other technologies for hard-to-abate emissions (Figure 5). For instance, the abatement costs of sustainable aviation fuels are projected to range from $245 to $409/tCO2 by 2050.82 The feasibility of DACCS is further emphasized when contrasted with carbon prices that reflect the marginal abatement cost on an economy-wide level, such as the $2022464/tCO2 projected in the EU ETS by 2050.81

For faster deployment, DACCS could benefit from strong demand-pull. To date, very few stringent DACCS policies have been enacted. Key examples of deployment policies include the US Inflation Reduction Act, which raised the 45Q tax credit to $180/tCO2 and extended construction deadlines by 7 years,86 and the US Department of Energy’s funding for direct air capture (DAC) hubs.87 Additionally, private sector actions, including AMCs88,89 and pre-purchase agreements, complement these policies, driving DACCS deployment by generating extra revenue streams.

Yet, considering the uncertainty and possible shifts in cost rankings, a diversified DACCS technology portfolio remains crucial. Here again, policy can be instrumental. In the US, the recently introduced CREATE Act aims to accelerate CDR R&D.90 Under the EU Innovation Fund in Europe, €1 billion is earmarked for innovating low-carbon technologies, including DACCS.91 While these policies serve as a catalyst for advancing DACCS at various stages of maturity, beyond financing, support policy should be contingent on cost and performance reporting, thereby enabling further improvements in DACCS cost projections. Indeed, future DAC plants may differ from current designs, possibly incorporating new or replacing existing components. For example, Carbon Engineering, a company involved in liquid solvent DAC technology, has filed a patent for bipolar membrane electrodialysis (BPMED),92 potentially eliminating the need for a pellet reactor in future plants.

Experimental procedures

Resource availability

Lead contact

Inquiries regarding the data and method associated with this paper can be directed to the lead contact Tobias Schmidt ([email protected]).

Materials availability

Materials and methods are described in the experimental procedures.

Data and code availability

All data are available in the main text or the supplemental information. The Python model used to perform the analysis of this study has been deposited on GitHub available at: https://github.com/kfdsievert/Cost-Model–DAC.93 Users are required to cite the code when using it for their research. Detailed licensing information is available within the repository.

Approach

The key contribution of this study lies in leveraging recent advancements in understanding inherent technology characteristics to derive component-level experience rates for emerging low-carbon technologies for comprehensive cost projections (see below). Methods commonly used for cost projections include expert elicitations,41,42,43 and projections such as Wright’s law experience rates with cost as a function of cumulative capacity.8,28,30,45,46,47 In projecting future costs of emerging energy technologies, experience-curve-based methods have shown superior performance over expert elicitations.27

To perform probabilistic cost projections, we apply a model-based approach combining a cost analysis at the initial capacity (equivalent to the first plant) with experience curves to evaluate cost reductions as cumulative capacity increases. This approach is in line with previously suggested cost estimation methodologies for carbon capture, utilization, and storage.8,24,34,94 We project the net removed cost of DACCS for the first commercial scale plants of three different DAC technologies. We consider a representative location with an overall favorable cost position (including for the cost of energy) in a stable investment environment, using the US as a proxy. Considering the US as a proxy is further motivated by the fact that much of the existing and planned DACCS deployment to date is in the US. Accordingly, most of the empirical data used for the cost estimates are taken from the US context.

Initial plant capacity and cost of equipment

Initial plant capacity establishes the starting point for cost reductions and determines the number of doublings in cumulative design capacity up to 1 Gt-CO2/year scale. From the first commercial plant, costs usually decline because of operational optimization, process design improvements, mass manufacturing, and innovation.34,95 The selection of plant capacities differs across DAC technologies (Table S9).

The cost of DACCS: Net removed cost

The cost of DACCS represents the average net cost of removing one ton of CO2 for a given DAC technology including transport and storage, accounting for energy-related greenhouse gas (GHG) emissions (Tables S24 and S25) and the capacity factor. This approach ensures a comprehensive assessment of the true cost of CO2 removed, considering only energy-related emissions as they dominate in life cycle analysis of DACCS.62,66,96

DACCS cost in $/tCO2 (in US$2022 terms) is the sum of levelized capital cost (LEVCAPEX), annual fixed operating and maintenance cost (fixed OPEX) and annual variable operating cost (variable OPEX) where

LEVCAPEX=Ctoc∗CRFplantsize∗cf∗(1−x)

(Equation 1)

LEVCAPEX, in $/tCO2 removed, is calculated by first determining the total overnight cost (Ctoc)60 (Equation 2) which is subsequently annualized and levelized taking into account the capital recovery factor (CRF) (Equation 3), the plant size (plantsize) and energy related emissions (x), and the capacity factor (cf).

Total overnight cost Ctoc are calculated from direct materials cost of components (Cd), installation cost (Ci), and engineering and procurement cost (Cep), process and project contingencies (Cc), owner’s spare parts (Co), and startup cost (Cs)34,60,97 (Equation 2) (see Table S15).

Ctoc=Cd+Ci+Cep+Cc+Co+Cs

(Equation 2)

CRF is derived from a 7% discount rate (r), and a 25-year plant life (T) (see Tables S16–S18).

CRF=r(1+r)T(1+r)T−1

(Equation 3)

Energy related GHG emissions (Equation 1, represented as x) are calculated based on the carbon intensity of the energy sources and their respective energy consumption, and subsequently deducted from the plant size (refer to Tables S24 and S25 for details on the carbon intensities of different energy sources).

Fixed OPEX are calculated in $/tCO2 removed by considering direct and indirect labor, maintenance, insurance, local taxes and fees, and adjusting for plant capacity, and employee salaries34 (Table S19). Variable OPEX are calculated in $/tCO2 by accounting for electricity, natural gas, low-grade heat, water, chemicals and sorbent replacement,34 and CO2 transport and storage cost98 (Tables S20 and S21).

Scaling liquid solvent and solid sorbent DAC process equipment

For consistency between technologies, we scale the cost data provided for a first commercial solid sorbent DAC plant from the original reference plant size (960 tCO2/year) to a 4,000 tCO2/year plant size. For liquid solvent DAC, we scale down the cost from 980,000 to a 500,000 tCO2/year plant size. To this end, we scale the direct materials cost using a scaling relationship based on scaling factors from academic literature99,100,101 (see detailed description in Tables S10–S14).

Approach to derive component-level experience rates

Experience mechanisms are defined as those related to learning within and across organizations and economies of scale, as previously identified in innovation literature.102,103,104,105,106 Projecting cost reductions through experience for emerging technologies is a challenging task due to the limited historical data available for constructing experience curves. Existing guidelines recommend decomposing technologies into components and utilizing experience rates reported for identical or similar technologies.24,34,57,58,59,60 However, when such data are scarce, this method becomes inapplicable.

Accordingly, we introduce a conceptual framework for estimating experience rates based on the technology- and component-inherent characteristics of the technologies under evaluation. Innovation research has explored how cost reductions through innovation differ between technologies, suggesting that technology-inherent characteristics can help explain the variations in technology experience rates.53,107,108 Key factors include the complexity of a technology’s design52,53 and the need for customization,55 with simple and standardized technologies more likely to succeed and progress down the experience curve faster. We build on an existing technology typology that distinguishes technologies based on their design complexity and need for customization, providing insight into whether their anticipated experience rates are higher or lower.56

To apply this typology-based approach to DACCS, we follow a four-step process where first, the plant of a DAC technology is decomposed into components57,58,59,60 through a qualitative, exploratory analysis (Table S3), drawing from existing academic literature and expert interviews. Second, expert interviews are conducted to validate the division of DAC technologies into components and assess the design complexity and need for customization of the components and the overall systems (Figure 1B). In total, 33 interviews were conducted (Table S2) out of which 22 interviewees rated each technical component and system level along these complexity dimensions:

(1)

Design complexity: “1: simple,” “2: design-intensive,” and “3: complex”

(2)

Need for customization: “1: standardized,” “2: mass-customized,” and “3: customized”

All interviews were conducted under the “Chatham House Rule,” ensuring that neither the identity nor the affiliation of the speaker may be disclosed.109 Interviewees were experts involved in DACCS research or projects, representing both academia and industry. Experts were identified through academic publications and from lists of employees at DAC companies and relevant engineering firms in networks like LinkedIn. Interviews took place during March–December 2022, lasting between 45 min and 1 h, primarily via video conference, and were recorded if the experts agreed. A detailed description of the expert interviews is described in Note S3. For a detailed list of interviewees see Table S2. Third, based on the interview results (22 data points), components are arranged in a matrix according to their design complexity and need for customization (Figures 2A–2C) (numerical ratings from all interviews in Tables S4–S7 and further elaboration in Note S3). Fourth, the matrix position of each data point is compared with technology types outlined in innovation literature56 (Figure 1B) to select an experience rate that aligns with the technology type.

Component capacity and experience rates assigned

For the novel components we assume that experience will commence with the first commercial-scale DAC plant. Hence, we evaluate these components through expert interviews and assign experience rates based on ratings of the design complexity and need for customization (Figures 2A–2C). Non-novel components are deployed in several other industrial applications, where the appropriate experience base is not always clear110 or explicitly documented. Due to lack of such data, we assume that the experience process for all DAC plant components commences with the first commercial-scale plant; even non-novel components will be modified to suit the DAC system. We expect modest cost reductions for non-novel components in their DAC-specific use due to process improvements110 and economies of scale due to increased manufacturing output111 or procurement volumes. Consequently, we assign experience rates of 2.5% for the non-novel components, with an uncertainty range of 0%–5%, which aligns with prior literature examining cost reductions of non-novel components in offshore wind and CO2 capture technologies39,110,112 (details in Table S8). We assign experience rates to all remaining Ctoc cost components (Equation 2), based on the system-level experience rates derived for DAC technologies, reflecting experience-related cost reductions as “system integration” for new technologies combining process components in novel ways.58 Fixed OPEX experience rates follow the same approach. Variable OPEX learn independently at an experience rate of 2.5% (range 0%–5%),34 constrained by the thermodynamic minimum (Table S21).

Multi-component experience curves

To project the net removed cost of DACCS, we employ a multi-component one-factor experience curve,38 based on the assumption that accumulating experience in deployment leads to cost reductions driven by innovation in core components. We decompose capture technologies, transport, and storage into cost components (CAPEX—further broken down into (Ctoc), fixed OPEX, and variable OPEX) (Equation 2). Individual experience rates are then assigned to all TOC cost components, as well as fixed OPEX and variable OPEX, allowing the cost of CO2 net removed Ctotal to encompass all cost factors influenced by their respective experience rates. The overall cost relation can be expressed as the sum of the component costs of a DACCS system38

Ctotal(xt)=∑i=1nC0i(xtix0i)−bi

(Equation 4)

ERi=1−2−bi

(Equation 5)

Where index i represents a specific cost component, with C0i denoting the cost of component i at installed capacity 0 and xtix0i indicating the ratio of existing cumulative design capacity to initial cumulative design capacity of component i. Each component is characterized by a distinct experience parameter bi (Equation 4), derived from the experience rate ER of component i (Equation 5).

The potential for cost reduction varies for each component, indicated by distinct positive experience parameters bi. Hence, the aggregate experience rate will decrease over time as the relative cost shares of components with high experience rates decrease more quickly.72 To determine a system-level aggregate experience rate from these multi-component experience curves, we establish a linear relationship between costs at an earlier and an increased capacity.

The aggregate experience rate ER can