Anticipating a potential deficit in global carbon capture demand in 2030 despite benchmarking strategies

Abstract

Carbon capture, utilization, and storage (CCUS) technology is pivotal in climate mitigation but lags behind expectations. This investigation employs machine learning methods grounded in collaborative governance theory to analyze multiscale drivers and development models in global CCUS deployment. We observe a policy-driven predominant paradigm, with cost barriers significantly impeding economies of scale. Hierarchical clustering reveals three distinct typologies, coordinative, single-axis and constrained models, that illustrate a Matthew Effect, characterized by “major-power dominance and minor-nation catch-up”. Crucially, the Gini coefficient for CCUS development inequality persists at 0.70–0.84, exhibiting tripartite asymmetry through policy convergence, cost equilibrium, and technological agglomeration, alongside emergent spatial counter-agglomeration trends in recent years. Counterfactual analysis indicates that a comprehensive optimized strategy could boost historical growth by 22.7% and double capture scale by 2030. Nevertheless, a persistent one-third deficit in meeting climate targets underscores the urgency for multilateral governance mechanisms to implement more aggressive global actions.

Introduction

CCUS remains the critical technological pathway enabling low-carbon utilization of fossil fuels while critically supporting global climate goals^1,2. Current projections estimate its global carbon capture capacity could increase tenfold to 1.67 Gt by 2030 and 7.6 Gt by 2050 under net zero scenarios, contributing approximately 8% of cumulative emission reductions³. From the source capture to innovative utilization and secure storage, CCUS technology not only reduces atmospheric CO₂ concentrations but also inaugurates new avenues for resource recycling, possessing the potential to achieve cumulative emission reductions of hundreds of billions of tons by mid-21st century⁴. Since the 2019 G20 Summit, global CCUS deployment has accelerated at an unprecedented pace, particularly with a surge in projects after 2021. Given its immense environmental and economic value, urgent expansion of CCUS is essential to meet Paris Agreement targets and accelerate its transition from demonstration to large-scale commercialization^5,6.

Despite being widely recognized as a potential option for deep decarbonization, CCUS deployment remains insufficient to deliver meaningful emission reductions^7,8. The record-high global CO₂ emissions of 37.4 Gt in 2023 starkly contrast with current CCUS capture capacity, which accounts for less than 1% of total emissions⁹. The evolving CCUS value chain now includes over 300 projects worldwide, yet exhibits marked heterogeneity in both implementation rates and technological sophistication across nations¹⁰. This nascent development phase is characterized by uneven geographical distribution that fails to meet net-zero transition requirements¹¹. These disparities primarily originate from institutional deficiencies in certain regions, where inadequate policy frameworks and implementation mechanisms hinder progress¹². The lack of standardized technology hubs and knowledge-sharing platforms further impedes the formulation of cohesive international CCUS strategies¹³. Consequently, comprehensive large-scale deployment demands integrated solutions addressing three critical dimensions: policy frameworks, technological innovation, and economic viability^14,15.

The framework design should prioritize policy instruments such as direct fiscal incentives, targeted project funding, and tax credits^16,17, while meticulously assessing policy efficacy and potential implementation lag^10,18. The subdued projected returns for CCUS primarily stem from insufficient project transparency and short-term economic drivers¹⁹, fostering a pronounced misalignment between policy frameworks and market realities^20,21,22. Furthermore, advancements in CCUS technology and shifts in the competitive ecosystem exert a direct influence on carbon capture scalability^23,24,25. Critical determinants include technology readiness levels (TRLs), long-term sustainability, and external catalytic factors^26,27,28,29. Future CO₂ capture capacity will be predominantly shaped by technologies currently in the prototype or demonstration phases³⁰, with the synergistic integration of carbon capture and storage (CCS) and carbon capture and utilization (CCU) proving pivotal for accelerating technological maturation³¹. Within the CCUS value chain, CO₂ capture represents the most cost-intensive phase, typically constituting 60% to 80% of total expenditures^32,33. As stipulated by the U.S. Department of Energy (DOE), achieving commercial viability necessitates reducing carbon capture costs to approximately $30 per ton. These costs exhibit an inverse correlation with CO₂ concentration—higher concentrations generally yield lower capture expenses^34,35,36,37. Although economies of scale drive down capture costs, transitioning from high- to low-concentration emission sources may induce a twofold cost escalation^38,39.

Although CCUS research has surged with remarkable momentum in recent years, previous studies primarily concentrate on policy support analysis^22,40,41, business model design^14,42,43, and technical-economic evaluations^44,45,46 of storage/utilization solutions across different carbon-source scenarios. These evaluations typically assess emission reduction potential, economic feasibility, and technological maturity through a technological evolution perspective, yet have largely overlooked the mechanisms behind differentiated CCUS deployment performance from a technology management perspective, and causal analyses of heterogeneity in these drivers. Moreover, projections of future CCUS trajectories rely on linear extrapolations or the wholesale transplantation of experiences from early-adopting countries, failing to consider collaborative frameworks that could foster positive technological spillovers globally. Current global deployment practices predominantly follow a “leader-first, follower-imitative” pattern commonly referred to as the herd effect, which inadequately incorporates multi-factor synergistic mechanisms. To overcome the technological gaps and institutional bottlenecks hindering global CCUS technological system evolution, comprehensive research is urgently needed to map deployment patterns, identify structural barriers, and explore transnational policy-technology-cost frameworks.

To address these gaps, we first develop an integrated analytical framework combining Random Forest (RF), K-means clustering followed by principal component analysis (PCA), and Support Vector Machine (SVM) classification within a collaborative governance theory. This multi-algorithmic approach identifies various distinct global CCUS deployment typologies across policy, technology, and cost dimensions. We then systematically characterize the heterogeneous pathways driving CCUS deployment for the first time at a global scale, offering new interpretations of structural asymmetries in industrial development trajectories. Subsequently, we innovatively design counterfactual scenarios based on historical trends and empirical driving modes to evaluate both individual and synergistic effects of improvements in policy, technology, and cost factors. Collectively, this study provides critical insights and a robust empirical foundation for accelerating CCUS deployment worldwide.

Results

Global trends and Key indicators in CCUS development

Fig. 1a unveils the latest deployment landscape of operational CCUS projects worldwide. By 2024, 21 countries have active projects, with a total annual capture capacity of 0.172 Gt CO₂—an increase of 138.89% compared to 2013—underscoring the growing role of CCUS in climate mitigation. However, deployment remains disproportionately concentrated in North America, with the U. S. (possessing the largest scale) and Canada together accounting for over 75% of global operational scale. This reflects strong policy frameworks, industrial maturity, and integrated infrastructure that have enabled these countries to move from pilot phases to full-scale deployment. In contrast, projects in other regions remain fragmented and lack scalable implementation. Globally, 773 projects are in the pipeline (either under construction or planned), with North America (55.76%) and Europe (27.94%) leading, supported by strategic planning and institutional readiness. Meanwhile, Asia and the Middle East contribute just 11.77%, with projects remaining largely small-scale and disjointed, constrained by absent pipeline infrastructure and weak implementation mechanisms^47,48. Despite overall growth, many projects worldwide are still stuck in research or demonstration phases due to persistent policy gaps, technology limitations, and high costs.

Fig. 1: Global layout of carbon capture capacity and distribution of policy, technology, and cost (PTC) indicators.

a Global operational CCUS projects in 21 countries. b The distribution of three key PTC indicators of CCUS from 2013 to 2024. The quartile points of policy shift rightward reflecting positive accumulation, while patents and costs show leftward quartile shifts. c The scatter plot and fitted line graph depict the relationship between PTC and capture scale. The size of each point represents a country’s average capture scale, with darker colors indicating more recent years. The red shaded area denotes the 95% confidence interval.

Figure 1b presents the temporal dynamics of key indicators. Policy accumulation surged by 1.5 times in 2021 relative to 2020, with several countries introducing over 10 relevant policies. Technological indicators exhibited notable volatility between 2018 and 2022, followed by a period of relative stabilization. Meanwhile, the average CO₂ capture cost declined by 17.79% from 2013 to 2024, primarily reflecting economies of scale driven by expanded deployment. Figure 1c further reveals that countries with larger CCUS deployment scales tend to possess more advanced policy frameworks and stronger technological capabilities, even though their unit capture costs remain comparatively high. A strong positive relationship is evident between policy maturity and capture scale, as well as between technological progress and capture scale. In contrast, a very weak correlation is found between cost levels and capture capacity, suggesting that while cost factors have not been the main determinant of deployment to date, ongoing cost reductions remain essential to ensure the long-term economic viability of CCUS systems.

Driving factors for global CCUS development

The global deployment of CCUS manifests pronounced cross-national disparities. As shown in Fig. 2a, the USA, CAN, and BRA constitute the first tier, demonstrating strong external influence as reflected by their contribution to global carbon capture capacity. A second tier—including the UK, AUS, CHN, NOR, and nine others—exhibits SHAP values above the external influence trendline yet below the global average, suggesting notable but not yet prevailing diffusion capacity. Despite moderate capture volumes, these countries augment their external influence through targeted policy instruments (e.g., cluster strategy in the UK) and technological strengths (e.g., carbon utilization in CHN; deep storage in NOR). In contrast, several MENA and European countries fall below or well beneath the trendline, indicating limited outward influence. To elucidate the underlying mechanisms of national stratification and pronounced deviations, we apply SHAP value decomposition (Fig. 2b, c), revealing structural contributions of 44.80%, 37.82%, and 17.38% from policy, technology, and cost, respectively—yielding an approximate 6:5:2 ratio.

Fig. 2: Drivers of CCUS scaling up across countries.

a National CCUS development impact on global deployment. “Size of influence” represents the contribution of a country’s carbon capture capacity to global carbon capture capacity (excluding the country itself) based on its SHAP value. b Average contributions of policy, technology, and cost indicators. c Distribution of contributions from policy, technology, and cost. d Effect of policy intensity on capture capacity. e Effect of technological progress on capture capacity. f Effect of cost changes on capture capacity and its interactions. g Country-level contributions of policy, technology, and cost. h Proportional relationships of policy, technology, and cost contributions across countries.

This structural weight is consistent with the interaction patterns observed in Fig. 2d–f. Figure 2d indicates that policy variables rapidly increase SHAP values after surpassing an intensity threshold (quantitatively identified at 12.00 via a piecewise linear regression method), affirming their primacy as key drivers. Notably, this positive effect persists even under high-cost conditions (transitioning from green to yellow). During the early stages of emerging technologies like CCUS, policy should primarily focus on establishing market demand foundations to support future large-scale deployment⁴⁹. Even under high-cost conditions, strong incentive signals and targeted funding can enable a shift toward initial deployment, subsequently reducing costs. A typical example is the U.S. Inflation Reduction Act (IRA), utilizing substantial tax credits to accelerate deployment despite high costs. Figure 2e highlights a nonlinear amplification of technology’s marginal contribution after surpassing critical thresholds, accompanied by declining costs, underscoring the dual role in scaling and cost mitigation. Indeed, the system-wide cost reduction potential of CCUS thus hinges on technological breakthroughs and learning effects^50,51. Figure 2f further demonstrates that cost variables predominantly exert a negative influence, turning positive only under conditions supported by advanced technology (yellow regions).

In fact, this structural configuration is far from globally uniform; rather, it predominantly mirrors the deployment trajectories of leading nations. Figures 2g, h systematically delineate cross-national variations in the relative contributions and internal composition of key drivers. The results reveal that key contributions from countries like USA, GBR, CAN, CHN, and JPN demonstrate the strongest policy-led contributions, consistent with their substantive policy frameworks in reality, such as the UK’s roadmap planning nine large offshore and onshore CCUS clusters; North America’s tax credit incentives (e.g., 45Q); China’s demonstration projects driven by its “dual-carbon” strategy; and Japan’s emphasis on carbon recycling and hydrogen-based conversion pathways. From a technological perspective, USA, CHN, and NOR lead globally, reflecting deep-rooted innovation ecosystems and sustained patent output. Cost-related factors also exert notable influence in USA (17.80%), CHN (28.24%), and JPN (25.86%), shaped in part by comparatively lower labor costs in parts of Asia and the economic efficiencies of large-scale deployment in the U.S. context.

Structurally (Fig. 2h), approximately 60% of countries fall within a technology-driven contribution band of 0.35–0.65, suggesting the widespread adoption of a globally converging knowledge-driven deployment paradigm. NOR exhibits the highest proportion of technological contribution at 53.46%, exemplifying a technology-centric model anchored in standardized, transparent storage infrastructures. The U.S. presents a balanced profile, with policy, technology, and cost contributions at 47.66%, 34.55%, and 17.80%, respectively, resembling Norway’s structure and reflecting commercial deployment consistency. By contrast, China’s policy-driven proportion of 39.52% is lower, ranked 18th, largely reflecting its reliance on voluntary corporate actions in the absence of robust fiscal subsidies or formalized incentive mechanisms.

Development models and driving mechanism of CCUS industry

To further validate the observed heterogeneity, we employ PCA-K-means clustering to identify typological constraints shaping cross-country CCUS deployment models. In Fig. 3a, CCUS development models are categorized into three typical pathways. The first is a policy-led, single-axis model (e.g., GBR), marked by frequent policy interventions and institutional steering but lacking in technological depth and cost efficiency. The second is a coordinated, dual-driven model (e.g., USA, CHN), characterized by strong technological innovation and divergent policy instruments—market-based mechanisms in the U.S. (e.g., tax and carbon credits), versus state-led industrial integration in China. The third is a constrained model, subdivided into three sub-clusters. Subset 0 (e.g., ARE, BEL, NOR, DEU) is defined by high deployment costs despite existing policy or technological frameworks, primarily due to limited market size and infrastructural deficits; Subset 1 (e.g., CAN, JPN) reflects policy-cost coordination, with strong policy support and manageable costs driving stable growth, despite moderate technological levels; Subset 2 (e.g., BRA, AUS, QAT, KWT et al.), exhibiting low-level technology-driven paths mainly reliant on oil and gas companies’ associated capture during enhanced oil recovery (EOR), limited core technological accumulation, external research and development (R&D) dependence, and lacking systematic deployment and policy support.

Fig. 3: Development model and driving mechanism of global CCUS industry.

a PCA-Based K-means clustering on the three driving factors for 21 countries. In primary cluster, Principal Component 1 is driven by technology and cost, while Principal Component 2 is influenced by policy in Table S3. In Cluster 1, sub-clustering shows Principal Component 1 linked to policy, and Principal Component 2 to technology and cost. b SVM Classification of Carbon Capture Capacity. The SVM model accurately separates countries into low (red) and high (blue) carbon capture capacity groups. c Analysis of the driving mechanism of carbon capture capacity.

Here, we further elucidate the underlying mechanism of policy-driven dynamics by linking the driving factors to carbon capture scale at the global level. Figure 3b presents a binary classification of CCUS capture capacities, elucidating the underlying mechanisms driving national differentiation (in Table S4). The high-capacity countries include Australia (2022–2024), Canada (2020–2024), China (2021–2024), the United Kingdom (2024), Japan (2017–2024), the Netherlands (2024), Norway (2020–2024), New Zealand (2024), Oman (2024), Qatar (2019–2024), and the United States (2018–2024), have progressed from pilot demonstrations to scalable deployment. By contrast, low-capacity countries (e.g., BRA, SAU) exhibit persistent deficits in policy continuity, technological efficiency, and cost management, lacking the systemic foundations necessary for large-scale implementation.

Figure 3c delineates the intricate mechanisms underpinning the expansion of carbon capture capacity. At the global level, both robust policy support (p < 0.05) and significant technological advancement (p < 0.001) exhibit a pronounced positive effect on capture scale, affirming their central roles as pivotal enablers of CCUS proliferation. Across both country groups, the volume of technology-related patent filings consistently serves as a significant catalyst for deployment, irrespective of national capacity levels. This underscores the universal relevance of technological innovation as a foundational pillar in CCUS development pathways.

By contrast, the influence of policy and cost factors reveals clear structural asymmetries. In low-capacity countries, policy support (p < 0.05) emerges as a decisive determinant, playing an indispensable role in catalyzing early-stage deployment. The absence of institutional maturity in these contexts amplifies the importance of state intervention as a compensatory mechanism. Conversely, in high-capacity countries, the influence of policy is subject to diminishing marginal returns across different stages of deployment. To facilitate the transition from large-scale infrastructure build-out to sustained market integration, more sophisticated economic instruments—such as capital subsidies, carbon pricing regimes (e.g., carbon taxes and emissions trading), carbon credit mechanisms, and contracts for difference—will be imperative to maintain investor confidence and deployment momentum. Nevertheless, such market-oriented framework remains largely confined to a handful of frontrunner nations, with limited diffusion and coordination at the global level—presenting a persistent governance gap in scaling CCUS beyond national boundaries.

Counterintuitively, a significant positive correlation emerges between cost levels and carbon capture scale at the global level, with this effect particularly evident among high-capacity countries (p < 0.05). This finding directly challenges the conventional assumption of economies of scale, which posits that unit costs should decrease as project scale expands. The capture cost often rises with the increase of deployment capacity, which indicates that CCUS has not yet reached the expected stage of economies of scale. This counterintuitive relationship underscores two fundamental insights. First, large-scale deployments in high-capacity nations are predominantly driven by capital-intensive pathways—most notably, negative emissions technologies such as Direct Air Carbon Capture and Storage (DACCS), Bioenergy with Carbon Capture and Storage (BECCS), and deep geological sequestration—all of which substantially elevate average cost levels. Second, the prevailing level of technological maturity in most countries remains limited to the demonstration stage, constraining the emergence of systematic learning effects and the development of economies of scale. By contrast, in low-capacity countries, the cost factor exerts a negative but statistically insignificant impact, indicating that CCUS deployment remains primarily concentrated in the single segment of carbon capture, without industrial chain integration or large-scale expansion. Overall development remains in the exploratory phase, making it difficult to approach the critical threshold for transitioning from high-concentration to low-concentration sources.

Inequality in CCUS deployment across countries

To thoroughly reveal the complex driving mechanisms, it is essential to examine how the pronounced imbalance in global CCUS deployment constrains its diffusion capacity. A comprehensive analysis of CCUS development dynamics and trends from 2013 to 2024 reveals significant disparities among countries, highlighting the striking inequality in CCUS development. As shown in Fig. 4a, the overall and inter-group Gini coefficients of global carbon capture capacity have persistently remained between 0.70 and 0.84, exhibiting a notable upward trend since 2020. This change indicates that while a few countries are rapidly scaling up deployment, most nations remain at low development levels, demonstrating a typical Matthew Effect. Currently, global CCUS development presents a structural imbalance characterized by simultaneous “concentration at the top” and “marginalization at the bottom”.

Fig. 4: Development inequality index of the global CCUS industry.

Overall and between-group Gini Coefficients a CCUS Scale, b Policy, c Technology, and d Cost. Within-group Gini Coefficient by each continent e CCUS Scale, f Policy, g Technology, and h Cost.

Figures 4b–d show that in the policy dimension, the Gini coefficients of policy support at global and continental levels generally exhibit a declining trend, approaching 0.5 by 2024. This suggests the global policy environment is gradually converging, with increasing numbers of countries enhancing CCUS support through climate policies. However, this downward trend experiences a temporary rebound in 2023. On one hand, this indicates that while the inclusion of emerging countries has improved coverage, their fragmented policies and weak implementation capacity limit policy effectiveness. On the other hand, high-capacity countries like the United States and Norway have taken the lead in exploring commercial deployment, with more market-oriented policy support creating structural bias against countries on a small scale and exacerbating policy distribution imbalances. At the technological level, the technology Gini coefficient continues to rise, reaching 0.8 by 2024, reflecting the high concentration of CCUS-related patents and technical capabilities in a few nations. High-capacity countries have accumulated technological advantages through early deployment experience, while low-capacity countries primarily rely on “purchased” or “open-access” technology transfers^52,53,54, resulting in global technological disparities. In contrast, the cost Gini coefficient remains relatively low (consistently below 0.3). However, cost inequality has begun to widen gradually, indicating that while a few countries have achieved significant cost optimization, carbon capture costs remain generally high worldwide.

Intra-continental disparities in CCUS deployment have widened across most regions, with North America as the only exception showing a convergence trend (Fig. 4e–g). While persistent structural imbalances—marked by concentrated leadership and peripheral underdevelopment—drive inequality in policy and technology, North America’s decline in disparity reflects transboundary policy coordination and joint infrastructure investment, exemplified by U.S.–Canada project integration, harmonized subsidies, and shared pipeline networks. Cost gaps remain relatively limited (Fig. 4h) but are rising in resource-rich areas, such as the Arabian Peninsula, where carbon-intensive industries and divergent strategies have amplified inequality. This reflects a distinct “resource-first mover” advantage⁵⁵, as countries like Saudi Arabia, Qatar, and Kuwait leverage international governance engagement and multinational CCUS initiatives to strengthen both project economics and geopolitical influence.

Global spillover effects in CCUS deployment

Despite pronounced disparities in national technological innovation capacities, as revealed by the inequality analysis of CCUS development and the intensifying polarization of technological advancement, the global trajectory of CCUS deployment remains predominantly policy-driven rather than technology-led. As demonstrated in Table 1, prior to 2021, the spatial effects of policy, technology, and cost drivers on CCUS deployment scales are not statistically significant. However, from 2021 to 2024, Moran’s I indices exhibit consistently negative and statistically significant values, revealing pronounced negative spatial spillover effects. These results indicate that recent patterns in global CCUS deployment are increasingly defined by spatial dispersion rather than agglomeration, underscoring a growing trend toward geographic fragmentation.

Table 1 Spatial autocorrelation test results for CCUS spillover effects

The simultaneous rise in policy and technological inequalities, coupled with spatial dispersion rather than agglomeration, can be attributed to the nonlinear structural characteristic of “policy convergence and technological polarization”. Current global CCUS deployment faces a dual bottleneck. On one hand, most countries lack systematic policy planning and effective market incentive mechanisms, and global deployment standards remain fragmented, leading to unclear development paths. On the other hand, the fundamental barrier arises from technological constraints. Low-capacity countries generally lack local R&D capabilities, engineering infrastructure, and industrial chain support, hindering effective absorption of external technological achievements. For instance, Australia’s Gorgon project underperformed due to insufficient preliminary research; India and South Africa face longstanding limitations from high costs, low efficiency, and inadequate infrastructure; Saudi Arabia, Qatar, Singapore, and Turkey heavily rely on oil and gas multinational cooperation due to insufficient domestic capabilities; countries like Indonesia and the Philippines remain at research stages due to inadequate geological assessments and financing mechanisms.

Despite participating in international cooperation or demonstration projects, these countries struggle with effective replication and diffusion due to inadequate institutional support and local absorptive capacity, exemplifying typical structural technological barriers severely limiting spatial synergy and agglomeration potential. Therefore, policy convergence has not translated into deployment agglomeration, and technological barriers significantly inhibit positive spatial spillover effects, reinforcing the Matthew Effect whereby leading nations advance rapidly while smaller nations lag behind, simultaneously resulting in pronounced spatial dispersion and polarized deployment patterns, exacerbating the technological gap within the global CCUS deployment system.

Counterfactual analysis based on driving mechanisms

To systematically explore scalable pathways for enhancing carbon capture capacity, we construct four counterfactual scenario archetypes that incorporate heterogeneous national contexts and apply targeted improvements to the compound annual growth rates (CAGRs) of key drivers^56,57,58. These scenarios reflect distinct optimization logics, from single-factor adjustments to integrated enhancement strategies, enabling a structured assessment of development trajectories under varying country conditions (Fig. 5a–c). Policy benchmarks are calibrated using CAGRs of JPN and USA for developed economies, with the U.S. serving as the regional reference for the Americas (e.g., CAN), and regional means employed across Asian contexts. Technological reference trajectories are derived from the observed CAGRs of SAU, BRA, NOR, and USA, whereby technological frontrunners (e.g., USA, CHN) retained their existing growth trajectories, while others are constrained by empirically grounded and regionally feasible enhancement rates. Cost baselines are anchored to the most substantial capture cost reductions recorded in AUS, CAN, and CHN, with further adjustments accounting for heterogeneity in resource endowments and agglomeration effects (see Supplementary Notes and Table S5).

Fig. 5: Carbon capture potential based on historical improvements.

Counterfactual scenario a policy, b technology, and c cost. Based on the SVM model classification, countries are divided into two categories, with counterfactual analysis performed for each group. Counterfactual improvement outcomes under d policy-Only change, e technology-Only change, f cost-Only change, and g all changes. The region circled in red in d represents the high capture capacity countries as classified by the previously applied SVM model. h Historical and Counterfactual scenario forecasts.

Global CCUS expansion remains uneven, with low-capacity countries showing far greater incremental potential from policy, technology, and cost improvements than high-capacity counterparts. Between 2013 and 2024, cumulative capture reached 0.9 Gt CO₂. Policy gains (Fig. 5d) are modest overall (5.28%) but more than double in low-capacity countries (7.69% vs. 3.58%). Technology (Fig. 5e) is the most decisive factor (17.71%), again yielding higher marginal returns in low-capacity systems (22.29% vs. 14.48%). Cost reductions (Fig. 5f) contribute globally (5.82%), yet effects are minor in high-capacity countries (1.60%) compared to lower-capacity peers (11.80%). Under the comprehensive scenario (Fig. 5g), gains reach 28.57% in low-capacity versus 18.53% in high-capacity countries. These disparities reflect not only resource endowments but also asymmetric policy responsiveness and absorptive capacity. For high-capacity countries, diminishing technology returns shift the frontier toward cost-driven expansion, hinging on financial instruments such as tax credits, carbon pricing, and contracts for difference.

Technological enhancement remains the dominant driver of CCUS expansion, while policy and cost factors exert limited influence due to weak incentives and scale inefficiencies. Between 2021 and 2024, Canada and Japan saw the strongest gains from technology improvements (Fig. 5e). Cost reduction scenarios provided some benefits but rarely exceeded 2 Mt CO₂ annually (Fig. 5f). Policy responsiveness was similarly weak, as most frameworks lacked targeted CCUS support; in China, for example, 2022–2024 policies encouraged demonstration projects, yet firms largely favored carbon markets over CCUS deployment. The integrated scenario raised capture capacity by 22.7% (Fig. 5g), doubling the baseline projection for 2030 (1 Gt vs. 0.5 Gt). Still, this remains below the IEA’s 1.67 Gt target, leaving a 0.53 Gt gap (Fig. 5h). In effect, current improvements unlock only one-third of the potential capacity, insufficient to meet 2030 goals and threatening the achievement of 2040 and 2050 climate targets, underscoring the need for far more ambitious strategies from leading nations.

Discussion

The global deployment of CCUS is marked by spatial heterogeneity, reflecting a Matthew Effect that necessitates differentiated pathways tailored to national capacities. Leading nations (e.g., USA, CHN, GBR, NOR) are mobilizing substantial fiscal resources (e.g., UK’s £20 billion package, the U.S. Infrastructure Act) and diversified market instruments such as carbon trading platforms and contracts for difference (CFDs) to accelerate the transition from demonstration to commercial-scale deployment and consolidate their leadership in standard-setting. By contrast, lower-capacity countries (e.g., IDN, SAU) face financial constraints and must rely more heavily on multilateral partnerships and international climate funds. In practice, these frameworks are being operationalized through concrete cases: the Council for Geoscience-Sasol collaboration in South Africa explores CCUS feasibility with Green Climate Fund support; the Asia-Pacific Partnership (APP) has advanced pilot projects, capacity-building, and phased regulatory design; Mexico’s engagement with the Pacific Alliance signals regional carbon pricing momentum; while delays in Kazakhstan’s carbon market and South Africa’s carbon tax reveal persistent institutional and fiscal hurdles.

The “scale disadvantage” of smaller nations exacerbates structural asymmetries in global CCUS deployment, underscoring the need for strengthened international cooperation to bridge technological and equity gaps. While multilateral mechanisms such as the IEA CCUS Network, CSLF, and GCCSI have established an initial collaborative framework, advanced economies should prioritize solvent performance improvements, direct air capture technologies, long-term storage monitoring, and risk-sharing arrangements such as long-term service contracts and government-backed liability frameworks. Even at the corporate level, partnerships like the one between Shell’s carbon capture expertise and Technip Energies’ engineering capabilities show how complementary alliances can share risks and speed up projects. Developing countries further face financing, regulatory, and data transparency barriers in accessing global technology networks. Moving forward, “North-South dialogue” should promote core technology spillover through cross-border carbon transport agreements and open sharing of expertise and databases, bolstering local absorption capacity. Simultaneously, “South-South cooperation”—exemplified by initiatives such as AZEC and Sinopec’s innovation platform—can enhance synergies through shared infrastructure and equipment manufacturing. Mid-level powers like China and Brazil can act as strategic intermediaries, using initiatives such as the Belt and Road to promote local technology adaptation and industrial integration.

The transition of CCUS policies from “powerful catalyst” to “placebo” reflects a paradigm shift from external policy-drive to internal technology-driven development. To achieve large-scale deployment by 2030, priority should be given to technologies above TRL 6 (e.g., amine absorption, salt cavern storage), while strategically incorporating experimental (TRL 4–5) and early-stage (TRL 1–3) innovations to establish a robust R&D pipeline. Notably, modular solutions such as Technip Energies’ “Canopy by T.EN” demonstrate how standardized designs can accelerate deployment through reduced costs and enhanced scalability. Crucially, CCUS must evolve beyond single-point deployment toward system-oriented frameworks characterized by clustering and integration. Leading initiatives—such as U.S. hubs (Midwest Express, Bayou Bend), the UK’s Industrial Cluster Plan, and Norway’s Northern Lights—demonstrate the efficacy of hub-and-cluster architectures with multi-source capture, shared transport, and coordinated storage. Future advancement hinges on harmonized progress across three dimensions: standardized CO₂ transport protocols, modular storage platforms, and intelligently dispatched multi-source capture systems. Anchored in technological maturity and economic viability, CCUS innovations serve to enhance inter-facility connectivity, enable regional replication, and boost chain-coupling efficiency.

Although non-technical barriers are not this study’s primary focus, they significantly constrain large-scale CCUS deployment. Ambiguities in post-storage liability—covering long-term monitoring, leakage compensation, and remediation—erode investor confidence and deter private investment. Concurrently, low public awareness of storage safety undermines social acceptance. Norway’s Sleipner project exemplifies best practice through transparent data sharing and proactive risk communication. Cross-regional CO₂ transport and storage also frequently trigger “Not-in-My-Backyard” (NIMBY)⁵⁹ resistance. Integrating Social License to Operate (SLO)⁶⁰ into early project approval may enhance stakeholder trust. In China, despite technical standards such as GB/T 42797-2023, the lack of dedicated legislation and formal industry classification results in regulatory gaps. Current oversight relies on the Oil and Gas Pipeline Protection Law, which fails to address CCUS-specific risks. Exclusion from the National Industrial Classification further creates institutional bottlenecks during permitting. These issues reflect broader regulatory fragmentation, highlighting the need for a coordinated legal framework, potentially under a national climate law⁵⁵. Future research could examine how social acceptance affects deployment costs, compare risk premiums across liability regimes, and evaluate the adaptability of SLO frameworks in different contexts.

This study has several limitations in terms of data and methodology: first, inconsistencies in data completeness, reporting frequency, and project status definitions may introduce unobservable errors, as unreported regions and inconsistent classifications persist even after cleaning; second, time lags constrain explanatory power, since policies and patents often take years to affect capture capacity, while recent surges may reflect expectations rather than realized deployment; third, model optimization is limited, as parameter searches covered only a few values of n_estimators and max_depth, and outcomes remain sensitive to sample splits (1:4 train–test ratio), leading to variation in parameter choice and accuracy; finally, the transferability of counterfactual scenarios is restricted, as they assume comparability across institutional and geological contexts, whereas real-world improvements are constrained by supply chains, CO₂ transport and storage siting, labor, and financing frictions.

Methods

Indicator selection

All data employed in this study are obtained from authoritative and publicly available sources, primarily integrated from IEA CCUS Projects Database, the Clean Air Task Force (CATF), and the China Annual Report on Carbon Capture, Utilization, and Storage (2023). These datasets provide comprehensive statistics on operational CCUS projects globally from 2013 to 2024, including key indicators such as the number of CCUS-related policies, patent counts, and average carbon capture costs by country. To enhance clarity regarding the data structure and source attribution, Table 1 summarizes all variables and their corresponding data origins. In this table, P denotes policy support intensity, T represents the number of CCUS-related patents, and C refers to average capture costs. Detailed information on data collection procedures, variable definitions, and indicator processing methodologies is provided in the Supplementary Notes and Table S1.

SHAP Interpretability

Random Forest Regression is employed to model the complex, nonlinear relationships among policy, technology, and cost variables and national carbon capture capacity, providing a robust and interpretable predictive foundation under conditions of high-dimensional interaction complexity^61,62,63. We evaluate the performance of four commonly used regression algorithms—Random Forest Regression (RFR), Support Vector Regression (SVR), Gradient Boosting Regression (GBR), and Lasso Regression—using a consistent data partitioning strategy and six-fold cross-validation. Model performance is assessed with error metrics including MSE, RMSE, and MAE. (see Table S2).

The results show that both SVR and Lasso exhibit significant prediction errors. Specifically, SVR has a testing RMSE of 16.12 and MAE of 5.62, indicating limited ability to capture complex nonlinear patterns. Lasso performs the worst, with a testing RMSE of 25.99 and MAE of 13.54, showing severe underfitting. While GBR achieves a low training RMSE of 0.52, its testing RMSE remains at 8.49, suggesting overfitting and limited generalizability. In contrast, RFR delivers the most balanced performance, with a testing RMSE of 11.84 and MAE of 4.66, significantly outperforming SVR and Lasso. RFR maintains stable error metrics and shows no signs of overfitting, as indicated by consistent variance across folds. Overall, RFR is the most reliable model, particularly well-suited for capturing the nonlinear relationships in complex CCUS datasets. To further improve performance and generalizability, we performed systematic hyperparameter optimization using grid search, with parameters including the number of estimators (10, 50, 100), max depth (None, 5, 10), minimum samples to split a node (4, 6, 8), and minimum samples per leaf (2, 3, 4), evaluated by negative mean absolute error. After model fitting, we implement SHAP analysis to interpret model outputs. We have carefully examined the data and confirm that no significant multicollinearity exists among the variables. The SHAP analysis decomposes model predictions and quantifies the marginal contribution of each explanatory variable, enabling a transparent assessment of the dominant deployment drivers and their relative influence across national contexts. This method provides both local and global interpretability by quantifying each variable’s marginal contribution to the prediction, as described in Eqs. (1) and (2):

$$\varphi (f,x)=\mathop{\sum }\limits_{{x\in c}_{x}}\frac{|z|!(M-|z|-1)!}{M!}|{f}_{x}(z)-{f}_{x}(z|i)|$$

(1)

$${I}_{j}^{SHAP}=\frac{1}{N}\mathop{\sum }\limits_{j=1}^{N}|\varphi (f|{x}^{(i)}|)$$

(2)

where ${\varphi }_{i}(f,x)$ represents the SHAP predicted value of the CCUS capture scale,$\,f$ is the machine learning model, $x$ is the machine learning model,$\,z$ is a subset of the predictive variables, $M$ is the number of predictive variables. $j$ and $i$ represent the input variables and data samples, respectively, and N is the number of samples.

More broadly, SHAP has emerged as a standard tool in interpretable machine learning, enabling both local and global explanation of complex model behavior. In Eq. (1), SHAP values reveal how each input feature shifts a specific prediction away from the model’s baseline (expected value), thereby enabling individualized explanation of predictions. In Eq. (2), overall feature importance can be assessed by averaging the absolute SHAP values across all samples, while dependence plots help visualize marginal effects and feature interactions. This level of interpretability facilitates the identification of key drivers, reveals potential model biases, and helps detect overfitting or underfitting, ultimately supporting feature engineering, model refinement, and hyperparameter optimization.

K-means clustering based on principal component analysis (PCA)

PCA combined with K-means clustering facilitates dimensionality reduction and identifies three distinct typologies of CCUS development pathways, revealing cross-national heterogeneity in policy–technology–cost configurations and underlying structural mechanisms. To enhance the robustness and interpretability of the clustering analysis, we conduct systematic data preprocessing and dimensionality reduction prior to cluster modeling⁶⁴. Given the variation in scales and the results of initial feature importance assessments, all variables are standardized using z-score normalization to ensure comparability and improve the effectiveness of subsequent Principal Component Analysis (PCA). PCA is applied to two datasets: the full sample set, to examine global structural patterns, and the Cluster 1 subset identified through preliminary K-means clustering, to explore internal heterogeneity within that subgroup. Meanwhile, the optimal category k = 3 of the two clusters is determined by the elbow plot method. In both cases (see Table S3), the cumulative variance explained by the first two principal components reaches 92.34% and 87.57%, respectively—well above the commonly accepted threshold of 80%⁶⁵. This indicates a successful reduction of dimensionality while preserving the essential structure and dominant information content of the original datasets.

The method is formally described as follows in Eq. (3):

$$J={\mathop{\sum }\limits_{j=1}^{k}\mathop{\sum }\limits_{i=1}^{n}\mathop{\sum }\limits_{d=1}^{D}({x}_{jd}^{(j)}-{C}_{jd})}^{2}$$

(3)

where ${x}_{{jd}}^{(j)}$ represents the coordinate of data point $i$ in dimension$\,d$, ${C}_{{jd}}$ is the coordinate of the cluster center ${C}_{j}$ in dimension $d$, while $D$ represents the total number of dimensions for data points and cluster centers.

Supervised support vector machine (SVM) classification

SVM is applied to classify countries based on capture performance and to evaluate whether stratified deployment outcomes correspond to systematic differences in the strength and composition of underlying drivers. We employ the SVM model to classify data by defining labels based on median capture capacity and identifying the optimal separating hyperplane. The primary goal of SVM is to maximize the margin between support vectors and the hyperplane, optimizing the weight vector w and bias term b. The linear kernel is used for linearly separable data, with the regularization parameter C optimized through grid search. The voting ensemble method further improves classification accuracy, achieving a training accuracy of 1.0 and a test accuracy of 0.91. The mathematical formulation of the model is presented in Eqs. (4) and (5).

$$\min \,\frac{1}{2}{\Vert w\Vert }^{2}+C\mathop{\sum }\limits_{i=1}^{N}{\varepsilon }_{i}$$

(4)

$$s.t.{y}_{i}({w}^{T}{x}_{i})\ge 1-{\varepsilon }_{i},{\varepsilon }_{i}\ge 0,i=1,\ldots ,N$$

(5)

where $w$ is the weight vector, $x$ is the input feature vector, and $b$ is the input feature vector, and ${y}_{i}$ is the input feature vector, and ${x}_{i}$ is the input feature vector. C is the regularization (penalty) parameter, and ${\varepsilon }_{i}$ is the classification error term.

Gini coefficient calculation

The Gini coefficient is used to measure inequality in CCUS deployment across multiple dimensions, scale, policy, technology, and cost. at both global and continental levels, thereby identifying the most unevenly distributed mechanisms and their temporal evolution. In this study, we utilize the Gini coefficient to assess disparities in the development of the CCUS industry, building on its application in energy systems. This approach provides a comprehensive understanding of the distribution of CCUS technology deployment across different regions and socio-economic groups, as demonstrated in Eqs. (6–12).

$$G=\frac{{\sum }_{j=1}^{k}{\sum }_{hr=1}^{k}{\sum }_{i=1}^{nj}{\sum }_{r=1}^{nh}|{y}_{ji}-{y}_{hr}|}{2{n}^{2}\mu }$$

(6)

$${G}_{ij}=\frac{1}{2\bar{{y}_{j}}}\left(\mathop{\sum }\limits_{i=1}^{{n}_{i}}\mathop{\sum }\limits_{r=1}^{{n}_{j}}|{y}_{ji}-{y}_{jr}|\right)$$

(7)

$${G}_{jh}=\frac{({\sum }_{i=1}^{{n}_{j}}{\sum }_{r=1}^{{n}_{h}}|{y}_{ji}-{y}_{hr}|)}{{n}_{j}{n}_{i}(\overline{{y}_{i}}+\overline{{y}_{h}})}$$

(8)

$${G}_{w}=\mathop{\sum }\limits_{i=1}^{{n}_{j}}{G}_{jj}{P}_{j}{S}_{j}$$

(9)

$${G}_{nb}=\mathop{\sum }\limits_{j=2}^{k}\mathop{\sum }\limits_{h=1}^{j-1}{G}_{jh}({P}_{j}{S}_{h}+{P}_{h}{S}_{j}){D}_{jh}$$

(10)

$${G}_{t}=\mathop{\sum }\limits_{j=2}^{k}\mathop{\sum }\limits_{h=1}^{j-1}{G}_{jh}({P}_{j}{S}_{h}+{P}_{h}{S}_{j})(1-{D}_{jh})$$

(11)

$$G={G}_{w}+{G}_{nb}+{G}_{t}$$

(12)

where $G$, ${G}_{jj}$, and ${G}_{jh}$ represents denote overall, between-group, and within-group Gini coefficient, respectively, ${G}_{W}$, ${G}_{nb}$, and ${G}_{t}$ represent the contribution portions of between-group, within-group and hypervariable density. Variables such as ${y}_{i}$ and y, compare values, while $n$ indicates group size. Proportions like ${P}_{j}$, ${S}_{j}$, and ${D}_{j}$ capture the differences within and between groups, facilitating a detailed analysis of disparities in CCUS technology deployment across various groups.

Moran’s I index measurement

Moran’s I is employed to assess spatial dependencies in CCUS deployment patterns, identifying whether recent global developments exhibit clustering, dispersion, or fragmentation, and thus uncovering emerging spatial disintegration trends. Moran’s I index assesses spatial autocorrelation⁶⁶, ranging from −1 (dispersion) to 1 (clustering), with 0 indicating random distribution. To calculate it, we require the geographic locations and CCUS capture scales for each region. Moran’s I for policy, technology, and cost factors helps evaluate spatial distribution, revealing regional associations and inequalities, and informing policy-making. The formula for Moran’s I can be expressed in Eq. (13):

$$I=\frac{n}{{\sum }_{i=1}^{n}{\sum }_{j=1}^{n}{w}_{ij}}\ast \frac{{{\sum }_{i=1}^{n}{\sum }_{j=1}^{n}{w}_{ij}\ast {Z}_{i}\ast Z}_{j}}{{\sum }_{i=1}^{n}{z}_{i}^{2}}$$

(13)

where $n$ is the number of spatial units, ${w}_{{ij}}$ denotes the spatial weight between units i and j, ${Z}_{i}$ and ${Z}_{j}$ are the values of the variable.

Prophet model selection

The Prophet model is utilized to simulate the evolution of national carbon capture capacity through 2030, incorporating historical trajectories and cross-country heterogeneity in driver configurations to assess the differentiated impacts of policy, technology, and cost interventions. This model is well-known for its ability to handle complex time series components, including trends, seasonality, and structural changes, while being flexible enough to account for irregular events such as holidays and multiple changepoints. The model demonstrates consistent performance across different scenarios, with R²values ranging from 0.965 to 0.982. The Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) remain relatively low across all scenarios, suggesting minimal dispersion between predicted and actual values (in Table S6). Furthermore, the Mean Bias Error (MBE) is exactly zero in each case, indicating the absence of systematic over- or under-estimation, as expressed in Eqs. (14–16):

$$y(t)=g(t)+s(t)+h(t)+\varepsilon (t)$$

(14)

where $y(t)$ represents the observed value, in this case, CCUS capture scale (t/a), $g(t)$ is the long-term trend component, $s\left(t\right)$ is the seasonality component, $h\left(t\right)$ accounts for holiday effects, and $\varepsilon \left(t\right)$ is the error term. In this study, the model is configured with yearly seasonality and a custom seasonality component (a 10-year period) to capture non-standard cyclical patterns.

The formula $g(t)$ is a piecewise linear trend model used to capture long-term changes in a time series. $k$ represents the growth rate, $a\left(t\right)$ is a time function, $\delta$ is the adjustment rate, m is the migration parameter, and $\gamma$ ensures the continuity of the trend at structural turning points. By setting changepoints at specific times (e.g., January 1, 2023), we enable the model to reflect structural changes caused by historical events.

$$g(t)=(k+a(t)\ast \delta )+m(t+a(t)\ast \gamma )$$

(15)

To handle seasonality in the data, a Fourier series is applied to model periodic variations:

$$s(t)=\mathop{\sum }\limits_{i=1}^{n}\left({a}_{n}\,\cos \left(\frac{2\pi nt}{P}\right)+{b}_{n}\,\sin \left(\frac{2\pi nt}{P}\right)\right)$$

(16)

where $P$ represents the period of the seasonality, and ${a}_{n}$ and ${b}_{n}$ are the Fourier coefficients.