Leveraging artificial intelligence and dynamic supply chains for renewable energy development in Africa's frontier markets

Theoretical foundations

The strategic convergence of AI and dynamic supply chain management represents a transformative yet underexplored avenue for accelerating renewable energy deployment across Africa's frontier markets. While existing scholarship has extensively examined renewable energy adoption through conventional policy, financing, and technological perspectives, ⁹ the role of AI-driven supply chain optimization remains conspicuously absent from the discourse. This oversight is particularly striking given the logistical complexities inherent in Africa's decentralized energy systems, where traditional supply chain models prove inadequate. Africa's energy landscape presents a paradoxical dichotomy: Despite possessing the world's most abundant renewable resources—including solar irradiation exceeding 2000 kWh/m²/year in the Sahel region ¹⁰ and wind potential surpassing 59 GW ¹¹ —the continent continues to grapple with chronic energy poverty, with over 600 million people lacking reliable electricity access. This disparity stems not only from technological constraints but also from systemic supply chain inefficiencies, including fragmented logistics, infrastructural deficits, and last-mile distribution challenges.^12,13

Conventional linear supply chain models ¹⁴ were designed for centralized, fossil fuel-based energy systems and failed to accommodate the unique characteristics of Africa's frontier energy markets. These markets are defined by extreme decentralization (with off-grid and mini-grid systems accounting for nearly 60% of new rural energy access, ⁴ demand volatility, and severe infrastructural constraints. While recent advancements in SCR theory ¹⁵ provide valuable insights, their applicability remains limited to mature manufacturing ecosystems, leaving a critical research gap in adapting these frameworks to Africa's renewable energy sector.

Emerging evidence suggests that AI can serve as a transformative solution to these challenges. Research indicates that AI enhances supply chain forecasting accuracy by up to 30% while reducing logistics costs by 15–20% in developed markets. ¹⁶ In the African context, three key AI applications show particular promise: Reinforcement learning algorithms for optimizing last-mile distribution in infrastructure-scarce regions ¹⁷ ; blockchain-enabled systems for improving procurement transparency in weak institutional environments ¹⁸ ; and computer vision and IoT networks for maintaining remote renewable energy installations. Complementary to these approaches, game-theoretical models of electricity markets provide analytical insights into how multiple stakeholders—including regulators, private developers, and community energy cooperatives—strategically interact to optimize supply chain decisions. ¹⁹ Despite this potential, empirical data reveals that fewer than 5% of renewable energy projects in Africa currently incorporate advanced AI solutions, ²⁰ underscoring a significant implementation gap and a critical area for future research. Recent literature highlights that as AI-driven supply chains evolve, cybersecurity risks and vulnerabilities present emerging challenges that cannot be overlooked. Radanliev et al. ²¹ review cybersecurity threats, exploits, and vulnerabilities in software bills of materials associated with AI-driven systems, emphasizing that these risks could undermine trust and operational stability in digital supply networks. Similarly, Radanliev et al. ²² discuss AI security and cyber risks in IoT-based energy systems, identifying potential exploit vectors in decentralized sensor networks and data-driven decision platforms. Our study complements this emerging body of work by recognizing that cybersecurity preparedness is integral to fully realizing AI's transformative potential in frontier energy markets.

AI's transformative role in renewable energy development

The strategic integration of AI into renewable energy systems is reshaping global energy markets^23,24 by enabling innovative approaches to sustainable energy deployment, maintenance, and scalability. ²⁵ Whereas mature economies primarily leverage AI to optimize existing infrastructure,^26,27 emerging markets demonstrate the potential for technological leapfrogging through context-specific AI adaptations that circumvent structural barriers.^11,28 AI's most significant contribution lies in its ability to process and analyze unconventional, large-scale datasets—ranging from mobile payment transactions to high-resolution satellite imagery—enabling more accurate demand forecasting in regions where traditional metrics prove inadequate. Studies demonstrate that AI-driven design optimization tools can reduce renewable energy project development costs by 18–22% in underserved markets. ²⁹ Furthermore, AI facilitates the customization of solar home systems based on regional economic cycles and environmental conditions, leading to higher adoption rates and lower maintenance expenses. Beyond demand-side applications, AI accelerates the development of context-appropriate renewable technologies. Advanced computer vision systems now guide the assembly of solar panels engineered for high-temperature environments, while natural language processing (NLP) technologies refine pay-as-you-go (PAYG) solar interfaces for low-literacy populations. ⁹ These innovations highlight AI's capacity to foster indigenous technological development rather than mere replication of solutions from advanced economies.

The human capital dimension of renewable energy development has also benefited from AI integration. Virtual reality training programs, enhanced by AI algorithms, have reduced skills acquisition timelines for renewable energy technicians by up to 40% compared to conventional training methods. ¹⁰ This advancement addresses a critical bottleneck in workforce development, particularly in regions experiencing rapid expansion of renewable energy infrastructure. In supply chain management, AI enhances operational efficiency through real-time supplier matching and sophisticated counterfeit detection mechanisms. Digital trade platforms utilizing AI algorithms have demonstrated significant reductions in procurement lead times, while blockchain integration has proven effective in mitigating counterfeit risks, which previously accounted for a substantial share of equipment failures in certain markets. ³⁰ Global renewable energy supply chains exhibit unique vulnerabilities characterized by import dependencies, infrastructure limitations, and fragmented trade networks. ¹⁰ Solar PV components remain highly concentrated in terms of production geography, while wind energy projects frequently face logistical bottlenecks at key transportation nodes. Recent cybersecurity research emphasizes that these vulnerabilities are compounded by digital transformation. AI-enabled IoT systems and software-driven logistics platforms can become targets for malicious cyber exploits, potentially disrupting predictive analytics, autonomous logistics, and blockchain traceability mechanisms critical to renewable energy supply chains.^21,22 Addressing this challenge requires coordinated investment in AI security frameworks, resilient software architectures, and cross-border cyber risk management policies. These dynamics reveal a paradox in global renewable energy markets: while experiencing the world's fastest growth rates, they remain disproportionately vulnerable to supply chain disruptions. Addressing this challenge will require coordinated efforts in regional manufacturing expansion, policy harmonization, and sustained AI integration to build resilient, self-sufficient energy ecosystems for the future.

Theoretical mechanism

Drawing on the theoretical framework of Onukwulu et al. ⁶ and Song et al., ³¹ this study posits that AI-driven strategies can effectively address structural weaknesses in Africa's renewable energy sector. Given that many frontier markets rely on a limited number of suppliers for critical energy infrastructure, ³² AI facilitates supplier diversification, logistics optimization, and predictive analytics, ³³ thereby reducing dependence on single points of failure. We review how AI fosters multilateral cooperation by enhancing data-sharing platforms, automating contract enforcement through blockchain integration, and improving cross-border trade efficiencies. These mechanisms are particularly relevant in Africa, where fragmented regulatory environments and infrastructural constraints often hinder seamless energy distribution.

Dynamic supply chains and ai-driven risk mitigation

Integration of AI and dynamic supply chains is essential for mitigating risks and enhancing the resilience of renewable energy development.^31,34 This study leverages how AI-driven risk prediction models and real-time supply chain monitoring proactively address disruptions, ensuring the steady flow of renewable energy components (e.g., solar panels, wind turbines) across diverse geographies. AI's ability to streamline procurement, enhance transparency, and mitigate vulnerabilities ⁸ positions it as a transformative force in bolstering the sustainability and scalability of renewable energy projects in Africa's emerging markets. African frontier markets face multiple supplier-supply chain risks, including shared risk events ( X ) which include geopolitical instability (e.g., Sahel region), cross-border logistics delays (e.g., avg. 14-day delays at African ports vs. 3 days in China), and currency volatility (e.g., devaluation impacting import costs). To quantify these risks, we model:

By definition,

X = { 0 otherwise 1 if any geopolitical , logistical , or currency risk occurs

(1)

Here the probability of occurrence,

P a = ω 1 . P g e o + ω 2 . P l o g + ω 3 . P c u r r

(2)

where, P g e o is the geopolitical instability risk (e.g., Sahel region = 0.25), P l o g is the logistics delay risk (14-day delay → P l o g = 1 − e − 14 λ , λ = 0.05 ), P c u r r is the currency volatility risk (e.g., Naira devaluation → P c u r r = 0.18 ) , and ω 1 , ω 2 , ω 3 ​ are the weights (e.g., 0.4, 0.3, 0.2). The rationale is that logistical risk uses exponential decay to model delay probability, ³⁵ while weights reflect the World Bank's Africa Pulse risk prioritization. ¹¹

Consequently, supplier failure (Yᵢ) now accounts for informal suppliers (30–40% of renewable component trade in East Africa), and mobile money payment defaults (15–20% rate in PAYG solar markets). ¹¹

By definition , Y i = { 0 o t h e r w i s e 1 if supplier i fails ( formal or informal )

(3)

Here also the probability of failure,

P i = α . P f o r m a l + ( 1 − α ) . P i n f o r m a l

(4)

where α is the share of formal suppliers (e.g., 0.6 in Kenya), P f o r m a l is the failure rate of formal suppliers (e.g., 0.10), P i n f o r m a l is the failure rate of informal suppliers: P i n f o r m a l = 0.35 + 0.15 ( Mobile Money Default Rate ) , eg . , 20 % default rate → P i n f o r m a l = 0.38 . Our rationale is that informal suppliers have 35% baseline failure rates, ³⁶ while mobile money defaults add 15–20% incremental risk. ³⁷

Our proposed supply chain risk management model incorporates critical factors influencing risk probability and associated costs in emerging markets. The probability of supply risk ( P n ( D ) ) is expressed as:

P n ( D ) = P a + ( 1 − P a ) ∏ i ϵ S n ( α P f o r m a l + 0.2 ⏟ I n f o r m a l i t y p r e m i u m )

(5)

where P a represents the baseline risk probability, α P f o r m a l accounts for formal supplier risks, and 20% informality premium ³⁶ reflects the additional risk inherent in informal supply channels. This formulation acknowledges that while diversification across multiple suppliers ( S n ) reduces vulnerability to external shocks through risk distribution, it simultaneously introduces complex management challenges. ³¹

The corresponding cost function ( C ( S n )) captures the economic trade-offs in supplier network management:

C ( S n ) = c + ∑ i ϵ S n ⏟ Formal costs b i + 0.3 c × Informal i ⏟ Informal monitoring

(6)

This structure incorporates three essential cost components: First, a fixed cost (c) that accounts for baseline logistics expenditures, ³⁸ which are typically elevated in developing economies due to infrastructure limitations. Second, variable costs ( b i ) associated with managing each formal supplier. Third, a 30% monitoring premium is applied to informal channels, reflecting the additional oversight required for unofficial suppliers. The model demonstrates an important optimization challenge for supply chain managers. While expanding the supplier base ( S n ) enhances resilience by reducing P n ( D ) through risk pooling, this benefit must be weighed against rising coordination costs. Each additional supplier introduces incremental management expenses, with informal suppliers demanding particularly high monitoring investments (30% premium) due to quality assurance and compliance requirements. This framework enables decision-makers to optimize supplier networks, balancing risk mitigation against cost containment—a crucial consideration for renewable energy projects in frontier markets.

Optimizing renewable energy supply chains through AI-driven supplier management

The strategic diversification of suppliers represents a critical lever for reducing supply chain vulnerability. ³¹ Our AI-powered approach addresses this challenge through three integrated components that balance risk mitigation with cost efficiency.

Supplier diversification engine

A NLP platform can continuously scan over 200 African trade databases (e.g., TradeMark Africa, national procurement registries) to identify and evaluate alternative suppliers, reducing supply risk probability ( P n (D)) by 18–25% in pilot implementations. ³⁹ The platform's machine learning algorithms assess supplier viability across multiple dimensions, enabling dynamic adjustments to supplier networks in response to emerging risks.

Intelligent supplier scoring framework

A proprietary model that assesses suppliers across multiple dimensions, including delivery reliability, AI-predicted stability, transport costs, and currency risk. Each potential supplier undergoes comprehensive evaluation through our proprietary scoring model:

Supplier Scor e i = Delivery Reliability × AI − Predicted Stability Transport Cost + Currency Risk

(7)

The scoring incorporates innovative data inputs including mobile money transaction histories (for liquidity risk assessment) and satellite-tracked delivery routes (for logistical risk evaluation). This multidimensional scoring allows for a more accurate prediction of supplier performance compared to traditional assessment methods.

Cost-optimized network configuration

A reinforcement learning system that dynamically balances risk reduction against management costs, identifying the optimal supplier count for a given market. This is expressed as:

b ′ = n ( L − 1 ) ( 1 − P a ) ( p ′ ) n − 1 − n l ( 1 − P a ) ( 1 − p ′ ) n − 1

(8)

The model incorporates region-specific parameters reflecting the unique challenges of African renewable energy supply chains where we employed the parameters L values 2.5 × higher than Chinese benchmarks, accounting for the severe economic impact of energy poverty, pʹ set at 0.35 (versus 0.15 in China) based on AfDB ³⁶ supply chain surveys, and our continuous adjustment of n (optimal supplier count) based on real-time risk assessments. This AI-driven approach effectively addresses two critical African supply chain challenges: mitigating risks from informal suppliers (through enhanced due diligence) and overcoming logistical bottlenecks (through predictive routing optimization).

Hypothesis development

Theoretical framework

The theoretical framework (Figure 1) underpinning this research integrates these three complementary perspectives that collectively address current limitations in the literature. Dynamic Capabilities Theory ⁴² provides a foundation for understanding how AI enables renewable energy firms to sense demand shifts through innovative data analysis techniques, seize opportunities through predictive maintenance systems, and reconfigure resources using automated logistics optimization. Complex Adaptive Systems Theory ⁴⁸ offers valuable insights into managing the nonlinear dynamics characteristic of African energy markets, where reinforcement learning algorithms can optimize routes amid unpredictable variables like road conditions and fuel availability. Finally, the Technology-Organization-Environment Framework ⁴⁹ provides crucial implementation insights by systematically evaluating AI suitability given Africa's digital infrastructure constraints, required hybrid capabilities, and varying regulatory environments.

OPEN IN VIEWER

Figure 1.

Theoretical framework. Notes: The figure illustrates how artificial intelligence (AI) technologies influence dynamic supply chains, which in turn drive renewable energy development in Africa's frontier markets. The framework incorporates key mediating factors (predictive analytics and blockchain-based tracking) and moderating factors (digital infrastructure and regulatory support). Source: Authors’ conceptualization.

This integrated approach addresses three critical limitations in current research: first, the lack of comprehensive frameworks that adequately consider the dynamic interplay between technological, logistical, and institutional factors in frontier energy markets; second, the striking absence of empirical studies validating AI applications in real-world African energy supply chains; and third, the inflexibility of current models to adapt to both the rapid evolution of renewable technologies and the unique characteristics of African market conditions. This research makes significant theoretical and practical contributions by developing an integrated framework that combines AI-driven predictive analytics with context-aware dynamic logistics models specifically designed for Africa's renewable energy sector. The framework advances academic understanding in energy economics, logistics, and digital transformation and provides actionable insights for policymakers and industry practitioners seeking sustainable energy access across Africa's frontier markets. Through its dual focus on technological innovation and contextual adaptation, the study offers replicable models for other developing regions facing similar energy challenges while contributing new perspectives to the ongoing discourse on sustainable development and Industry 4.0 applications in emerging economies.

Data architecture and measurement

The empirical analysis utilizes a multi-source panel dataset spanning the period 2015–2024 across twenty African frontier markets, selected through stratified sampling based on IRENA's Renewable Energy Readiness Index. ¹⁰ The data infrastructure incorporates three primary data streams: trade flow records from UN Comtrade (HS codes 8541, 8501, 8542), AI adoption metrics from World Bank Enterprise Surveys, and supply chain performance indicators from AfDB Logistics Performance Assessments.

The variable construction framework operationalizes key constructs through rigorous measurement protocols. SCR is derived through principal component analysis (PCA) of three operational metrics: Lead times (in days), inventory turnover ratios, and disruption frequency indices. The AI integration index represents a weighted composite score incorporating predictive analytics adoption rates (measured on a 0–1 scale), autonomous logistics implementation status (binary), and blockchain traceability deployment (binary). Policy effectiveness is quantified as the interaction between World Bank Governance Indicators’ Regulatory Quality scores and RISE Database Energy Policy Intensity metrics.

To account for the substantial heterogeneity in technological infrastructure and AI readiness across African nations, this study extends the AI integration index by incorporating country-specific digital infrastructure metrics (mobile broadband penetration, cloud computing availability), electricity access levels (World Bank SDG7 tracking data), and AI human capital indicators (AI-related patent filings, AI workforce skill indices). These dimensions allow the index to reflect not only the nominal presence of AI technologies but also the enabling environment required for effective adoption.^60,61 We standardize these measures within each country and year to mitigate bias from uneven data availability. Furthermore, to explicitly handle variance across nations, the dataset integrates FE capturing country-level structural differences (e.g., ICT development, regulatory frameworks) and interaction terms between AI adoption and digital readiness indicators. This design ensures that countries with weaker technological infrastructure are not overrepresented in AI impact estimates. Sensitivity analyses segment the sample into high-readiness and low-readiness groups based on median broadband coverage and ICT capacity, enabling a comparative assessment of AI's role in differing technological contexts. ¹⁷ This allows for a comparative analysis of AI adoption impacts under diverse infrastructure conditions. Outlier detection techniques (Mahalanobis distance and Cook's D) are applied to remove influential observations that could skew cross-country results. Lastly, recognizing that supply chain optimization may be constrained by skill shortages, the study introduces an AI Skills Gap variable, derived from and data, capturing shortages in machine learning and data engineering expertise relative to labor market needs. This variable serves as a moderator in econometric specifications, allowing the analysis to quantify how workforce readiness influences AI-driven supply chain performance.^62,63 Figure 2 illustrates the variables’ flowchart causal pathway.

OPEN IN VIEWER

Figure 2.

Variables flowchart. Notes: This figure illustrates the flow of data and variables used in econometric and machine learning models. Source: Authors’ elaboration.

Core econometric framework

System generalized method of moments

The analysis employs a system generalized method of moments (GMM) estimator to address potential endogeneity, autocorrelation, and heteroskedasticity concerns inherent in the AI-renewables relationship. The baseline specification models renewable energy capacity per capita (log-transformed) as a function of AI adoption intensity, dynamic supply chain capabilities, and their interaction effect, while controlling for macroeconomic and institutional factors. To capture cross-country differences in technological infrastructure and workforce readiness, we incorporate measures of digital readiness (mobile broadband penetration, cloud computing adoption), AI Skills Gap, and interaction effects that test how digital infrastructure amplifies the impact of AI adoption on renewable energy deployment. The baseline model is specified as follows:

RE i t = 0 + 1 AI i t + 2 DSC i t + 3 ( AI i t . DSC i t ) + α 4 DigInfra i t + α 5 ( AI i t × DigInfra i t ) + α 6 SkillsGap i t + α 7 ( AI i t × SkillsGap i t ) + 8 X i t + λ t + μ i + i t

(9)

where RE i t represents renewable energy development (measured as installed renewable energy capacity in MW per capita in country i at time t ). AI i t denotes the extent of AI adoption in supply chains (standardized AI adoption index ranging from 0 to 1), measured through an AI utilization index. DSC i t ​ captures dynamic supply chain capability scores, operationalized via inventory optimization, demand forecasting accuracy, and logistics efficiency metrics. AI i t . DSC i t captures the interaction effect, testing whether AI enhances dynamic supply chain performance for renewable energy deployment. DigInfra i t is the composite digital infrastructure readiness index (mobile broadband subscriptions, cloud computing services). SkillsGap i t is the AI workforce skills gap measure (shortage index for data and AI engineers). X i t is a vector of control variables including GDP per capita, mobile internet penetration rates, trade openness, and regulatory quality, λ t represents year-FE that account for temporal macroeconomic shocks, μ i denotes the country-specific FE, and i t is the error term. Interaction terms ( AI i t × DigInfra i t ) and ( AI i t × SkillsGap i t ) capture how technological infrastructure and workforce readiness condition the effect of AI on renewable energy capacity, aligning with technology diffusion literature (see Aghion et al.). ⁶¹

To mitigate reverse causality (AI adoption may increase in response to supply chain shocks), we use the Arellano-Bover/Blundell-Bond Estimator:

Δ SC V i t = α Δ SC V i t − 1 + β 1 Δ A I i t + β 2 Δ DS C i t + β 3 Δ DigInfr a i t + β 4 Δ SkillsGa p i t + Δ X i t + Δ i t

(10)

The identification strategy employs three complementary approaches: Lagged AI adoption variables (t-2, t-3) as internal instruments, heteroskedasticity-based instruments variability following Lewbel, ⁶⁴ and robustness checks through Limited Information Maximum Likelihood estimation. We then test our Arellano-Bond AR(2) (for autocorrelation) and Hansen J-test (for overidentification), ensuring that inclusion of digital readiness and skills gap variables does not compromise model validity.

Model estimation

Our empirical analysis employs a dual-model approach to examine the heterogeneous effects of AI-enhanced dynamic supply chains on renewable energy deployment across African frontier markets, building on recent methodological advances in energy economics.^6,40 The FE and random-effects (REs) specifications enable rigorous assessment of both within-country temporal dynamics and cross-country structural variations in supply chain performance. Country-specific intercepts ( α i ) and hierarchical modeling elements ensure that national characteristics, such as regulatory capacity and grid infrastructure, are properly captured.

FE model

Building upon established econometric theory,^65,66 this approach effectively controls for unobserved country-specific heterogeneity that could otherwise bias estimation results. The model specification captures persistent national characteristics including institutional frameworks, grid infrastructure quality, and geographical constraints that may affect renewable energy adoption patterns. The core econometric specification takes the following form:

RE i t = α i + β 1 AI it + β 2 DSC it + β 3 ( AI it × DSC it ) + β 4 X it + μ i + i t

(11)

RE model

The RE model provides an alternative analytical framework for examining the relationship between AI adoption, dynamic supply chain capabilities, and renewable energy development across African frontier markets. Building upon the foundational work of Bell and Jones, ⁶⁷ this specification treats country-specific unobserved characteristics as random variables that are uncorrelated with the explanatory variables, allowing for both within-country and between-country variation in the analysis. The model specification takes the following form:

RE i t = α i + β 1 AI i t + β 2 DSC i t + β 3 ( AI i t × DSC i t ) + β 4 X i t + μ i + λ i + i t

(12)

The FE and RE specification captures the logarithmic transformation of renewable energy capacity per capita ( RE i t ) measured in megawatts for country i at time t, capturing the full spectrum of grid-connected and off-grid energy solutions from utility-scale projects to decentralized systems as documented in IRENA's 2023 global energy statistics. This structure addresses several critical dimensions of renewable energy deployment in frontier markets. The intercept term α i represents the baseline renewable energy capacity across country FE that account for time-invariant national characteristics ⁶⁷ in the sample.

AI i t constitutes a standardized AI adoption index (ranging from 0 to 1) that comprehensively measures supply chain automation through three validated components: Robotic process automation in energy logistics, ⁶⁸ AI patent filings specific to renewable technologies, ⁶⁹ and mobile-based AI applications for energy distribution systems. ³⁷ This multidimensional approach captures the diverse pathways through which AI transforms energy systems in developing economies. DSC it denotes dynamic supply chain capability as a principal component score derived from six World Bank Logistics Performance Indicators, providing a multidimensional assessment of operational efficiency across customs clearance, infrastructure quality, international shipment processing, logistics workforce competency, tracking systems, and delivery reliability ⁷⁰ ; the interaction term ( AI it × DSC it ) tests the contingent relationship between AI technologies and supply chain maturity, building upon the conceptual framework developed by Acemoglu and Restrepo. ⁶²

The model incorporates several innovative features to address frontier market specificities. The RE component ( μ i ) accounts for unobserved country-level heterogeneity in institutional quality, resource endowments, and geographical constraints, assumed to follow a normal distribution with mean zero and constant variance ( μ i ∼ N(0, α μ 2 )). Building on recent methodological advances in hierarchical modeling, ⁷¹ we introduce λ i to capture firm-level variation within the renewable energy sector, reflecting Africa's multi-level structure of energy supply chains. ⁶⁰

Control variables ( X it ) include digital infrastructure penetration (measured by mobile broadband subscriptions), regulatory quality (as assessed by the World Governance Indicators), energy import dependence, and climate vulnerability (as indicated by the ND-GAIN Index). The error term ( ξ i t ) maintains standard normality assumptions ( ξ i t ∼ N(0, α 2 )), with the RE approach assuming no correlation between country-specific effects and explanatory variables. ⁶⁶ This proves particularly valuable for analyzing technology diffusion patterns in frontier markets where between-country variation contains meaningful structural information. ⁶¹

The model's logarithmic transformation of the dependent variable enables the interpretation of coefficients as percentage changes while normalizing the distribution of renewable energy capacity measures. This specification addresses key econometric challenges in frontier market analysis by: (1) Controlling for unobserved heterogeneity across diverse institutional contexts; (2) testing the contingency hypothesis that technological innovations require supportive organizational capabilities; and (3) accounting for Africa's distinctive energy landscape where off-grid solutions represent over 60% of new electricity access. ⁴ Through the Hausman specification test, ⁶⁵ this approach provides a robust analytical framework for understanding how AI and dynamic supply chain capabilities interact to transform renewable energy development while accounting for both country-level FE and enterprise-specific factors that influence technology adoption and implementation in Africa's frontier markets.

Analytical approaches

Nonlinear autoregressive distributed lag model

Given the asymmetric nature of technology adoption in frontier markets, we utilize a nonlinear autoregressive distributed lag (NARDL) model ⁷² to differentiate between positive and negative shocks in AI adoption and supply chain development. This approach, based on recent advancements in econometrics, ⁷² effectively captures short- and long-run effects while addressing potential endogeneity issues in technology-growth dynamics. The core NARDL specification for renewable energy deployment is structured as:

Δ RE i t = α i + ∑ i = 1 p Φ i Δ RE t = i + ∑ j = 0 q θ j + Δ AI t − j + + ∑ j = 0 q θ j − Δ AI t − j − + β 1 + AI t − 1 + + β 1 − AI t − 1 − + γ DS C i t + t

(13)

where AI t + and AI t − are the positive and negative changes in AI adoption, derived from a composite index of robotic automation, ⁶⁸ AI patents, ⁶⁹ and mobile applications ³⁷ ; θ j + and θ j − capture short-run asymmetric effects, while β 1 + and β 1 − measure long-run asymmetric impacts.

To ensure robustness, the study employed a panel data testing framework where we applied a structured econometric validation process:

Stationarity testing: We verify the integration order of variables using Phillips-Perron (PP) and Augmented Dickey-Fuller tests, with the PP test preferred for its robustness against heteroskedasticity. ⁷³ The general test equation is:

Δ Y t = α + β t + γ Y t − 1 + ∑ i = 1 k δ i Δ RE t = i + ε t

(14)

Model selection: To determine whether fixed or RE are appropriate, we conduct the ⁶⁵ specification test:

H = ( β F E − β R E ) ′ [ Var ( β F E ) − Var ( β R E ) ] − 1 ( β F E − β R E ) ∼ X 2 ( K )

(15)

If the Hausman test is significant (p < 0.05) → Use FE

If not significant (p > 0.05) → Use RE

To establish cointegration, we apply bounds testing, ⁷² while the error correction mechanism quantifies the speed of adjustment toward equilibrium. Given Africa's mobile-first technology diffusion ⁶⁰ and fragmented supply chains, ⁷⁰ this approach effectively captures the nonlinear adoption patterns shaping renewable energy transitions.

Network analytics module

The study models renewable energy supply chains as directed networks G t = ( V , E t , W t ), where V represents nodes (countries and major firms), E t captures directed edges (component flows), and W t assigns trade volume weights. The analysis focuses on two key structural metrics:

The polarization index ( PI i t ) quantifies supply chain asymmetries through the formula:

PI i t = 1 − ∑ i = 1 n [ Export i − Import i ] Total Trad e i Volume

(16)

Betweenness centrality measures identify critical hubs in the continental supply network:

C B ( v ) = ∑ s ≠ v ≠ t σ s t ( v ) σ s t

(17)

These metrics are computed annually using NetworkX algorithms and visualized through Gephi's Force Atlas 2 layout, with node sizing proportional to eigenvector centrality scores. Figure 3 depicts the network centrality and PI.

OPEN IN VIEWER

Figure 3.

Network centrality and polarization index. Notes: Node size: Represents Eigenvector centrality, indicating a country's overall importance in the artificial intelligence (AI)-driven supply chain. Edge thickness: Represents the strength of collaboration between countries. Color intensity: Represents betweenness centrality, highlighting key intermediary hubs like South Africa and Kenya in AI-enabled renewable energy networks. South Africa and Kenya act as major connectors in the network, bridging different regions. Source: Author via Gephi + NetworkX (Python).

Machine learning augmentation

Our predictive analytics framework employs an ensemble machine learning approach to assess supply chain disruption risks in African renewable energy markets. The methodology combines two complementary techniques to capture both classification and regression perspectives on supply chain vulnerabilities. The primary classification model utilizes an XGBoost algorithm ⁷⁴ trained on 34 predictive features spanning five critical dimensions: AI adoption metrics, digital infrastructure indicators, trade policy variables, macroeconomic conditions, and institutional factors. Through Bayesian optimization, we achieve cross-validated area under receiver operating characteristic curve (AUC)-receiver operating characteristic curve (ROC) scores of 0.87, with SHAP value analysis revealing mobile broadband penetration and AI-driven demand forecasting accuracy as the most significant predictors of SCR.

To ensure rigorous validation, we implemented a stratified 10-fold cross-validation procedure that maintains the proportional representation of high- and low-adoption countries in both training and testing folds. We further performed bootstrap resampling (1000 iterations) to assess coefficient stability and prediction intervals. Placebo tests were conducted by randomly permuting AI adoption values across countries to verify that observed predictive power did not arise from spurious correlations. The model consistently outperformed these placebo simulations, with accuracy drops exceeding 35% under permutation tests. For enhanced predictive accuracy, we complement this with a random forest regressor that specifically analyzes the continuous relationships between AI adoption levels, digital infrastructure development, and trade policy frameworks. This regression approach identifies key drivers that inform our spatial econometric modeling.

To address potential biases in data collection and sample representativeness, we applied country-level weighting schemes based on renewable energy capacity size and adjusted for policy environment heterogeneity through RE modeling. Sensitivity analyses were performed by sequentially excluding dominant economies (South Africa, Morocco, Egypt) and re-estimating model performance; the AUC-ROC scores remained stable (variation < 0.02), indicating that no single country disproportionately drives the results. We also examined the distribution of AI adoption and supply chain indicators using skewness and kurtosis measures to confirm balanced representation across the 20-country sample.

The spatial analysis component incorporates a spatial autoregressive model to quantify neighborhood effects and technology spillovers across African markets. We extend this with a Spatial Durbin specification:

SCR i = ρ W SCV j + β A I i + W AI θ + γ X i + i

(18)

where SCR i represents SCR, W is the queen-contiguity spatial weights matrix, ρ captures spillover effects, and the model captures both direct effects ( β ) and spatial spillovers ( θ ) of AI adoption (Figure 4). This dual machine learning and spatial econometrics approach, combined with the expanded validation and bias mitigation procedures, provides comprehensive and reliable insights into the determinants and geographic diffusion patterns of supply chain risks in Africa's renewable energy sector.

OPEN IN VIEWER

Figure 4.

Spatial spillovers choropleth. Notes: Geographic representation of spatial spillover effects of artificial intelligence (AI) adoption on supply chain resilience. Darker shades indicate stronger positive spillover effects between neighboring countries. Source: Author via QGIS + GeoPandas (python).

Further assessment

Analytical techniques

The following analytical techniques complement the empirical strategy:

(a) Mediation analysis (structural equation modeling—SEM): To examine whether dynamic supply chains mediate the relationship between AI adoption and renewable energy outcomes, a structural equation model is estimated:

DSC i t = β 0 + β 1 AI i t + β 2 Z i t + i t

(19)

RE i t = γ 0 + γ 1 DSC i t + γ 2 AI i t + γ 3 Z i t + 2 , i t

(20)

where Z i t includes digital infrastructure and policy environment as mediating variables. The indirect effect is computed as:

Indirect Effect = β 1 . γ 1

(21)

A Sobel test will assess whether the mediation effect is statistically significant.

(b) Moderation analysis (hierarchical regression models): To test the moderating effects of digital infrastructure and regulatory policies, the study employs an interaction model:

RE i t = δ 0 + δ 1 AI i t + δ 2 DSC i t + δ 3 ( AI i t . DSC i t ) + δ 4 ( AI i t . DigInfra i t ) + δ 5 ( AI i t . RegQual i t ) + δ 6 X i t + λ t + i t

(22)

where DigInfra i t represents digital infrastructure quality (measured by mobile broadband penetration and cloud computing adoption), RegQual i t denotes regulatory quality (measured by an index of energy sector policies).

Statistical significance of the interaction terms ( δ 4 , δ 5 ) will indicate whether AI's impact is contingent on digital and policy environments.

(c). Threshold effect analysis (panel threshold regression—PTR): Given the possibility of non-linear effects, the study employs Hansen's ⁷⁵ Threshold Regression Model to determine whether AI adoption impacts renewable energy development differently at varying levels of digital infrastructure:

RE i t = θ 0 + θ 1 AI i t .1 ( DigInfra i t ≤ τ ) + θ 2 AI i t .1 ( DigInfra i t > τ ) + θ 3 X i t + λ t + i t

(23)

where τ represents the threshold value of digital infrastructure. The model estimates whether AI adoption has stronger effects in high-connectivity markets.

Policy-Simulation

Our analytical framework employs advanced simulation techniques and rigorous case selection protocols to assess AI's impact on renewable energy supply chains across African markets. The methodology integrates Bayesian hierarchical modeling^63,76 with synthetic control approaches ⁷⁷ to generate robust policy insights while maintaining scientific rigor and practical relevance.

The policy simulation extension utilizes a Bayesian hierarchical model structure to evaluate counterfactual scenarios and estimate marginal effects:

SCR i t ∼ N ( α i + β A I i t , σ 2 )

(24)

α i ∼ N ( μ , τ 2 )

(25)

This formulation enables country-specific intercepts ( α i ) while estimating the global effect of AI adoption ( β ) on SCR . Posterior distribution analysis quantifies how AI's effectiveness varies under different digital infrastructure development levels, ¹⁷ providing nuanced policy guidance.

For the synthetic control method implementation, we establish strict case selection criteria following Abadie. ⁷⁷ The treatment group comprises countries demonstrating greater than 50% AI adoption growth between 2015 and 2024, with primary case studies including Kenya (notable for M-Pesa integration), Morocco (featuring the Noor Ouarzazate solar complex), and South Africa (demonstrating industrial AI adoption). We exclude nations with more than three years of missing pre-treatment data to ensure analysis quality.

The donor pool construction follows a rigorous two-stage process. ⁷⁸ First, we match on pre-treatment characteristics (2015–2019) including renewable energy capacity, mobile broadband penetration, trade openness metrics from the World Bank, and Logistics Performance Index scores. Second, we remove outliers using Mahalanobis distance thresholds at the 95th percentile, ⁷⁷ resulting in a final donor pool of 20 countries including Senegal, Ghana, and Tunisia.

The optimization framework aims to minimize the weighted discrepancy between the treated and control units by solving the following objective function:

min w ( X 1 − X 0 w ) ′ V ( X 1 − X 0 w )

(26)

Subject to: w j ≥ 0 ( weights must be non - negative ) ; ∑ w j = 0 ( weights sum to one )

where X 1 represents the characteristics of the treated unit, X 0 is the matrix of potential control units (donor pool), and V is an optimized weight matrix, determined using genetic algorithms. ⁷⁴ This approach ensures an optimal counterfactual construction, improving the robustness of policy impact evaluation. Validation protocols ⁷⁸ maintain analytical rigor through multiple checks. We require pre-treatment fit with RMSE below 0.25 on the standardized index, placebo tests where treatment effects must exceed 90% of simulated outcomes, and leave-one-out stability checks maintaining coefficient variation under 15%.

The computational infrastructure (Table 1) combines specialized software tools with high-performance resources. The software stack includes Synth (R) for SCM optimization with permutation inference, ⁷⁷ Stan (Python/R) for Bayesian estimation using Hamiltonian Monte Carlo sampling, ⁶³ NetworkX and Gephi for network analysis including community detection, ⁷¹ and GPU-accelerated XGBoost via cuML ⁷⁹ for machine learning tasks with SHAP analysis. ⁸⁰ High-performance computing resources enable efficient execution, featuring AWS Batch parallelization for 1000 bootstrap replications, NVIDIA RAPIDS acceleration delivering 8× speedup for XGBoost training, ⁷⁹ and distributed memory architecture using 64-core EC2 instances. This integrated methodology combines advanced statistical modeling with cutting-edge computational techniques to produce robust, policy-relevant findings about AI's transformative potential ⁶² for Africa's renewable energy supply chains.

OPEN IN VIEWER

Table 1.

Software implementation.

Component	Tool	Capabilities
SCM optimization	Synth (R)	Permutation inference, MSE minimization
Bayesian estimation	Stan (Python/R)	Hamiltonian Monte Carlo sampling
Network analysis	NetworkX + Gephi	Community detection, polarization metrics
Machine learning	XGBoost (cuML)	GPU-accelerated SHAP analysis

Robustness validation

The study implements a comprehensive validation strategy to ensure the reliability and generalizability of our findings as presented in Table 2.

OPEN IN VIEWER

Table 2.

Robustness checks.

Method	Purpose
Leave-one-out GMM	Excludes each country sequentially
Bartik instrument	Uses global AI trends as IV
Nonlinear interactions	AI² terms for diminishing returns
Subsampling	Urban vs. rural supply chains

GMM: generalized method of moments; AI: artificial intelligence.

Seven distinct robustness checks were systematically conducted to verify the consistency of our results. First, we examine alternative AI measurement by substituting our composite AI utilization index with direct firm-level expenditures on AI-based supply chain technologies. This alternative operationalization helps verify whether our core findings hold when using concrete investment metrics rather than adoption indices. Second, we employ alternative estimation methods by re-estimating all models using both FE and RE specifications. This approach confirms that our results are robust to different assumptions about unobserved heterogeneity and between-country variation. Third, we conduct subsample analysis by stratifying our dataset between high and low-regulatory-quality countries. This test evaluates whether the relationship between AI adoption and supply chain performance varies across different institutional environments. Fourth, we explore alternative AI proxies including counts of AI research publications and measures of cloud computing usage. These supplementary indicators provide complementary perspectives on technological diffusion beyond our primary AI adoption metrics. Fifth, we perform a regional subsample analysis comparing results between East and West African markets. This geographical breakdown assesses whether regional economic and infrastructural differences moderate AI's impacts. Sixth, we test for nonlinear effects by incorporating quadratic AI terms in our specifications. This examination reveals whether diminishing returns emerge at higher levels of AI adoption in supply chain applications. Seventh, we implement placebo tests (Figure 5) by randomly reassigning AI adoption levels across countries while preserving other variables. This validation check helps ensure our identified effects are not artifacts of spurious correlations or model misspecification. Together, these seven validation procedures provide strong evidence that our core findings are not sensitive to specific measurement choices, estimation techniques, or sample compositions. The consistency of results across these varied tests significantly strengthens confidence in the study's conclusions about AI's transformative potential for renewable energy supply chains in African frontier markets.

OPEN IN VIEWER

Figure 5.

Distribution of placebo test. Notes: Placebo test results validating the robustness of synthetic control methods used for policy simulation. The treated countries show effects significantly larger than 90% of placebo outcomes. Source: Author's via Synth (R).

Descriptive statistics

The AI adoption index integrates predictive analytics, autonomous logistics, and blockchain traceability, following composite index construction methods validated in supply chain automation research. ¹⁶ SCR scores, derived via PCA, are standardized (z-scores), with negative values indicating below-average performance relative to the sample mean. All monetary variables are adjusted to 2024 constant USD. Our dataset comprises 200 country-year observations (20 African frontier markets, 2015–2024), capturing stark disparities in renewable energy capacity (e.g., South Africa and Morocco vs. Chad and Niger). This balanced panel enables robust multivariate analysis, with log-transformed dependent variables (renewable energy capacity) and standardized resilience metrics facilitating cross-country comparability. ⁷¹ The operationalization reflects Africa's unique energy landscape, where AI-driven supply chains interact with institutional factors,^60,70 while control variables, such as GDP per capita and trade openness, account for structural heterogeneity. ⁶⁶ Importantly, we address potential discrepancies in AI adoption rates across countries by decomposing the AI adoption index into three weighted components—predictive analytics (0.6), autonomous logistics (0.3), and blockchain traceability (0.1)—and augmenting it with country-level digital infrastructure and AI workforce indicators. This adjustment ensures that the index reflects both technological deployment and enabling capacity, avoiding bias from assuming uniform readiness across all nations. To validate this approach, we conducted robustness checks where the index was alternatively weighted, and outlier countries (top and bottom 5% in AI adoption rates) were excluded. Results remain qualitatively unchanged, indicating that the findings are not driven by high- or low-adoption extremes. Furthermore, descriptive trend analysis reveals significant heterogeneity: high-connectivity countries (e.g., Kenya, South Africa, Morocco) exhibit average AI adoption scores of 0.58 compared to 0.26 in low-connectivity markets (e.g., Chad, Niger), with corresponding differences in renewable capacity growth. Sensitivity tests (Figure 5, right panel) illustrate that AI-driven supply chain efficiency gains are approximately 40% stronger in countries with above-median digital infrastructure, confirming the moderating role of ICT readiness. Table 3 presents the descriptive statistics and variable definitions, while Figure 6 depicts the trend analysis of AI adoption (y-axis) vs. renewable capacity (secondary y-axis) over time and sample 5 frontier markets by AI-driven supply chain efficiency gains.

OPEN IN VIEWER

Figure 6.

Trend analysis (artificial intelligence (AI) vs. renewable capacity). Notes: Temporal evolution of AI adoption and renewable energy capacity across selected frontier markets. The solid line represents mean capacity, while the dashed line represents AI-driven supply chain efficiency gains. Source: Author via Stan (Python/R).

OPEN IN VIEWER

Table 3.

Descriptive statistics and variable definitions.

Variable/source	Description	Dimension	No.	Mean	SD	Min	Max
Renewable energy capacity (IRENA, AfDB country reports)	Installed renewable energy capacity per capita	MW per 100,000 population (log-transformed)	200	2.34	0.87	0.12	4.56
AI adoption index (World Bank Enterprise Surveys, Industrial Robotics Database, Firm-level surveys: M-KOPA, Zola Electric)	Degree of AI integration in supply chain operations	Composite index (0–1 scale) of: (1) Predictive analytics (0.5 weight), (2) autonomous logistics (0.25), (3) blockchain traceability (0.10), (4) digital infrastructure readiness adjustment (0.10), (5) AI workforce skills availability (0.05)	200	0.42	0.18	0.05	0.89
Supply chain resilience (World Bank LPI, AfDB operational data, UNCTAD Logistics Reports)	Ability to withstand and recover from disruptions	PCA-derived score combining: Lead times (−0.5), inventory turnover (0.3), disruption frequency (−0.2)	200	0.00	1.00	−2.31	2.15
Digital Infrastructure (GSMA mobile connectivity index, ITU indicators)	Enabling environment for AI applications	Mobile broadband subscriptions per 100 people (z-scores)	200	0.32	0.45	−1.20	1.80
AI workforce skills gap (UNESCO skills report, LinkedIn talent insights)	Shortage of skilled AI professionals in renewable energy supply chains	Index (0–1 scale, higher = larger skills gap)	200	0.46	0.22	0.10	0.88
Regulatory quality (World Bank WGI, RISE database, energy policy scores)	Effectiveness of energy sector governance	World Bank Governance Indicator (range −2.5 to 2.5)	200	0.32	0.45	−1.20	1.80
GDP per capita (World Bank WDI, IMF outlook)	Economic development level	Constant 2015 US$ (log)	200	7.85	1.12	5.23	9.78
Trade openness (UN Comtrade, World Bank WDI)	Integration in global trade networks	(Exports + Imports) / GDP (%)	200	62.3	18.7	22.1	89.5
Energy import dependence (IEA, AfDB energy statistics)	Vulnerability to external energy shocks	Net energy imports/total consumption (%)	200	41.2	25.6	3.5	92.7

Notes: The revised AI adoption index adjusts weights to capture enabling capacity via digital infrastructure readiness and AI workforce skills, ensuring cross-country comparability. All variables standardized (z-scores) where appropriate; monetary values in 2024 constant USD. Robustness checks include alternative weighting schemes and exclusion of outliers (top and bottom 5% AI adoption), with consistent results. PCA: principal component analysis; AI: artificial intelligence.

Dynamic panel GMM estimation

The analysis employs a two-step system GMM estimator with Windmeijer-corrected standard errors to address endogeneity, using lagged levels (t-2, t-3) of endogenous variables as instruments (Table 4). Country and year FE are included across country observations yielding an R-squared (within) of 0.682. Results indicate that AI adoption significantly enhances renewable energy capacity, with a 1-unit increase in the AI index associated with a 42.8% rise in renewable deployment. A complementary effect emerges between AI and SCR with energy import dependence negatively impacts outcomes. Dynamic persistence is evident (lagged DV β = 0.517, p < 0.01), and diagnostic tests confirm robustness: no autocorrelation, valid instruments, and low multicollinearity. The Hausman test supports FE specification, aligning with findings on technology-driven energy transitions in frontier markets. ⁴⁰ Interaction terms AI i t × DigInfra i t and AI i t × SkillsGap i t were tested to capture moderating effects. Results show that AI's impact on renewable capacity is 38–45% stronger in countries with above-median broadband penetration and smaller skills gaps (both interactions significant at p < 0.05). Conversely, in low-readiness countries, the marginal effect of AI adoption drops to 0.27 (p < 0.10), confirming that technological infrastructure and human capital are critical enablers.

OPEN IN VIEWER

Table 4.

Dynamic panel GMM estimation results.

Variable	Coefficient	Std. error	z-value	p-value	95% Conf. interval
AI adoption index	0.412***	0.095	4.34	0.000	[0.226, 0.598]
Supply chain resilience	0.305**	0.121	2.52	0.012	[0.067, 0.543]
AI × SCM Interaction	0.176*	0.101	1.74	0.082	[-0.021, 0.373]
AI × Digital Infrastructure	0.189 **	0.088	2.15	0.032	[0.016, 0.362]
AI × Skills Gap	-0.162 **	0.079	-2.05	0.041	[-0.318, -0.006]
GDP per capita (log)	0.124**	0.056	2.21	0.027	[0.014, 0.234]
Digital infrastructure	0.196***	0.071	2.76	0.006	[0.056, 0.336]
Trade openness	0.091	0.065	1.40	0.162	[-0.036, 0.218]
Energy import dependence	-0.149**	0.061	-2.44	0.015	[-0.268, -0.030]
RE capacity (t-1)	0.508***	0.087	5.84	0.000	[0.338, 0.678]
Diagnostic tests
Test category/name		Statistic	p-value	Conclusion
Model specification (Ramsey RESET test)		F = 1.21	0.289	No specification error
Serial correlation (Breusch-Godfrey LM Test)		χ² = 3.08	0.215	No serial correlation
Normality (Jarque-Bera test)		χ² = 2.79	0.247	Residuals normally distributed
Heteroskedasticity (ARCH test, 4 lags)		χ² = 5.59	0.133	No ARCH effects
Endogeneity (Hausman FE vs. RE)		χ² = 19.02	0.015	Reject H0: FE preferred
Autocorrelation (Arellano-Bond AR(2))		z = 0.85	0.392	No autocorrelation
Instrument validity (Hansen J)		χ² = 9.36	0.155	Valid instruments
Instrument exogeneity (Diff-in-Hansen)		χ² = 5.27	0.261	Exogeneity valid

Notes: Estimates use a two-step System GMM with Windmeijer correction. Instruments include lagged levels (t-2, t-3) of endogenous variables. AI effectiveness is amplified by stronger digital infrastructure and reduced by higher skills gaps. Hansen and AR(2) tests confirm instrument validity and absence of second-order autocorrelation. FE: fixed effects; GMM: generalized method of moment; AI: artificial intelligence.

Robustness checks were conducted to ensure findings were not driven by outliers or index construction assumptions. (1) Estimation, excluding the top and bottom 5% of AI adoption countries produced coefficients within ±5% of the baseline. (2) Alternative weightings of the AI integration index (e.g., equal weights or higher emphasis on blockchain) yielded consistent results (coefficients ranging 0.41–0.45). (3) A subgroup analysis comparing high-readiness vs. low-readiness clusters (based on digital infrastructure and regulatory quality) revealed stronger AI effects in digitally advanced markets, highlighting heterogeneity across nations. Additionally, validation procedures for the machine learning models included stratified 10-fold cross-validation, bootstrap resampling, and placebo testing, while country-level weighting and sensitivity analyses confirmed that the 20-country sample is broadly representative and not disproportionately influenced by high-capacity economies. Diagnostic tests, including Hansen J and Arellano-Bond AR(2), remained valid across all robustness checks, indicating the stability of instruments and absence of second-order autocorrelation. Figure 7 (right panel) depicts Q-Q plots and marginal effects plots, illustrating how digital infrastructure amplifies AI's contribution to renewable energy deployment.

OPEN IN VIEWER

Figure 7.

Q-Q plot (residual normality). Notes: Quantile-quantile plot assessing normality of residuals from dynamic panel generalized method of moments (GMM) estimation. Points along the 45-degree line indicate residuals cosely follow a normal distribution. Source: Author via Stan (Python/R).

Machine learning XGBoost SHAP analysis

The XGBoost model demonstrates strong predictive performance in identifying supply chain disruption risks, trained on 5214 firm-quarter observations across our 20 African markets with 10-fold cross-validation. SHAP analysis reveals mobile broadband penetration as the most protective factor, highlighting digital infrastructure's critical role in mitigating disruptions. AI capabilities, particularly demand forecasting accuracy, and regulatory quality emerge as significant resilience enhancers, while operational factors like inventory turnover and cross-border delays present trade-offs between efficiency and vulnerability. These findings, generated using Python's SHAP library (KernelExplainer) with normalized values, underscore the dual importance of technological adoption (AI, connectivity) and institutional strengthening for resilient renewable energy supply chains in frontier markets, consistent with recent evidence on digital transformation in emerging economies.^17,24 Figure 8 gives the ROC curve predictive performance of the XGBoost classifier and Table 5 presents the top 5 predictors of supply chain disruptions (XGBoost SHAP analysis).

OPEN IN VIEWER

Figure 8.

ROC curve (XGBoost classifier). Notes: ROC curve from XGBoost classification model predicting supply chain disruption risks. AUC score of 0.82 indicates strong predictive performance. AUC: area under the receiver operating characteristic curve; ROC: receiver operating characteristic curve. Source: Author via XGBoost (cuML).

OPEN IN VIEWER

Table 5.

Top 5 predictors of supply chain disruptions (XGBoost SHAP analysis).

Predictor variable	SHAP value	Impact direction	Interpretation	Relative importance
Mobile broadband penetration	0.211	+	Higher mobile connectivity reduces disruption risk	21.1%
Artificial intelligence (AI) demand forecasting accuracy	0.187	+	Improved AI forecasting decreases disruptions	18.7%
Regulatory quality index	0.162	+	Stronger regulations lower disruption likelihood	16.2%
Inventory turnover ratio	0.143	-	Faster inventory turnover increases disruption risk	14.3%
Cross-border logistics delay	0.121	-	Longer customs delays lead to more disruptions	12.1%

Policy simulation

Bayesian hierarchical modeling with 10,000 Monte Carlo iterations reveals that coordinated AI adoption and digital infrastructure improvements generate non-linear returns for renewable energy development in African frontier markets. The optimal scenario (40% AI adoption with targeted digital upgrades) yields 4680 MW in additional renewable capacity and 46% fewer supply chain disruptions, outperforming isolated interventions by 18% in cost–benefit efficiency. Digital infrastructure demonstrates threshold effects, with the strongest marginal impacts below 90% penetration (SHAP-derived inflection points), while AI adoption shows diminishing returns beyond 70% implementation. System GMM estimates confirm country-specific elasticities, with posterior predictive checks and historical back-testing validating the model. These findings, based on firm-level adoption metrics and mobile broadband penetration rates, demonstrate how strategic investments in AI-driven supply chains and priority digital upgrades (grid connectivity, customs automation, rural cloud access) can accelerate Africa's energy transition while enhancing resilience, consistent with recent evidence on technology-enabled development pathways.^37,53,56 The simulation framework provides policymakers with empirically grounded tools to identify cost-effective intervention points and anticipate non-linear system responses across varying implementation horizons. Figure 9 visually depicts the policy impact boxplot distribution of renewable capacity gains under AI adoption scenarios, while Table 6 gives the policy simulation outcomes under varying AI/digital infrastructure scenarios.

OPEN IN VIEWER

Figure 9.

Policy impact boxplot (renewable capacity gains). Notes: Boxplot showing renewable energy capacity gains under varying artificial intelligence (AI) adoption levels across policy scenarios, derived from Bayesian policy simulations. Source: Author via Python/R.

OPEN IN VIEWER

Table 6.

Policy simulation outcomes under varying artificial intelligence (AI)/digital infrastructure scenarios.

Scenario	AI adoption level	Digital infrastructure penetration	Projected RE capacity gain (MW)	Reduction in supply chain disruptions	Required investment (USD million)	Time horizon
Baseline (2023)	42%	58%	-	0%	-	-
Conservative	+15% (to 57%)	+20% (to 78%)	1820	22% reduction	380	5 years
Moderate	+30% (to 72%)	+35% (to 93%)	3450	37% reduction	740	5 years
Ambitious	+50% (to 92%)	+50% (to 108%)	5210	51% reduction	1250	5 years
Optimal*	+40% (to 82%)	+45% (to 103%)	4680	46% reduction	920	5 years

Hypothesis validation through advanced analytical techniques

Our empirical analysis employed robust methodological approaches to test the three proposed hypotheses. Table 7 shows the key analytical techniques. The SEM results provide compelling evidence for H1, revealing that AI adoption strengthens renewable energy development by enhancing SCR. This relationship manifests through both a significant indirect effect and a complementary interaction between AI and dynamic supply chains. For H2, the moderation analysis demonstrates that regulatory frameworks play a crucial amplifying role, with AI's effectiveness increasing significantly in environments with stronger governance. This effect becomes particularly pronounced when digital infrastructure penetration reaches 90%, as shown by our threshold analysis. The PTR results for H3 indicate that AI's positive impact on trade network efficiency is most substantial in low-connectivity markets, supporting our proposition about AI's transformative potential in frontier economies. These findings align with theoretical expectations from the Technology-Organization-Environment Framework ⁴⁹ and Dynamic Capabilities Theory, ⁴² while building metho7dologically on established approaches to mediation ⁸¹ and threshold effects. ⁷⁵ Collectively, the results underscore how AI-driven supply chain innovations, when combined with supportive institutional environments and digital infrastructure, can significantly accelerate Africa's renewable energy transition, as recent empirical work has suggested.^6,40

OPEN IN VIEWER

Table 7.

Key analytical techniques.

Mediation analysis (SEM)
Component	Specification	Expected outcome	Key metrics
Mediator (DSC)	DSC i t = β 0 + β 1 AI i t + β 2 Z i t + i t	AI adoption ( β 1 ) significantly improves DSCs.	β 1  = 0.35*** (p < 0.01)
Outcome (RE)	RE i t = γ 0 + γ 1 DSC i t + γ 2 AI i t + γ 3 Z i t + 2 , i t	DSC mediates AI's impact on renewable energy (RE) (γ₁ > γ₂).	γ₁ = 0.28** (p < 0.05); γ₂ = 0.12 (n.s.)
Indirect effect	β 1 . γ 1	Significant mediation (Sobel test: z = 2.89, p < 0.01).	Indirect effect = 0.098*
Controls (Z)	Digital infrastructure, policy environment	Digital infrastructure strengthens mediation.	β₂ (Digital) = 0.41***
Moderation analysis (hierarchical regression)
Interaction term	Specification	Interpretation	Results
AI × DSC	δ 3 ( AI i t . DSC i t )	Tests complementarity (H1: AI + DSC synergy).	δ 3  = 0.189* (p < 0.1)
AI × Digital infrastructure	δ 4 ( AI i t . DigInfra i t )	Digital infrastructure amplifies AI's impact (H2).	δ 4  = 0.204*** (p < 0.01)
AI × Regulatory qualityq	δ 5 ( AI i t . RegQual i t )	Strong policies enhance AI's effectiveness (H2).	δ 5  = 0.162** (p < 0.05)
Threshold effects	Non-linear returns beyond 70% AI adoption (H3).	Diminishing marginal gains post-threshold.	Threshold (τ) = 0.70
Threshold regression (panel threshold regression—PTR) for non-linear effects
Threshold variable	Specification	Hypothesis link	Key findings
Digital infrastructure (τ)	RE i t = θ 0 + θ 1 AI i t .1 ( DigInfra i t ≤ τ ) + θ 2 AI i t .1 ( DigInfra i t > τ ) + θ 3 X i t + λ t + i t	Tests H3 (trade network efficiency).	τ = 90% penetration (SHAP-derived)
Low connectivity	θ 1 ( DigInfra i t ≤ τ )	AI has stronger impact in low-connectivity markets.	θ 1 = 0.38***
High connectivity	θ 2 ( DigInfra i t > τ )	Diminishing returns in mature digital markets.	θ 2  = 0.12 (n.s.)

Notes: The notation “n.s.” in the statistical result (n.s.) stands for not statistically significant. DSC: dynamic supply chain; RE: random effect; AI: artificial intelligence.

PP test

The PP tests (Table 8) confirm all variables are integrated of order one [I(1)], with the first-differenced series showing strong stationarity as indicated by test statistics ranging from −3.95 to −5.12. This statistical property validates the use of bounds testing for cointegration analysis, ⁸² crucial for examining long-run relationships between AI adoption, supply chain dynamics, and renewable energy development in Africa's frontier markets. The non-stationarity at levels reflects the transitional nature of these economies, ⁷² while the I(1) characteristics enable robust modeling of technology-driven energy transitions. The results’ robustness to heteroskedasticity ⁸³ ensures reliable inference about nonlinear interactions between digital infrastructure, regulatory quality, and renewable capacity expansion across diverse national contexts. The trend of PP test statistics is visually depicted in Figure 10.

OPEN IN VIEWER

Figure 10.

Trends of Phillips–Perron (PP) test statistics. Notes: Time series of PP test statistics for key variables confirming non-stationarity and justifying the use of advanced econometric techniques. Source: Author via Synth (R).

OPEN IN VIEWER

Table 8.

PP unit root test results.

Variable	Level (I(0))	First difference (I(1))	Stationarity conclusion	Implications for analysis
Renewable energy capacity	−1.24 [0.65]	−4.87*** [0.000]	I(1)	Suitable for cointegration analysis
AI adoption index	−0.98 [0.76]	−5.12*** [0.000]	I(1)	Valid for long-run equilibrium tests
Supply chain resilience	−1.56 [0.50]	−3.95*** [0.001]	I(1)	Confirms dynamic relationships
Digital infrastructure	−2.01 [0.28]	−4.33*** [0.000]	I(1)	Supports threshold effects analysis
Regulatory quality	−1.89 [0.34]	−4.76*** [0.000]	I(1)	Valid for moderation analysis
GDP per capita (log)	−0.75 [0.83]	−5.04*** [0.000]	I(1)	Ensures reliable control variable

PP: Phillips-Perron; AI: artificial intelligence.

Independent test

The Brock–Dechert–Scheinkman (BDS) test statistics, conducted with embedding dimensions (m) of 2–3 and ε=1.5 as the distance parameter, uniformly reject the null hypothesis of independent and identically distributed data across all variables (Table 9). The significant BDS statistics confirm the presence of nonlinear dependence in renewable energy capacity, AI adoption, SCR, and digital infrastructure measures. These findings, consistent with the methodology of Brock et al. ⁸⁴ empirically justify the study's use of nonlinear modeling approaches to capture the complex interactions between AI adoption and dynamic supply chains in Africa's frontier energy markets. The results particularly support the application of threshold effects and interaction term specifications in modeling these technology-driven renewable energy transitions.

OPEN IN VIEWER

Table 9.

BDS test results for nonlinear independence.

Variable/dimension	Statistic	Std. Dev.	z-Stats	p-value	Conclusion
Renewable energy capacity (m = 2)	0.021	0.008	2.625	0.009	Reject H₀
Renewable energy capacity (m = 3)	0.035	0.012	2.917	0.004	Reject H₀
AI adoption index (m = 2)	0.018	0.007	2.571	0.010	Reject H₀
AI adoption index (m = 3)	0.029	0.010	2.900	0.004	Reject H₀
Supply chain resilience (m = 2)	0.015	0.006	2.500	0.012	Reject H₀
Supply chain resilience (m = 3)	0.025	0.009	2.778	0.005	Reject H₀
Digital infrastructure (m = 2)	0.017	0.007	2.429	0.015	Reject H₀
Digital infrastructure (m = 3)	0.028	0.010	2.800	0.005	Reject H₀

Notes: The BDS test evaluates the null hypothesis (H₀) of linear independence in the residual series. Results across all embedding dimensions (m = 2, 3) indicate rejection of H₀ at the 5% significance level, confirming the presence of nonlinear dependence in the examined variables. BDS: Brock–Dechert–Scheinkman; AI: artificial intelligence.

NARDL test

The nonlinear ARDL analysis reveals significant asymmetric effects in AI adoption's impact on renewable energy development (Table 10). Positive AI shocks exhibit stronger long-run effects compared to negative shocks, confirming path-dependent technology adoption in frontier markets. ⁷² The error correction term (−0.211, t = -3.89) indicates a 21.1% quarterly adjustment toward equilibrium, while the bounds test validates cointegration. With optimal lags (p = 2, q = 1) selected via AIC and strong model fit, these results demonstrate that AI-driven supply chain enhancements create persistent renewable energy capacity gains in Africa's frontier markets, supporting targeted deployment in high-impact nodes (Brock et al., 1996). The absence of residual autocorrelation (Q-test p = 0.32) further confirms the robustness of these asymmetric dynamics.

OPEN IN VIEWER

Table 10.

Nonlinear autoregressive distributed lag (NARDL) test results for asymmetric effects.

Variable	Panel A: Short-run coefficients (t-stat)	Panel B: Long-run coefficients (t-stat)	Significance	Lag structure (AIC)
AI adoption (positive shocks)	0.142** (2.31)	0.387*** (4.02)	p < 0.01	Lag 2
AI adoption (Negative shocks)	−0.091* (1.85)	−0.203** (2.54)	p < 0.05	Lag 2
Supply chain resilience	0.067 (1.42)	0.184*** (3.27)	p < 0.01	Lag 1
Digital infrastructure	0.053* (1.96)	0.129** (2.89)	p < 0.05	Lag 1
Regulatory quality	0.038 (1.12)	0.097* (1.98)	p < 0.1	Lag 1

Discussions

The empirical findings of this study both confirm and extend existing literature on AI-driven supply chains in Africa's renewable energy sector, while revealing novel insights specific to frontier markets. The robustness validation framework, comprising seven distinct methodological checks, establishes the reliability of our results beyond what previous studies have demonstrated. While Wamba et al. ¹⁶ validated composite indices for AI adoption in developed markets, our alternative measurement approaches using firm-level expenditure data confirm these metrics remain valid in African contexts, addressing concerns raised by Akinyele et al. ²⁰ about measurement validity in frontier economies.

The dynamic panel GMM results strongly confirm Ahmad et al's ⁵³ findings about AI's positive impact on renewable capacity while extending their work by quantifying the complementary effect between AI and SCR. This aligns with Modgil et al.'s ⁸ theoretical framework but contradicts Blimpo and Cosgrove-Davies, ¹² who argued for policy rather than technological solutions to Africa's energy challenges. Beyond these energy capacity gains, our analysis also identifies important economic spillovers. Specifically, increased AI adoption in renewable energy supply chains correlates with higher rates of SME participation in decentralized solar distribution networks and localized manufacturing of renewable components. ³⁶ Job market analyses conducted in parallel with this study indicate that each 10% increase in AI-driven supply chain efficiency is associated with a 4–6% rise in skilled technical employment, particularly in predictive maintenance services, drone-assisted site inspections, and AI-enabled energy trading platforms. ³⁰ Importantly, our results show that this relationship is not uniform across countries: interaction effects indicate that AI's marginal contribution to renewable deployment is up to 45% stronger in nations with above-median digital infrastructure and significantly weaker in markets with large AI workforce skills gaps. This heterogeneity highlights that enabling environments—comprising ICT infrastructure and human capital—are critical for fully realizing AI benefits.

The significant role of digital infrastructure supports Khalid's ²⁴ digital transformation thesis while adding nuance through our identification of a 90% penetration threshold beyond which marginal benefits diminish. Sensitivity analyses, including the exclusion of outlier countries and alternative index weightings, confirm the stability of these findings. Subsample analyses for high- vs. low-connectivity clusters further reveal that AI adoption yields more resilient supply chains and greater renewable capacity gains in digitally advanced economies, demonstrating robustness to model assumptions. Our machine learning analysis reveals mobile broadband penetration as the strongest protective factor against disruptions, confirming GSMA's ³⁷ industry findings while providing empirical validation through rigorous SHAP analysis. The predictive performance (AUC-ROC: 0.82) exceeds that reported in similar studies by Fioravanti et al., ¹⁷ suggesting frontier markets may derive greater relative benefits from AI applications than more developed contexts. Additional sensitivity tests indicate that removing the top and bottom 5% of AI-adopting countries does not materially change model accuracy or feature importance rankings, further reinforcing the reliability of these machine learning results.

The policy simulation results demonstrating 46% fewer disruptions under optimal AI adoption contrast with Mangla et al.'s ¹⁴ more pessimistic assessment of supply chain risks in developing markets while supporting Ukoba et al.'s ⁴⁰ arguments about technology-driven transitions. These simulation results also show that localized AI deployment promotes regional supply chain development, reducing import dependence by up to 15% over a five-year horizon and stimulating domestic entrepreneurship. Frontier energy markets like Kenya and Rwanda illustrate this trend, where AI-powered microgrid platforms have enabled the emergence of dozens of locally owned SMEs providing AI-augmented logistics and after-sales services. ¹¹ This aligns with recent game-theoretical approaches to stakeholder interaction in electricity markets,^19,85 which emphasize the role of cooperative strategies and incentive-compatible mechanisms in achieving efficient, decentralized renewable energy supply chains. Our findings similarly demonstrate that collaborative, multi-stakeholder decision-making amplifies AI's benefits, leading to more resilient and adaptive energy networks.

The SEM results for H1 (indirect effect) provide empirical validation for Teece's (2007) Dynamic Capabilities Theory in an African context, while the moderation analysis for H2 resolves the debate between Breyer et al. ⁷ and Armstrong et al. ³² by demonstrating that both technological and institutional factors are essential. The inclusion of digital infrastructure and skills gap moderators offers a more granular view of these factors, showing that institutional quality amplifies AI effects only when paired with adequate infrastructure and workforce capacity. The threshold regression results for H3 extend Onukwulu et al.'s ⁶ work by quantifying AI's disproportionate impact in low-connectivity markets. The PP and BDS tests confirming non-stationarity and nonlinear dependence validate our methodological approach while addressing methodological concerns raised by Eberhardt and Presbitero. ⁷¹ The NARDL analysis revealing asymmetric effects provides empirical support for Adedokun et al.'s ⁷² conceptual work on path-dependent technology adoption in frontier markets. Collectively, these findings highlight that AI adoption not only enhances energy capacity but also contributes to broader socio-economic development—creating new employment opportunities, fostering SME-driven innovation, and gradually reducing Africa's reliance on imported renewable technologies. This dual impact strengthens the case for policies that jointly promote AI deployment and inclusive local economic participation.

Policy implications

The empirical findings of this study yield critical policy insights that align with both theoretical frameworks and practical implementation challenges in Africa's renewable energy sector. Grounded in Dynamic Capabilities Theory and the Technology-Organization-Environment Framework, the results demonstrate that strategic interventions at the intersection of technological innovation and institutional strengthening can significantly accelerate energy transitions while enhancing SCR. Governments should establish dedicated AI innovation hubs focused on renewable energy supply chain optimization, with priority given to predictive analytics and blockchain traceability applications, which our AI adoption index weighted at 60% and 10% respectively. These hubs could be funded through blended finance mechanisms, combining development bank resources with private sector matching funds. For instance, the AfDB's Sustainable Energy Fund could be expanded to include an AI-specific grant facility, modeled after Kenya's successful M-Akiba renewable energy bond program that mobilized $5 million for clean energy projects through retail investor participation.

Digital infrastructure development requires targeted investment to achieve the 70–90% mobile broadband penetration threshold identified in our SHAP analysis as delivering optimal returns. Governments should implement universal service obligation funds specifically for renewable energy corridors, drawing lessons from Nigeria's Universal Service Provision Fund which allocated $25 million to expand broadband connectivity to rural communities. Our findings show that every 10% increase in mobile coverage reduces supply chain disruption risk by 2.1%, making such investments particularly impactful for energy project viability. Regulatory frameworks need to evolve to create testing environments for AI applications in energy supply chains. Morocco's Renewable Energy Regulatory Sandbox, which reduced approval times by 40% while maintaining safety standards, offers a replicable model. Energy ministries should establish AI regulation units with mandates to streamline approvals for logistics solutions while maintaining performance monitoring, as our results demonstrate that stronger regulatory quality (SHAP value = 0.162) significantly amplifies AI's effectiveness.

For research funding, governments should create competitive grant programs tied to measurable supply chain performance indicators. Our policy simulations show each dollar invested in AI-driven supply chain research yields $2.30 in renewable energy capacity gains. South Africa's Technology Innovation Agency provides a potential model, having funded over 200 technology projects through its competitive grant system. These programs could be structured as challenge grants where universities and private sector consortia compete to solve specific supply chain pain points identified in our analysis, such as cross-border delays, which showed a SHAP value of −0.121. The nonlinear ARDL results showing asymmetric effects of AI adoption (0.387 for positive shocks vs −0.203 for negative shocks) suggest implementing “AI adoption credit” systems. Renewable energy projects could earn tradable credits for verifiable AI implementation that reduces supply chain vulnerabilities, redeemable for tax incentives or preferential procurement treatment. India's Renewable Energy Certificate mechanism, which created a market for clean energy attributes, demonstrates how such systems can drive technology adoption.

As such, the strong predictive performance of our machine learning models (AUC-ROC: 0.82) supports creating national AI observatories for renewable energy systems. These institutions could provide real-time monitoring and early warning systems for supply chain disruptions, funded through a modest levy on large-scale renewable projects. Chile's National Center for AI, which receives $12 million in annual funding from government and industry partners, illustrates how such centers can bridge research and practical applications in emerging markets. Integrating game-theoretical perspectives ⁸⁵ into these observatories and policymaking frameworks would further enable governments to design participatory regulatory sandboxes, balancing incentives among diverse stakeholders—including SMEs, investors, and community cooperatives—to foster widespread AI adoption and equitable renewable energy transitions.

Moreover, policymakers must address foundational barriers to inclusive AI adoption, see Figure 11. Our findings highlight that limited ICT infrastructure and workforce skills gaps significantly constrain AI-driven supply chain performance. Governments should therefore launch digital infrastructure stimulus programs—such as subsidizing rural broadband networks and deploying low-cost cloud computing nodes—to lower the entry barrier for SMEs adopting AI tools. In parallel, capacity-building initiatives are essential. Ministries of education and labor, in collaboration with development partners, should establish AI and data analytics training academies modeled on Rwanda's Centre for the Fourth Industrial Revolution, which has successfully trained over 3000 local engineers. Targeted SME support programs, including AI adoption vouchers and fiscal incentives for renewable energy startups investing in supply chain automation, can foster equitable technological diffusion. By integrating these measures, policymakers can ensure that AI adoption benefits not only large utilities but also smaller enterprises and community-led renewable energy projects, promoting inclusive, sustainable energy transitions across Africa's diverse markets.

OPEN IN VIEWER

Figure 11.

Policy roadmap. Notes: Practical roadmap for policymakers showing sequential steps: Infrastructure development → skills training → SME artificial intelligence (AI) adoption → regulatory sandbox → AI observatory for supply chain monitoring.

Conclusions

This study provides robust empirical evidence that AI-enhanced dynamic supply chains serve as a transformative catalyst for renewable energy deployment across Africa's frontier markets. The findings demonstrate that AI adoption drives a 42.8% increase in renewable energy capacity per unit improvement in the AI index, while simultaneously reducing supply chain disruptions by up to 46% under optimal implementation scenarios. These results validate the application of Dynamic Capabilities Theory in frontier market contexts, showing how AI enables organizations to sense demand fluctuations through predictive analytics, seize opportunities via automated logistics, and reconfigure supply networks to overcome infrastructural constraints (Figure 12). Importantly, our analysis reveals significant cross-country heterogeneity in these effects. The impact of AI adoption is up to 45% stronger in markets with above-median digital infrastructure and smaller AI workforce skills gaps, whereas low-readiness countries experience substantially diminished benefits. These moderating effects underscore that technological infrastructure and human capital are essential complements to AI deployment. The identification of critical thresholds—70% for AI adoption, 90% for digital penetration—provides policymakers with precise targets for intervention and highlights that surpassing these thresholds yields disproportionately large renewable energy gains. The research resolves several theoretical debates in the literature, particularly the tension between technology-first and institution-first approaches, by demonstrating that AI's effectiveness is significantly amplified (δ = 0.162) in environments with stronger regulatory frameworks and digital infrastructure. Robustness checks, including alternative index weighting, outlier exclusion, and subgroup analyses, confirm that these findings are stable across methodological specifications. Sensitivity analyses further validate that removing the top and bottom 5% of AI-adopting countries or adjusting composite index weights does not materially alter the main results. The asymmetric effects revealed through nonlinear analysis underscore the importance of maintaining implementation momentum and preventing reversals in AI adoption policies. From a policy perspective, these results indicate that accelerating renewable energy deployment requires not only technological adoption but also strategic investment in ICT infrastructure and AI skills development, particularly to support small and medium enterprises. The study makes significant methodological contributions through its integration of advanced econometric techniques with machine learning approaches, offering a comprehensive framework for analyzing complex technology adoption dynamics in developing energy markets. By systematically accounting for cross-country heterogeneity and validating findings through extensive robustness checks, this research provides a reliable evidence base for policymakers and industry stakeholders seeking to design inclusive, scalable AI strategies for sustainable energy transitions in Africa.

OPEN IN VIEWER

Figure 12.

Artificial intelligence (AI) adoption and renewable energy efficiency. Source: Author via Gephi.

Limitations and future outlook

While this study provides important insights, several limitations warrant consideration in interpreting the results and designing future research. The reliance on firm-level adoption metrics, though carefully validated through multiple robustness checks, could be strengthened through more granular data collection partnerships with industry associations and regulatory bodies. The 10-year study period captures medium-term effects but cannot assess the full lifecycle impacts of AI adoption in energy systems, suggesting the need for longitudinal studies spanning 15–20 years to evaluate sustainability and adaptation requirements. The geographic focus on African frontier markets, while providing depth of analysis, raises questions about generalizability to other developing regions with different institutional and infrastructural characteristics. Future research should pursue three key directions. First, investigations into human-AI collaboration models could yield important insights about workforce transition requirements and skill development pathways in renewable energy supply chains. Second, the integration of circular economy principles with AI-driven supply chains presents a promising avenue for enhancing both sustainability and resilience. Third, comparative studies across different developing regions could help establish boundary conditions for the findings and identify context-specific adaptation requirements. The rapid evolution of AI technologies suggests the need for ongoing research to assess the applicability of these findings to emerging tools like generative AI and quantum computing for supply chain optimization. These future research directions, combined with the robust findings of this study, can inform both academic understanding and practical implementation of AI solutions for sustainable energy development in frontier markets.