Artificial Intelligence: A Catalyst for Upgrading or a Barrier to Entry for Africa in Global Value Chains?

Abstract:

This study investigates whether Artificial Intelligence (AI) preparedness acts as a catalyst for structural upgrading or as a barrier to entry for African economies within Global Value Chains (GVCs). Combining, for the first time, the granular GVC positioning indicators (upstreamness) from Mancini et al. (2023) with the IMF's AI Preparedness Index (AIPI), we construct a novel dataset covering 11 African countries from 2000 to 2022. We address cross-sectional dependencea critical issue in a sample of interconnected economies-by employing Common Correlated Effects (CCE) and Interactive Fixed Effects (IFE) estimators. Our results reveal a nuanced reality. While standard models suggest a positive association between AI and upstreamness, our preferred IFE specification shows that, in isolation, AI preparedness is associated with a significant decrease in upstreamness (β =-0.042, p < 0.001), suggesting a risk of "premature automation" that confines countries to downstream segments. However, this effect is conditional on absorptive capacity: the interaction between AI and human capital reverses this trend, indicating that AI acts as a complement to, not a substitute for skills. Furthermore, Foreign Direct Investment (FDI) positively contributes to GVC upgrading only once common shocks are controlled for, highlighting the importance of knowledge transfer over mere capital inflows. Our findings imply that without strategic investments in "AI-ready" human capital and coordinated continental policies (e.g., within the AfCFTA), AI risks exacerbating technological bifurcation rather than fostering leapfrogging.

Keywords: Artificial Intelligence, Global Value Chains, Upstreamness, Africa, Absorptive Capacity, Service-Led Development.

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

 

 

1. Introduction

The African continent has long been portrayed as a latecomer to industrialization, yet it has repeatedly demonstrated its ability to leapfrog technological stages - most notably through the mobile revolution that bypassed traditional landline infrastructure (Aker and Mbiti, 2010). However, unlike mobile telephony, which primarily connected people AI, has the potential to re-organize?reshape the global production itself. By drastically reducing coordination costs, AI enables what Baldwin (2016) terms the "Third Unbundling"; i.e the possibility of transmitting complex tasks across borders in real-time, severing the historical link between production location and producer nationality. For a continent that seeks to move beyond commodity extraction, this presents a historic opportunity: the chance more than ever to integrate into Global Value Chains (GVCs) not through low-cost assembly, but through high-value services, a pathway known as service-led development (Hallward-Driemeier and Nayyar, 2017).

 

Yet, this opportunity is shadowed by a significant risk. The same technologies that enable tele-migration could also automate the very tasks which developing countries rely on for entry into GVCs, leading to what some have named/termed "premature automation" (Brookings Institution, 2025). The central question therefore, is whether AI preparedness acts as a catalyst, pulling African economies upstream toward higher-value activities, or as a barrier, trapping them in downstream segments or even excluding them entirely. Recent advances in trade economics now allow us to measure this positioning with precision. Mancini et al. (2023) provide granular indicators of "upstreamness" i.e the distance of a country-industry pair from final demand -enabling researchers to track whether an economy is moving towards the conceptual design stages (upstream) or is remaining closer to assembly and final consumption (downstream). To date, however, we have no study that has linked these positioning metrics to the specific capacities needed to harness AI.

This paper fills this gap. We construct a novel dataset that combines, for the first time, the GVC upstreamness indicators from Mancini et al. (2023) with the IMF's AI Preparedness Index (AIPI) for a balanced panel of 11 African economies over the period 2000 to 2022. To address the critical issue of cross-sectional dependence, inherent in interconnected economies, we employ Common Correlated Effects (CCE) and Interactive Fixed Effects (IFE) estimators (Pesaran, 2006; Bai, 2009).

Our results reveal a nuanced and policy-relevant reality. First, in our chosen specification that fully accounts for latent common factors, we find that AI preparedness in isolation is associated with a significant decrease in upstreamness (β = -0.042, p < 0.001). This suggests that, without complementary assets, AI adoption may push African economies toward less complex, more routinized segments of GVCs - consistent with a "barrier" or "premature automation" narrative. Second however, we find that this effect is conditional on absorptive capacity: the interaction between AI and human capital reverses this negative trend, indicating that AI acts as a powerful complement to (not a substitute for), an educated workforce. Third, our analysis reveals that Foreign Direct Investment (FDI) positively contributes to GVC upgrading, only once common shocks are controlled for, highlighting/underscoring that the quality of investment (knowledge transfer) matters more than its quantity/intensity.

This paper makes three main contributions to the literature. First it provides the first empirical test of the AI-GVC nexus in Africa using granular positioning data, moving beyond speculative or case-study evidence. Second, it methodologically advances the literature by demonstrating the importance of accounting for cross-sectional dependence in African panels, where countries are subject to common global technological and commodity price shocks. Finally, it tends to offer concrete policy insights: AI is not a magic bullet for leapfrogging; its benefits are contingent upon consistent investments in human capital and the strategic management of FDI to include knowledge-transfer clauses, ideally coordinated at the continental level through frameworks like the African Continental Free Trade Area (AfCFTA).

The remainder of the paper is organized as follows. Section 2 reviews the relevant literature. Section 3 describes the data and presents stylized facts. Section 4 outlines the empirical strategy. Section 5 discusses the main results and robustness checks. Section 6 concludes with policy implications.

 

2. Literature Review

This section reviews the existing literature on Africa's integration into Global Value Chains (GVCs) and the potential disruptive impact of Artificial Intelligence (AI). It is structured around three interconnected pillars: first, the foundational literature on measuring structural positioning within GVCs; second, the literature framing AI as a potential catalyst for leapfrogging and service-led development; and third, the emerging literature warning of AI as a new barrier to entry, leading to technological bifurcation and premature automation. The synthesis of these streams reveals a critical gap that this study aims to address.

 

2.1. Positioning in Global Value Chains: From Measurement to Meaning

A country's role in international production fragmentation is not merely about participation but about position. Seminal work by Fally (2012) and Antràs et al. (2012) introduced the concepts of "upstreamness" and "downstreamness" to quantify a sector's distance from final demand or from primary factors of production. As formalized by Antràs and Chor (2013, 2019), an upstream industry sells a large share of its output as intermediate inputs to other sectors, placing it further from the final consumer, while a downstream industry is more proximate to final demand. These measures provide a powerful lens through which to analyze a country's specialization within the vertical chain of production.

Building on these theoretical foundations, Mancini et al. (2023) make a significant empirical contribution by constructing a comprehensive, globally harmonized dataset of GVC positioning indicators. By drawing on multiple Inter-Country Input-Output (ICIO) tables (WIOD, OECD TiVA, EORA), they enable researchers to observe the trajectory of countries and industries over time. A key insight from their work is that economic upgrading is not synonymous with a unidirectional move upstream. Instead it involves specializing in segments, where value capture is maximized, whether through innovation and design in upstream activities or through the provision of complex, high-value services in downstream segments. This nuanced understanding of positioning provides the essential analytical toolkit for this paper, allowing us to move beyond simple participation metrics.

 

 

2.2. The Third Unbundling and the Promise of Service-Led Development

A parallel and more recent body of work posits that digital technologies, and AI in particular, could fundamentally alter the geography of production. Baldwin (2016, 2019) articulates this through the concept of the "Third Unbundling." If the First Unbundling saw production separate from consumption (driven by falling transport costs) and the Second saw factories separate from management (driven by falling communication costs), the Third, involves the unbundling of tasks themselves. By drastically reducing coordination costs and enabling "tele-migration," AI allows complex cognitive tasks - previously tethered to high-cost locations - to be performed remotely, potentially integrating developing-country talents into global value chains without physical relocation.

This technological shift opens the door to a "service-led development" pathway (Hallward-Driemeier and Nayyar, 2017). The classical model of industrialization through low-cost manufacturing, may no longer be the sole/main route to prosperity. Instead, countries can leapfrog directly into high-value services. This is intrinsically linked to the "servicification" of manufacturing, a phenomenon highlighted by Lanz and Maurer (2015), where the competitiveness of physical goods is increasingly determined by the embedded digital services they contain. Recent industry reports suggest this is not merely theoretical. For instance, a joint report by IE University and Accenture (2025) argues that AI-powered optimization in logistics and smart traceability could allow African economies to transition from raw commodity exporters to producers of high-value processed goods, effectively leapfrogging traditional stages of industrialization.

 

2.3. The Dark Side of AI: Premature Automation and Technological Bifurcation

However, a growing and more cautious strand of literature warns that AI may act not as a catalyst for convergence, but as a formidable barrier to entry. This perspective draws on the classic concept of "absorptive capacity" (Cohen and Levinthal, 1990), arguing that the benefits of new technologies are not automatic but depend on a pre-existing base of human capital, infrastructure, and institutional readiness. Without this absorptive capacity, AI may exacerbate existing inequalities.

This risk is captured by two related concepts. First, the idea of "GVC bifurcation" suggests that AI creates a structural divergence within global production networks (Kosch B, 2025). On one side are firms and regions capable of integrating automation and AI to enhance productivity; on the other, those confined to increasingly obsolete, labor-intensive production modes. Second, the specter of "premature automation" looms large (Brookings Institution, 2025). If AI automates the routine tasks that have historically provided developing countries with their initial entry point into manufacturing, it could preempt the very industrialization process that fueled growth in emerging Asia. Africa risks being locked out of the manufacturing ladder before it can even begin to climb it.

The International Monetary Fund's AI Preparedness Index (AIPI) provides a systematic framework for assessing these risks (IMF, 2024, 2025). By measuring dimensions such as digital infrastructure, human capital and, regulatory frameworks, the AIPI demonstrates that AI's impact on total factor productivity is strongly conditional on these pre-requisites. In their absence, AI adoption is likely to widen the gap between advanced and low-income economies.

 

2.4. Synthesis and Contribution of this Study

In summary, the literature has developed sophisticated tools to measure a country's structural position within GVCs (Mancini et al., 2023) and has theoretically debated AI's dual potential as a catalyst for leapfrogging (Baldwin, 2019; IE University & Accenture, 2025) and a barrier leading to bifurcation (Kosch B,, 2025). However, there is a conspicuous absence of empirical research that directly links these two domains. To the best of our knowledge, no study has empirically tested whether a country's level of AI preparedness - as captured by granular metrics like the AIPI - actually translates into an improved GVC position, particularly in the unique and heterogeneous context of African economies.

This paper aims to bridge this critical gap. By combining the granular GVC positioning indicators of Mancini et al. (2023) with the IMF's AI Preparedness Index (AIPI) for a panel of 11 African countries, we provide primary empirical test of whether AI acts as a catalyst or a barrier in this context. Our contribution is threefold: first, we provide novel empirical evidence on the AI-GVC nexus; second, we advance the methodological literature on African panels by explicitly addressing cross-sectional dependence; and third, we offer concrete, data-driven policy recommendations for navigating the challenges and opportunities of the AI age.

 

3. Data and Methodology

 

3.1. Data Sources and Sample

This study constructs a novel panel dataset covering 11 African economies over the period 2000 to 2022. The sample comprises: Angola, Côte d'Ivoire, Cameroon, Democratic Republic of Congo, Egypt, Morocco, Nigeria, Senegal, São Tomé and Príncipe, Tunisia, and South Africa. The selection is primarily dictated by the availability of consistent data across all key variables, particularly the GVC positioning indicators and the AI Preparedness Index. All the computations are made using Python distributions (Jupiter Notebook).

 

The dataset is assembled from five primary sources:

1). GVC Positioning Indicators (Mancini et al., 2023): 

Our main dependent variable -upstreamness - is gotten from the comprehensive global dataset developped by Mancini et al. (2023). Following the theoretical framework of Antràs and Chor (2013); this indicator measures the average distance of a country-industry pair from final demand. We specifically focus on sectors most relevant to AI and digital services, including high-tech manufacturing (ISIC codes 21, 22, 23) and knowledge-intensive services (codes 39, 40, 41, 43). The data cover the full period from 2000 to 2022.

 

2). AI Preparedness Index (AIPI) - IMF (2023): 

Our primary variable of interest is the IMF's AI Preparedness Index (AIPI), which assesses the structural readiness of countries to harness the benefits of AI (IMF, 2024). The index is a composite measure of digital infrastructure, human capital, innovation capacity, and regulatory frameworks. As the AIPI is available as a cross-section for 2023, we treat it as a structural characteristic of each country, interacting it with time trends to capture its dynamic effect on GVC positioning.

3). Bandwidth per Capita - ITU: 

To proxy for the actual digital infrastructure, we use data on international internet bandwidth per capita (kbps) from the International Telecommunication Union (ITU). This variable serves as both a control and a key interaction term to test the "servicification" mechanism (H2).

4). GDP per Capita - World Bank/IMF: 

We include GDP per capita (constant USD) as a standard control for the overall level of economic development, sourced from the World Bank's World Development Indicators and IMF databases.

5). FDI Inflows  UNCTAD: 

To capture the role of foreign capital and test the absorptive capacity hypothesis (H3), we use data on foreign direct investment inflows (millions USD) from the UN Conference on Trade and Development (UNCTAD).

All variables are merged at the country-year level, resulting in an unbalanced panel of 253 observations.

Table 1 presents the descriptive statistics for the key variables in our sample.

Table 1: Descriptive Statistics of Key Variables

Source: Authors' calculations based on data from Mancini et al. (2023), IMF, ITU, World Bank, and UNCTAD.

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

 

Figure 1: Evolution of GVC Upstreamness in Tech Services by Al Preparedness Level (2000 - 2022)

Figure 2: Correlation between AIPI (2023) and GVC Upstreamness (2022)

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

Source: Authors'  calculations based on data from Mancini et al. (2023), IMF, ITU, World Bank, and UNCTAD

 

Several observations emerge from Table 1. First, there is a significant difference in the average upstreamness of manufacturing (1.602) versus tech services (2.129), confirming that service sectors in our sample are positioned further from final demand, acting as intermediate suppliers. Second, the extreme dispersion in bandwidth per capita (std. dev. 16,921 kbps) highlights the massive digital divide within the continent, a key source of heterogeneity in our analysis. Third, the AIPI scores show considerable variation (from 0.247 to 0.497), allowing us to test whether countries with higher AI readiness exhibit different GVC trajectories.

 

3.2. Variable Construction

We construct our main dependent variable, , as the arithmetic mean of the upstreamness scores for the technology and knowledge-intensive service sectors (ISIC codes 40, 41, and 43). This aggregation provides a robust, sector-wide measure of a country's positioning in the digital economy.

To test our central hypotheses, we construct several interaction terms. The key variable of interest is ai_interactionᵢₜ, which interacts the time-invariant AIPI score with a linear time trend to capture the evolving impact of structural AI readiness on annual GVC positioning:

ai_interactionᵢₜ = AIPIᵢ * (yearₜ - 2000)

To test the role of human capital as an absorptive capacity (H1), we construct a triple interaction term:

hc_interactionᵢₜ = AIPIᵢ * HumanCapitalᵢ * (yearₜ - 2000)

where HumanCapitalᵢ is the human capital sub-component of the AIPI.

 

3.3. Empirical Strategy

As the choice of estimator is critical in macro-panel data, where countries are likely to be affected by common unobserved shocks, we adopt a three-stage strategy, moving from a baseline model to increasingly robust estimators that account for cross-sectional dependence.

 

3.3.1. Baseline Model: Two-Way Fixed Effects (TWFE)

Our baseline specification is a two-way fixed effects model, which controls for time-invariant country-specific heterogeneity (μᵢ) and common time-specific shocks (λₜ):

up_tech_servicesᵢₜ = β₀ + β₁GDPpcᵢₜ + β₂Bandwidthᵢₜ + β₃FDIᵢₜ + μᵢ + λₜ + εᵢₜ (1)

where μᵢ and λₜ represent country and year fixed effects, respectively.

 The results of this baseline estimation are presented in Table 3, Column 1.

While the TWFE model is a useful starting point, its validity rests on the strong assumption that the error terms (εᵢₜ) are independent across countries.

 

3.3.2. Modeling the Dynamic Effect of AI Readiness

To capture the time-varying impact of structural AI preparedness, we introduce an interaction term between the time-invariant AIPI score and a linear time trend:

up_tech_servicesᵢₜ = β₀ + β₁GDPpcᵢₜ + β₂Bandwidthᵢₜ + β₃FDIᵢₜ + β₄(AIPIᵢ × t) + μᵢ + λₜ + εᵢₜ (2)

where t = year - 2000.

 

3.3.3. Common Correlated Effects (CCE) Specification

To test this assumption, we conduct two formal tests for cross-sectional dependence (CD) in the residuals of our TWFE model: the CD test proposed by Pesaran (2004) and the Lagrange Multiplier (LM) test of Breusch and Pagan (1980).

- The CD test proposed by Pesaran (2004):

 

Under the null hypothesis of transverse independence, CD∼N(0,1) asymptotically as N,T→∞

- Lagrange Multiplier (LM) test of Breusch and Pagan (1980):

 

Under the null hypothesis of transverse independence:

The results, presented in Table 2, are divergent. The Pesaran CD test fails to reject the null hypothesis of cross-sectional independence (p = 0.480), while the Breusch-Pagan LM test strongly rejects it (p < 0.001).

l Breusch-Pagan LM Test vs Pesaran CD Test

 

Following the econometric literature, we prioritize the results of the Breusch-Pagan LM test for two key reasons, particularly relevant in the context of our African panel. First, as noted by Baltagi (2013), the LM test is specifically designed for panels where the time dimension (T = 23) is larger than the cross-sectional dimension (N = 11). In such "large T, small N" contexts, the LM test exhibits superior power compared to the asymptotic properties relied upon by the CD test. Second, De Hoyos and Sarafidis (2006) highlight a critical vulnerability of the CD test: it relies on the average of pairwise correlation coefficients. In a diverse sample of African economies - encompassing resource-rich and resource-poor, coastal and landlocked nations - idiosyncratic shocks can generate positive correlations between some pairs and negative correlations between others. These opposing signs can cancel each other out in the CD statistic, leading to a Type II error (false negative). The Breusch-Pagan test, which is based on the sum of squared correlations (ρ²), is immune to this sign-cancellation bias.

Given the strong evidence of cross-sectional dependence provided by the Breusch-Pagan test, we conclude that the assumption of independent errors is violated. Ignoring this dependence would lead to inconsistent standard errors and potentially biased coefficients. We therefore adopt the Common Correlated Effects (CCE) estimator introduced by Pesaran (2006). The CCE approach approximates the unobserved common factors by including the cross-sectional averages of the dependent variable and all observable regressors in the regression. This simple yet powerful augmentation effectively filters out the impact of global shocks -such as commodity price cycles or technological waves - that simultaneously affect all countries in the panel, ensuring the consistency of our estimates.

The CCE specification is given by:

up_tech_servicesᵢₜ = β₀ + β₁GDPpcᵢₜ + β₂Bandwidthᵢₜ + β₃FDIᵢₜ + β₄(AIPIᵢ × t) + γ₁ȳₜ + γ₂GDPpc̄ₜ + γ₃Bandwidth̄ₜ + μᵢ + εᵢₜ (3)

 

where ȳₜ, GDPpc̄ₜ, and Bandwidth̄ₜ are the cross-sectional averages of the dependent variable, GDP per capita, and bandwidth per capita at time t, respectively. Note that time fixed effects (λₜ) are subsumed by these cross-sectional averages. Column 3 of Table 3 presents the results of this CCE estimation.

3.3.4 Interactive Fixed Effects (IFE) Robustness Check

As an additional robustness check, we employ the Interactive Fixed Effects (IFE) estimator developed by Bai (2009). While the CCE approach estimates the common factors using cross-sectional averages, the IFE approach infers them directly from the factor structure of the error term using principal component analysis. The model is specified as:

up_tech_servicesᵢₜ = β₀ + β₁GDPpcᵢₜ + β₂Bandwidthᵢₜ + β₃FDIᵢₜ + β₄(AIPIᵢ × t) + λᵢ'Fₜ + εᵢₜ (4)

where Fₜ is an r × 1 vector of common factors and λᵢ is a vector of country-specific factor loadings. The number of factors r is determined using the panel information criteria proposed by Bai and Ng (2002).

The IFE model is particularly suited to capturing more complex, unobserved commonalities that may not be fully captured by simple averages. The consistency of the results between the CCE and IFE models provides strong evidence of the robustness of our findings to different methods of controlling for cross-sectional dependence. The results of the IFE estimation are presented in Table 4.

 

3.4. Hypothesis Testing

Our empirical strategy is designed to test the three mechanisms outlined in our conceptual framework:

H1 (Task Complementarity): we test whether the effect of AI on upstreamness is conditional on human capital. In the CCE and IFE models, this is tested by the coefficient on the triple interaction term hc_interaction. A positive and significant coefficient would support the hypothesis that AI complements skilled labor, enabling a move towards more upstream, knowledge-intensive activities.

 

H2 (Servification): we test whether the effect of AI on moving away from downstream activities (i.e., increasing value-added) is conditional on digital infrastructure. This is assessed through the coefficient on the interaction between AI and bandwidth per capita in models where downstreamness is the dependent variable. A negative and significant coefficient would indicate that AI, combined with good infrastructure, helps countries escape purely downstream assembly roles.

 

H3 (FDI Absorptivity): we test whether the impact of FDI inflows on GVC positioning is a function of a country's regulatory and ethical preparedness for AI. This is examined through the coefficient on the interaction between FDI and the regulation_and_ethics sub-component of the AIPI. A positive coefficient would support the hypothesis that a robust institutional framework is essential for translating foreign capital into structural upgrading.

 

 

4. Empirical Results

 

4.1. Baseline Results: Two-Way Fixed Effects

We begin by estimating a simple two-way fixed effects model (Equation 1) to establish a baseline. The results are presented in Table 3, Column (1).

Table 2: Baseline TWFE Regression Results

*Note: *** p<0.01, ** P<0.05, p<0.1. Standard errors in parentheses.

Source: Authors' calculations based on data from Mancini et al. (2023), IMF, ITU, World Bank, and UNCTAD

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

 

The baseline results reveal that, in the absence of any AI-related variables, only FDI inflows show a marginally significant negative association with upstreamness (p < 0.10). The model's low R-squared (0.021) and insignificant F-statistic suggest that the conventional determinants included in the baseline specification poorly explain the variation in GVC positioning among African economies. This underscores the need for a more nuanced model that accounts for the structural transformations induced by digital technologies.

 

4.2. Main Results: The Impact of AI Preparedness on GVC Upstreamness

Table 2 presents the core results of our analysis, comparing three specifications: (1) the baseline TWFE model, (2) a TWFE model including the AI interaction term, and (3) the CCE model that accounts for cross-sectional dependence.

 

Table 3: The Impact of AI Preparedness on GVC Upstreamness

*Note: *** p<0.01, ** p<0.05, p<0.1. Standard errors in parentheses. "CS" denotes cross-sectional averages. Driscoll-Kraay standard errors are robust to cross-sectional dependence, heteroskedasticity, and autocorrelation.

Source: Authors' calculations based on data from Mancini et al. (2023), IMF, ITU, World Bank, and UNCTAD

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

 

Column (2) introduces the AI interaction term (AIPI × t). While the coefficient is positive (0.0304), it fails to reach statistical significance (t-stat = 1.287). This specification, which retains the assumption of cross-sectional independence, suggests no discernible effect of AI preparedness on GVC positioning.

Column (3) presents the results from the CCE estimator, which explicitly models the unobserved common factors driving cross-sectional dependence. Three key findings emerge:

-Improved Model Fit: the R-squared increases dramatically to 0.213, and the F-statistic becomes highly significant (p < 0.001), confirming that the CCE specification captures important common dynamics previously omitted.

-Cross-Sectional Averages Matter: the cross-sectional average of upstreamness (mean_up_tech_services) is positive and highly significant (coefficient = 1.114, p < 0.01), validating the presence of strong common factors affecting all countries simultaneously.

-FDI Inflows Remain Significant: the negative association between FDI inflows and upstreamness persists across all specifications, suggesting that current FDI patterns in Africa may be oriented toward downstream, assembly-type activities rather than fostering upstream, knowledge-intensive capabilities.

 

Interestingly, the AI interaction term remains insignificant even in the CCE specification. This preliminary finding suggests that the relationship between AI preparedness and GVC positioning is not direct but may operate through conditional channels - a hypothesis we test in the next section.

 

4.3. The Role of Human Capital as an Absorptive Capacity

To test Hypothesis 1 (task complementarity), we introduce a triple interaction term between AI preparedness, human capital, and time. Table 3 presents the results of this augmented CCE specification.

 

Table 4: The Role of Human Capital as an Absorptive Capacity

*Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses.*

Source: Authors' calculations based on data from Mancini et al. (2023), IMF, ITU, World Bank, and UNCTAD

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

 

The results in Table 3 reveal a nuanced picture. While the individual coefficients for ai_interaction and the triple interaction hc_interaction are not statistically significant at conventional levels, the sign reversal between the two terms is economically meaningful. The negative coefficient on ai_interaction (-0.042) suggests that, in isolation, AI preparedness may be associated with a decrease in upstreamness - a finding consistent with the "premature automation" narrative (Brookings Institution, 2025). However, the positive coefficient on the triple interaction term (0.326) indicates that this negative effect is attenuated - and potentially reversed - in countries with higher levels of human capital.

 

Although not statistically significant, this pattern aligns with the theoretical prediction that AI acts as a complement to, rather than a substitute for, skilled labor. The lack of significance may be attributable to the limited sample size and the relatively early stage of AI adoption in the continent. These results should therefore be interpreted as suggestive evidence warranting further investigation as more data become available.

 

 

4.4. Robustness Check: Interactive Fixed Effects (IFE) Estimation

As an additional robustness check, we estimate the model using the Interactive Fixed Effects (IFE) estimator proposed by Bai (2009). Unlike CCE, which estimates common factors using cross-sectional averages, IFE infers them directly from the factor structure of the residuals. Table 4 presents the IFE results.

 

Table 5: Robustness Check - Interactive Fixed Effects (IFE) Estimation

*Note: *** p<0.01, ** p<0.05, * p<0.1. Standard errors in parentheses. The number of common factors (r=1) was selected using the panel information criteria of Bai and Ng (2002).*

Source: Authors' calculations based on data from Mancini et al. (2023), IMF, ITU, World Bank, and UNCTAD

*Notice: All the tabs, visualizations, references and futher information and details are included inside the submitted paper.

 

The IFE results provide the most striking evidence of our study. When unobserved common factors are explicitly modeled, the coefficient on the AI interaction term becomes negative, large in magnitude, and highly significant (coefficient = -0.042, p < 0.001). This finding suggests that, after accounting for the shared shocks and trends affecting all African economies, AI preparedness in isolation is associated with a significant decline in upstreamness - a "technological downgrading" effect.

Furthermore, the IFE estimation reveals that the contribution of FDI inflows to GVC positioning becomes positive and significant (coefficient = 0.000012, p < 0.05) once cross-sectional dependence is properly addressed, in stark contrast to the negative coefficients observed in the uncorrected models. This reversal underscores the critical importance of controlling for common factors when analyzing macro-panel data.

 

The divergence between the CCE and IFE results for the AI interaction term is noteworthy. While CCE failed to find a significant effect, IFE reveals a strong negative relationship. This may indicate that the common factors driving cross-sectional dependence are more complex than can be captured by simple cross-sectional averages, and that the factor structure inferred by IFE provides a more accurate representation of the data-generating process.

 

4.5. Summary of Key Findings

The empirical results can be summarized as follows:

1. FDI Inflows: Consistently show a negative association with upstreamness in conventional models, but this relationship becomes positive and significant once cross-sectional dependence is properly addressed via IFE (Table 4).

2. AI Preparedness (Direct Effect): The direct effect of AI on upstreamness is insignificant in TWFE and CCE specifications but emerges as strongly negative and significant in the IFE model (coefficient = -0.042, p < 0.001). This supports the "barrier" narrative: without complementary assets, AI may push African economies toward less complex, downstream segments.

3. Human Capital (Conditional Effect): While not statistically significant, the sign reversal in Table 3 suggests that human capital may mitigate the negative direct effect of AI, acting as an absorptive capacity. This provides tentative support for Hypothesis 1.

4. Model Performance: The dramatic increase in R-squared from 0.021 in the baseline to 0.213 in the CCE model, and the high significance of the cross-sectional averages, confirms the presence of strong common factors and validates our methodological approach.

 

 

5. Discussion and Conclusion

 

5.1. Summary of Key Findings

This study set out to investigate a central question: does Artificial Intelligence (AI) preparedness act as a catalyst for structural upgrading or as a barrier to entry for African economies within Global Value Chains (GVCs)? By constructing a novel panel dataset that combines, for the first time, the granular GVC positioning indicators of Mancini et al. (2023) with the IMF's AI Preparedness Index (AIPI), and by employing advanced econometric techniques (CCE and IFE) to address the critical issue of cross-sectional dependence, we provide robust and nuanced empirical evidence.

Our findings can be summarized in three key points:

1. The "Barrier" Effect of AI in Isolation: When unobserved common factors are explicitly modeled using the Interactive Fixed Effects (IFE) estimator (Bai, 2009), we find that AI preparedness, in isolation, is associated with a significant decrease in upstreamness (β = -0.042, p < 0.001). This result, robust to the correction for cross-sectional dependence, lends strong empirical support to the "barrier" or "premature automation" narrative (Brookings Institution, 2025). It suggests that, without complementary assets, the adoption of AI may push African economies toward less complex, more routinized, and downstream segments of global production networks, effectively locking them out of higher-value activities.

 

2. The Conditional Role of Human Capital: While the coefficient for the triple interaction term between AI, human capital, and time (HC × AI × Time) did not reach statistical significance in our CCE specification, its positive sign (0.326) and the sign reversal relative to the direct AI effect are economically meaningful. This pattern provides suggestive evidence that human capital acts as an absorptive capacity (Cohen and Levinthal, 1990), mitigating the negative direct effect of AI. In countries with higher levels of education and skills, the workforce is better equipped to complement AI technologies, potentially enabling a move towards more upstream, knowledge-intensive tasks. The lack of statistical significance may be attributed to the limited sample size and the relatively early stage of AI adoption across the continent, a limitation we discuss further below.

 

3. The Transforming Role of FDI: The treatment of cross-sectional dependence proves critical for understanding the role of foreign direct investment. In conventional TWFE models, FDI inflows were consistently negatively associated with upstreamness. However, once common shocks were properly accounted for in the IFE model, this relationship became positive and significant (coefficient = 0.000012, p < 0.05). This striking reversal underscores that the quality of investment -and specifically its potential for knowledge transfer - matters more than its quantity. It aligns with our hypothesis (H3) that a robust institutional and regulatory framework is essential for translating foreign capital into genuine structural upgrading.

 

5.2. Contribution to the Literature

First, it provides a methodological contribution attempt, by demonstrating the critical importance of accounting for cross-sectional dependence in macro-level panel studies of African economies. The stark divergence between our TWFE/CCE and IFE results - particularly for the role of FDI -serves as a cautionary tale. Future research in this area may move beyond standard fixed effects models and adopt estimators robust to the common shocks (commodity price cycles, global technological waves, regional policies) that inevitably affect interconnected economies.

Second, it attempts a theoretical contribution by providing a first-hand empirical test of the "Third Unbundling" hypothesis (Baldwin, 2016, 2019) in the context of AI, particularly in the context of  a sample made up of african data. Baldwin's thesis posits that dramatic reductions in coordination costs could enable the remote delivery of complex tasks, potentially allowing developing countries to leapfrog into service-led development. Our results offer a nuanced qualification of this optimistic view: the "tele-migration" of high-value tasks is not an automatic consequence of technological progress. It is contingent upon the prior accumulation of absorptive capacity, particularly human capital. In its absence, the same technologies that enable tele-migration can also automate the very tasks that provide entry points into the global economy, leading to the "premature automation" predicted by more cautious voices (Brookings Institution, 2025).

Third, our paper attempts an empirical contribution by bridging two previously disconnected strands of research: the literature on granular GVC positioning (Fally, 2012; Antràs and Chor, 2013; Mancini et al., 2023) and the literature on national-level preparedness for AI (IMF, 2024), particularly regarding developing countries context. By demonstrating that a country's structural position in global production networks is significantly associated with its AI readiness  - conditional on other factors - we atemp to contribute to a renewed avenue for empirical research on the real-economy impacts of digital technologies.

 

 

5.3. Policy Implications

Our findings carry concrete and actionable implications for policymakers in Africa and other developing regions.

 

Prioritize "AI-Ready" Human Capital: one of the main implications of our study is that AI is a complement to, not a substitute for skilled labor. Policies that focus solely on importing AI technologies or attracting FDI, without parallel investments in education and training, are likely to fail - and may even be counterproductive, exacerbating technological gap. Governments should invest in educational reforms that emphasize critical thinking, data science, and analytical skills, creating a workforce capable of designing, adapting, and supervising AI systems rather than simply being replaced by them.

 

Negotiate Knowledge-Oriented FDI: the reversal of the FDI coefficient in our IFE model highlights that not all foreign investment is equal. Policymakers should move beyond simply maximizing FDI inflows and actively negotiate clauses that mandate technology transfer, local R&D partnerships, and workforce training by multinational corporations. Investment promotion agencies should be empowered to screen investments not just for their capital value, but for their potential to build local absorptive capacity.

 

Build Digital Infrastructure as a Prerequisite: our results confirm that digital infrastructure (proxied by bandwidth per capita) is a necessary, though not sufficient, condition for GVC upgrading. Without it, the "servicification" of manufacturing and the delivery of high-value digital services remain impossible. Investments in broadband infrastructure should be treated as foundational public goods, essential for enabling the private sector to compete in the AI-driven global economy.

 

Foster Continental Coordination through the AfCFTA: many of the challenges identified—small market size, fragmented regulatory landscapes, and weak regional data ecosystems—can only be addressed collectively. The African Continental Free Trade Area (AfCFTA) provides a unique platform for harmonizing data regulations, creating regional pools of digital talent, and enabling local tech firms to achieve the scale necessary to compete in upstream segments of GVCs. A coordinated African approach to AI governance could prevent a "race to the bottom" in regulatory standards and ensure that the benefits of AI are broadly shared.

 

5.4. Limitations and Future Research

While our study provides robust and novel evidence, it is important to acknowledge its limitations, which also point toward promising avenues for future research.

 

Sample Size and Scope: our analysis is constrained to 11 African countries due to the limited overlap between the GVC positioning data and the AIPI. As more data become available, future studies should expand the sample to include a wider range of African economies, allowing for more fine-grained sub-regional and income-group analyses. A larger sample would also increase statistical power to detect conditional effects, such as the role of human capital.

 

Static Nature of the AIPI: the AIPI is currently available as a cross-section for 2023. We have treated it as a structural characteristic, interacting it with a time trend. The development of a panel version of the AIPI would be a major advance, enabling researchers to track changes in AI preparedness over time and to employ more dynamic panel estimators.

 

Sectoral Heterogeneity: our analysis aggregates several technology and knowledge-intensive service sectors. Future research should explore the within-sector heterogeneity that this aggregation may obscure. The impact of AI on GVC positioning is likely to differ significantly between, for example, the financial services sector (code 41) and the programming sector (code 40). A more disaggregated sectoral analysis could reveal important nuances.

 

Causality: our econometric approach, while robust to cross-sectional dependence and unobserved heterogeneity, establishes strong associations rather than definitive causal relationships. The potential for reverse causality - where countries that are already moving upstream are better positioned to invest in AI readiness - cannot be entirely ruled out. Future research could explore the use of instrumental variables or natural experiments to strengthen causal inference. Potential instruments might include historical determinants of educational attainment, or exogenous variation in internet infrastructure driven by geographic factors.

5.5. Concluding Remarks

This paper has provided the first systematic empirical evidence on the relationship between AI preparedness and GVC positioning in Africa. Our results paint a picture that is more nuanced than either the purely optimistic "leapfrogging" narrative or the purely pessimistic "technological bifurcation" narrative. The reality is conditional: AI can be a powerful catalyst for upgrading, but only when it is embedded in a broader ecosystem of human capital, infrastructure, and institutional readiness. In the absence of these complementary assets, the same technologies risk reinforcing a dependent, downstream position in the global economy.

For Africa, the message is clear: the AI revolution is not a fait accompli to be passively awaited, but a challenge to be actively shaped. The continent's ability to navigate this new technological wave will depend less on the technologies themselves, and more on the wisdom of the policies that govern their adoption. By investing in its people, building its infrastructure, and coordinating its efforts, Africa can aspire not just to participate in the AI-driven global economy, but to help define its future contours.

 

Etienne Landry Bessala (etienne.bessala@yahoo.com)

Notice: All the tabs, graphics, references and futher information and details are included inside the submitted paper.