As artificial intelligence is becoming a top priority for businesses and policymakers worldwide, understanding what makes a country competitive in AI-driven industries is crucial for ensuring success in the global economy. This column uses US import data over the period 1999–2019 to study the determinants of comparative advantage in AI-intensive industries. It shows that more STEM graduates, better digital infrastructure, larger markets, and less restrictions on digital trade give nations a competitive edge. A key challenge will be to enable countries to remain competitive while avoiding an AI arms race and ensuring nobody is left behind.
Artificial intelligence (AI) and machine learning are among the most significant driving forces of the global economy today. Applications include predictive maintenance, quality control, customisation, supply chain management, inventory optimisation, pricing algorithms, and customer service. AI and machine learning, coupled with big data, enable companies to increase their range of products and services and improve efficiency (Aghion and Bunel 2024, Babina et al. 2024). As a result, AI-related industries are among the fastest growing components of world trade.
While some studies have illustrated how AI and machine learning can foster trade (Baldwin 2018), less attention has been paid to what makes countries competitive in AI-driven industries. Although AI and machine learning are being widely implemented, their deployment varies significantly across sectors and countries. Hence, identifying the factors that promote or hinder their adoption can help governments and businesses win the AI race, dominate global markets, and foster growth. Such insights seem particularly needed in a period when governments are recognising the strategic importance of developing the AI industry and are mobilising funds for this purpose, such as the €200 billion public investments announced by the European Commission at the AI Action Summit in Paris in February 2025.
Trade patterns in AI-intensive industries
To address this question, in a recent work (Bonfiglioli et al. 2025a), we study the determinants of comparative advantage in AI-intensive industries using US import data from up to 68 countries across 79 4-digit manufacturing and service industries over the past two decades (1999–2019). We adapt a methodology from the trade literature (Romalis 2004, Chor 2010) that requires the combination of proxies for factors facilitating AI adoption at the country level (i.e. potential sources of comparative advantage) with data on AI intensity at the industry level.
One difficulty is that data on AI intensity are hard to come by, as there are no official statistics on it. To overcame this, we follow Bonfiglioli et al. (2024, 2025b) in measuring AI intensity as the relative employment of AI-related workers in the US. In turn, AI-related workers are defined as those employed in occupations that typically require knowledge of specialised software for machine learning and data analysis, as identified in the “Hot Technologies” section of the O*NET database. This measure shows that the top service industries for AI intensity include information and data processing, financial services, and other business services. Top manufacturing industries include the production of communication equipment, audio and video equipment, electronic products, navigational and control instruments, and computer equipment.
Next, we define ‘AI-intensive’ industries as those with a share of AI-related employment above the sample median, and we describe some new trade patterns in the data. First, we confirm the strong dynamism of AI-related sectors. In particular, comparing the growth rate of exports to the US in AI-intensive industries to that of other industries within countries, we find that the former have grown 27% faster. Second, we study the relative export performance of countries in the US market. We build a measure of revealed comparative advantage as the ratio of a country’s export share of AI-intensive industries to the global export share of those industries. A value greater than one indicates revealed comparative advantage of a country relative to all other countries in the sample, within the US market.
Figure 1 plots the average values of revealed comparative advantage over the sample period. The figure unveils significant variation in revealed comparative advantage across countries. The most technologically advanced economies, such as Ireland, Switzerland, Japan, the UK, and Northern European countries stand out for their high values. The index also exceeds one in countries like Mexico, Saudi Arabia, and Malaysia, due to their specialisation in industries such as automotive, chemicals, petroleum refining, insurance, and other business services, which exhibit relatively high levels of AI intensity. In contrast, the lowest values are observed in Latin America and Southeast Asia.
Figure 1 Revealed comparative advantage in AI-intensive industries
Notes: The map reports the average values of the revealed comparative advantage index. Averages are computed over all sample years.
Sources of comparative advantage in AI-intensive industries
We next turn to examine possible sources of comparative advantage. First, we focus on production factors that are critical in AI-intensive industries. The appropriate type of human capital is probably the most important input, as data analysts and engineers design and adjust the software that powers the algorithms. Therefore, our preferred proxy is the supply of bachelor’s, master’s, and PhD graduates in science, technology, engineering, and mathematics (STEM) disciplines from the OECD. Second, we consider factors that may enhance productivity in AI-intensive industries. We use the share of the population with internet access as a measure of a country’s digital infrastructure. Additionally, economies of scale can be important because AI-driven processes are easily scalable and because larger markets generate more data, which can make AI systems even more effective. Therefore, we use the total volume of exports as a proxy for market scale.
Finally, we also consider some policy measures. Although policy is clearly endogenous, including it in the analysis is nevertheless an interesting exercise. Given the synergies between algorithms and data flows, and following Sun and Trefler (2023), we use the Digital Services Trade Restrictiveness Index (DSTRI) developed by the OECD, which measures various dimensions of regulatory barriers in digital services trade and data flows across countries.
We find significant empirical support for all the main sources of comparative advantage. Figure 2 provides a graphical representation of our empirical results. It shows the relationship between countries’ exports in different industries and the sources of comparative advantage. To draw it, we have computed the total exports to the US from each country and year, separately for AI-intensive and non-AI intensive industries. The figure plots these exports to the US against possible sources of comparative advantage, after controlling for country fixed effects to neutralise differences in distance and other country-level determinants of trade. Moreover, to help visualise the data, the observations are grouped by deciles of the variable on the horizontal axis (the determinant of comparative advantage).
Figure 2 Exports, AI intensity, and country characteristics
Notes: The figure shows the average log of exports to the US, after controlling for country fixed effects, across ten deciles of the distribution of the country characteristic indicated on the horizontal axis of each graph. Solid lines and dots refer to the values and the linear regression lines for AI-intensive industries, while hollow dots and dashed lines refer to non-AI-intensive industries.
The plots reveal that in AI-intensive industries, countries’ exports increase sharply with each of the three sources of comparative advantage, while in other industries, they slightly decrease. Hence, countries with greater availability of scientific skills, more advanced digital infrastructure, and stronger economies of scale export relatively more to the US in industries with a higher AI intensity, consistent with these countries enjoying a comparative advantage in these industries. The bottom-right panel indicates instead that exports decrease across the deciles of the DSTRI index for AI-intensive industries, while this is not the case for other industries. Thus, countries with a more liberal environment for digitally enabled trade (lower deciles of DSTRI) export relatively more to the US in AI-intensive industries.
These results are confirmed when regressing exports to the US on an interaction term between the industry-level AI intensity and the country-level measures of comparative advantage, thereby exploiting the full power in the dataset. Moreover, they are robust to the inclusion of standard sources of comparative advantage considered in the literature, other measures of skill, and a host of fixed effects, as well as to the use of alternative econometric specifications to control for sample selection and confounding factors. Finally, by exploiting predetermined variation in the regressors and a historical instrument for STEM graduates, we also find that the results are unlikely to be driven by reverse causality.
Quantitatively, all factors driving comparative advantage in AI-intensive industries are roughly equally important in explaining differences in export volumes across countries and industries. For example, a larger population share of new STEM graduates, equivalent to the average difference between France and Mexico, is associated with 0.27 log points higher exports to the US in “Other General Purpose Machinery Manufacturings” (high AI intensity) compared to “Metal Forging and Stamping” (low AI intensity). Since 1999, however, the main driver of comparative advantage has been digital infrastructure, due to its greater time variation.
Conclusions
Our results identify factors and policy bottlenecks crucial for competitiveness in AI-intensive industries at the global level. At the same time, they provide insights into why building comparative advantage in these industries may be particularly important. Specifically, our data show that AI-intensive industries are among the most dynamic sectors in the global economy. Furthermore, these industries seem to be characterised by stronger scale economies compared to more traditional sectors, suggesting that comparative advantage may become self-reinforcing and that specialisation in these sectors could yield greater economic gains.
In light of this, our results underscore the importance of investing in the right skills, in digital infrastructure, and the role of market size to boost a country’s AI competitiveness. They also indicate that regulations can hinder AI adoption. Nevertheless, the growing integration of AI into business functions calls for appropriate policies (Gans et al. 2018). A key challenge for the global economy will be to enable countries to remain competitive in AI-intensive industries while avoiding an AI arms race and ensuring that nobody is left behind.
Source: cepr.org
