A new study has found significant differences in how emerging-market economies ( EMEs ) are positioned to reap the short-term productivity benefits of artificial intelligence ( AI ).

Such differences are likely to widen further over the medium to long term if gaps in AI preparedness persist, the Bank for International Settlements ( BIS ) warns in the study.

Released on February 17,  BIS Bulletin No 121 finds that differences in the use and effectiveness of AI are likely to reflect two main factors: the sectors of an economy and its AI preparedness.

“A country’s sectoral production structure plays a key role in shaping AI deployment,” it says, noting that AI-intensive industries like finance, education, and information are more prevalent in advanced economies.

AI preparedness – encompassing digital infrastructure, skills, and institutional capacity – shapes countries’ ability to absorb, deploy, and benefit from AI technologies.

Another source of difference is the high concentration of the global AI supply chain, which gives a few jurisdictions a structural advantage in capacity development and income generation.

Industry sectors

Compared with advanced economies, many EMEs are likely to see smaller near-term benefits as their output is less concentrated in sectors with intensive cognitive and information-processing tasks that can be enhanced by AI, the study says.

Industry-level exposures to AI and country GDP structures underscore this. Agriculture, transport, and construction show the lowest exposures. Their core activities rely more heavily on physical and manual tasks that are primarily affected by robotics rather than AI, the study says.

At the same time, the share of agriculture in total real value added is notably larger in EMEs than in advanced economies. The share of professional services is smaller, which tends to be less conducive to the deployment and effective use of AI .

AI preparedness

The study notes that the AI Preparedness Index of the International Monetary Fund ( IMF ) shows that advanced economies are generally better positioned than EMEs and low-income countries to adopt AI. This reflects their relatively strong digital infrastructure, innovation capacity, and regulatory frameworks.

Countries better prepared have greater proportions of workers in highly exposed jobs. South Korea and Singapore, along with Gulf countries like Saudi Arabia and the United Arab Emirates, score highly on digital infrastructure, including renewable energy and power grids.

“But many EMEs exhibit an uneven profile, particularly in terms of human capital and regulatory preparedness,” the study finds. For example, India, Turkey, and Brazil have relatively high AI-related skills – all above the world average in terms of skills penetration.

“However, as in the case of India, a substantial share of AI-related workers are employed abroad, effectively making the country a net exporter of AI talent and reflecting the higher international mobility of such workers.”   

Government capacity to design, implement, and oversee AI strategies is found to be another key element shaping effective AI deployment. In this area, six EMEs ranked in the Top 20 worldwide, with Singapore and South Korea in second and third place after the United States. Among the others, the UAE ranked 12th, Saudi Arabia 17th, China 18th, and Malaysia 19th.

Short-term impacts

In the short run, AI’s positive impacts on growth in EMEs are expected to lag those in advanced economies where AI investments are largely concentrated.

In sectors with high shares of cognitive and information-processing activities such as financial services, advanced economies are also likely to lead.

“However, AI-related activities have also fuelled a recent surge in exports from several EMEs, particularly in semiconductors and computing equipment,” the study notes.

Given these differences, an increase in the IMF index would raise average value-added growth from the global minimum by 0.6 percentage points in 15 advanced economies with little variation ( Luxembourg is the only outlier at more than 1.0 full point ).

But it would be only around 0.4 percentage points in 15 EMEs with considerable variation – ranging from less than 0.2 points in Indonesia to more than 0.8 points in Hong Kong.

The short-term impacts of AI on jobs are not clear. “EMEs have a larger share of workers in low-skill cognitive and clerical occupations that are susceptible to automation, raising the risk of initial job losses in specific tasks,” the study says. “Early evidence points to negative effects in call centres and business process outsourcing in India and the Philippines.”

At the same time, however, “labour market adjustments tend to unfold gradually, as firms often wait for clear productivity gains before reorganizing work”.

Divergence or convergence?

Over a decade, AI-driven 0.5% annual increases in total factor productivity – the efficiency with which inputs are turned into outputs – would raise average GDP in advanced economies by more than 2 percentage points compared with EMEs. In other words, considerable divergence.

But the authors also look at a scenario of partial convergence, whereby gaps in AI preparedness compared with the US shrink by half. In this case, average GDP in advanced economies rises by less than 1 percentage point.

The authors add that international trade and global linkages can significantly shape the cross-country distribution of AI-induced gains.

EMEs importing AI-intensive goods and services, for example, may benefit from lower prices, while those integrated into AI supply chains or specializing in AI-exposed sectors may experience stronger demand.

“As a result, trade openness can either amplify or mitigate domestic gains, depending on countries’ position in global value chains,” the study says. “In this context, open digital trade, lower barriers to technology transfer, and reskilling policies can support a more equitable distribution of AI gains.

“For EMEs, leveraging trade networks alongside continued investment in digital infrastructure, skills, and institutions will be essential to harness the full potential of AI,” it concludes.