Emerging Markets Economic Outlook & Strategy: Artificial Intelligence and the EM Growth Divide

A new Citi Research report from a team led by Head of Emerging Markets Economics Johanna Chua explores the possibility of an emerging-markets growth divide in the scramble for benefits from artificial intelligence (AI). There are only a few potential emerging-markets winners in the U.S.-led AI capex cycle, and if the 1990s information-technology (IT) revolution is any guide, future productivity gains could accrue more to users than producers. Here too, emerging markets look relatively disadvantaged vs. developed ones.

As Chua and team note, AI has become one of the most important forces shaping the global economy. Massive investment in AI infrastructure — especially in the U.S. — has helped support global growth even as governments and businesses have faced repeated shocks, from trade disputes to geopolitical tensions. Much of the public discussion has focused on how AI will boost company profits, but a broader question is equally important: How will the benefits of AI be distributed across countries, particularly between developed and emerging economies?

The answer remains uncertain. Estimates of AI’s long-term impact on productivity vary widely: Some economists expect relatively modest gains, while others see AI delivering a much larger boost to economic output. What is clear is that many investors assume the U.S. will capture a disproportionate share of the benefits, with only a handful of emerging economies seeing meaningful gains.

That view helps explain why U.S. tech stocks have significantly outperformed many other markets in recent years, alongside economies such as Taiwan and Korea that occupy critical positions in the semiconductor supply chain.

Three main arguments support the idea that emerging markets may be at a disadvantage.

The first is that most of the value created by AI will be captured by the companies developing the leading AI models and the infrastructure that powers them. Today, those firms are concentrated largely in the U.S., while a small number of companies in Taiwan and Korea dominate essential semiconductor technologies.

Under this view, businesses around the world will pay for access to AI through software subscriptions, cloud-computing services, and related products. As a result, much of the economic value generated by AI adoption could flow back to the firms providing the technology rather than the firms using it.

However, China presents an important challenge to this narrative. Chinese AI models have become increasingly competitive while often remaining substantially cheaper to use. Lower costs appear to reflect a combination of more efficient model designs, lower infrastructure and energy costs, subsidies, and aggressive pricing strategies.

At the same time, concerns about access to leading U.S. models have encouraged countries and companies to explore open-source alternatives, many developed in China. Recent releases from Chinese developers have also narrowed the performance gap with leading American models.

Even so, greater use of Chinese AI models doesn’t automatically translate into equivalent profits for Chinese firms. Many organizations may still rely on American cloud infrastructure and software tools while using Chinese-developed models, leaving the distribution of economic gains uncertain.

The second argument focuses on how quickly AI spreads through an economy.

History suggests the greatest benefits of transformative technologies often go not to the companies that create them but to the much larger number of businesses that successfully adopt them. In the IT revolution of the 1990s, for example, much of the productivity gains came from firms that learned how to use new technologies effectively.

Current measures of AI adoption suggest richer countries are better positioned to put the same lessons into practice. Indicators of AI use and readiness, digital infrastructure, innovation capacity, education levels, and regulatory frameworks are all strongly correlated with income levels.

Many emerging economies face practical barriers. Reliable electricity, data infrastructure, internet connectivity, and digital skills remain unevenly distributed. Language also matters: Because many advanced AI systems have been built around a limited number of major languages, large populations may initially benefit less if their primary languages are underrepresented.

As a result, AI adoption is likely to proceed most quickly in higher-income economies, potentially widening existing productivity gaps.

The third argument concerns labor markets.

AI is generally expected to have its greatest impact on jobs that involve information processing, analysis, communication and other cognitive tasks. These occupations are more common in developed economies than in many emerging markets.

Research measuring exposure to AI across occupations finds that countries with higher incomes tend to have a larger share of jobs that could be affected by AI. This may seem counterintuitive, but it also means those economies have more opportunities to generate productivity gains through AI adoption.

The impact on jobs remains highly uncertain. AI may replace some tasks while making workers more productive in others. It may also create entirely new tasks and occupations. Predicting these effects is difficult as jobs consist of many individual activities, and AI's influence continues to evolve rapidly.

For emerging economies, a key question is whether AI helps workers perform more sophisticated tasks that were previously inaccessible to them. While this could narrow skill gaps between countries, the likely macroeconomic impact may not be large enough to fundamentally alter existing economic hierarchies.

We are still in the early phase of an enormous global buildout of AI infrastructure. Investment continues to flow into semiconductors, software, data centers, power networks and the energy systems required to support increasingly intensive computing demands.

Among emerging economies, Taiwan and Korea appear to be the largest beneficiaries so far because of their dominant positions in semiconductor production. Strong demand for advanced chips has boosted exports, investment, and industrial activity in both economies.

Other participants include Singapore, Malaysia, Vietnam, Thailand, Mexico and China. However, when measured relative to the size of their domestic economies, the growth benefits have generally been more modest than headline export figures might suggest. Singapore stands out as one of the clearest beneficiaries because of its specialized role within the semiconductor ecosystem.

On the software side, Israel is perhaps the most significant beneficiary: The country has built strong positions in cybersecurity and enterprise software, areas that are increasingly important as businesses integrate AI into their operations.

Meanwhile, data-center investment is expanding across many emerging economies as technology companies seek to bring AI services closer to users. Yet the macro impact of such investment may be smaller than many assume. Data centers require substantial construction activity and supporting infrastructure, but much of the equipment is imported. Once operational, they also employ relatively few workers compared with their scale.

The longer-term benefits are therefore likely to come less from construction itself and more from the services that data centers enable, such as cloud computing, data processing and broader AI adoption by local businesses.

Countries such as Malaysia and Thailand appear to be attracting economically meaningful investments. By contrast, many African economies continue to receive relatively little AI-related infrastructure investment, risking a widening digital divide.

Another question: What if the AI bubble pops?

Technological revolutions have often been accompanied by periods of excessive optimism. While AI will likely generate genuine economic benefits, markets may still overestimate how quickly those benefits will arrive, or how large they will ultimately be.

If AI-related investments were to disappoint, a sharp market correction could follow. Such a selloff would affect emerging markets through several channels.

First, global investors would likely become more risk-averse, reducing capital flows to emerging economies and tightening financial conditions.

Second, economies closely tied to the AI supply chain — especially Korea and Taiwan — could face direct economic effects through weaker demand.

Third, a decline in U.S. equity markets could reduce household wealth and consumer spending, creating knock-on effects for countries dependent on U.S. demand.

In such a scenario, investors would probably still seek safety in U.S. government bonds. However, questions about U.S. fiscal sustainability and changing market dynamics could make traditional safe-haven behavior less predictable than in the past.

Some higher-quality emerging-market local-currency bonds could potentially benefit as alternative destinations for capital, particularly because many emerging economies have not experienced the large AI-driven investment inflows seen in U.S. markets.