Over the past several decades, the most reproducible pathway of development for countries has followed a familiar sequence: moving from low-skill agrarian production, to building a globally competitive manufacturing base, and eventually to exporting higher-value services and technology.
Such a progression is often described as a development ladder – a series of rungs that countries climb as they accumulate the capital and capabilities to compete in the global marketplace. This ladder could see increasing risks from this upcoming wave of AI-driven automation.
South Korea is a canonical example of a successful development ladder. In 1953, its GDP per capita was $67 – among the lowest in the world, with little infrastructure and no natural resources. To bring the country out of poverty, Park Chung-hee's government made a bet on competitive manufacturing. Korea would compete globally, starting with the simplest products.
Over the next fifty years, Korea built entirely new industries from scratch: steel mills without any iron, petroleum refineries despite importing all oil. Textile manufacturing in the 1960s led to heavy industry in the 1970s, which spurred the development of an electronics industry in the 1980s. By 2020, South Korea’s GDP per capita reached $33,000: a nearly 500x increase in just seventy years.
Though Korea had other critically necessary benefits, such as strong pro-capitalist economic institutions and strategic American support, its ascent depended on a long window of successful export growth – driven primarily by efficient labor arbitrage.
Labor arbitrage has been the key enabler of export-driven development
There are dozens of determinants for the success of countries leveraging the export-led development ladder. Crucial requirements have historically included strong political and economic institutions, effective education systems, and access to major waterways. Developing countries benefit in comparison to their developed counterparts from reduced land costs, minimal regulatory burdens, and occasionally, generous tax incentives.
The foundation of the export-led strategy, however, is based primarily on effectively applied labor arbitrage. A garment worker in a developing country earning $3 a day, competing against an American earning $150, creates a 50:1 wage differential that makes this comparative advantage inevitable – even after accounting for shipping costs, quality variance, and supply chain complexities. Export-driven development ladders are optimized around the efficient sourcing of abundant and relatively inexpensive human labor.
In many ways, this system is mutually beneficial for both producers and consumers. Consumers in wealthy countries receive access to cheaper goods. Developing countries gain capital investment and access to global markets, but also something far more valuable: systematic capability-building.
Through what development economists call "learning-by-doing" spillovers¹, foreign investment inadvertently transfers technology and know-how that can eventually transform economies. The mechanism is straightforward but powerful. Workers assembling electronics learn precision manufacturing, supervisors learn quality control systems, and so on. These workers eventually leave, starting businesses, joining competitors, or training others – dispersing skills throughout the economy.
Successes and trends in export-driven development
Export-led development has historically been the most proven pathway of economic growth for developing countries over the past half-century. Exemplified by East Asian states such as South Korea, Japan, China, and Taiwan, countries prioritizing export-driven manufacturing strategies have lifted over a billion people out of poverty since the 1950s. Large-scale manufacturing has succeeded by transferring capital, technology, and learning-by-doing to massive, previously unskilled labor forces over decades.
More recently, attention has shifted in many developing nations away from manufacturing and towards exporting digital services. Contemporary developing countries, especially those in Sub-Saharan Africa, have explicitly positioned themselves around fintech, business process outsourcing (BPO), and information technology (IT) rather than attempting to replicate East Asia's manufacturing-led model².
This has seemed to be a rational response for the 21st-century: why invest decades in building labor-intensive manufacturing when you can leverage educated populations, mobile connectivity, and English proficiency? Digital services offer 2-4x higher wages than manufacturing and better working conditions. Many modern countries have plans to potentially leapfrog directly from agriculture to knowledge work, or to pursue export diversification strategies to bypass manufacturing.
Transformative AI could remove the bottom rungs of the development ladder
Transformative AI systems³ could fundamentally break this equation by reducing opportunities for global labor arbitrage across the board. If AI systems have the capability to function as “drop-in remote workers” by the end of the 2020s, developing countries could lose their comparative advantages in two of the primary rungs of the export-driven development ladder – first, in digital services, and eventually in manufacturing.
Digital services could face rapid automation, with value moving from contractors in developing countries to AI systems operated by corporations in wealthy nations. Simultaneously, manufacturing could become increasingly capital-intensive, pricing out countries that cannot afford advanced automation. Together, these dynamics could close the primary pathways available to late developers.
This could happen in three specific ways:
First, AI could eliminate cognitive services as the most generalizable modern development route.
The obvious irony of this latest wave of AI automation is that cognitive services are being automated faster and more completely than manual labor – precisely when many developing countries were betting on digital services as their primary development pathway.
Kenya is an excellent example. It is launching a national business process outsourcing (BPO) policy seeking to bring in a million jobs over five years, and has been pouring resources into call centers, content moderation, and data labelling. Rwanda has explicit intentions to become the “Singapore of Africa”. Nigeria has started to build a tech hub in Lagos home to multiple unicorns.
Unfortunately, this strategy appeared most promising before 2023. Today, those countries are beginning to discover that the digital services pathway could be closing even faster than manufacturing.
Manufacturing automation still may operate on multi-decade timelines due to capital costs and the marginal cost of robotics. On the other hand, many forms of cognitive services could be automated within a few years of developing powerful AI systems because of near-zero marginal costs – deployment costs essentially nothing beyond API calls.
Many call centers are likely to face near-total automation within the next decade. Many tech leaders are predicting the advent of drop-in remote workers by 2027.⁴ If these predictions hold true, digital export strategies for developing countries will be devastated.
Developed countries – and particularly the US and China – are already capturing the vast majority of data center development, due to strengths in capital deployment and larger market sizes. This would leave little to no financial incentives to offshore AI cognitive labor to developing countries.
As a result, countries betting on digital services could have even less time to establish themselves and climb upward than countries pursuing more traditional manufacturing pathways: a total upheaval of decades of economic planning.
Second, AI-driven automation could continue to raise the capital requirements for competitive manufacturing to levels late-arriving countries cannot afford.
Export-led industrialization has succeeded historically because bottom-rung manufacturing required minimal upfront capital. Countries could enter with minimal technology and remain competitive through lower wages. The model was self-financing: profitability at each stage funded progression to the next level of development.
In the past, the technology gap between new entrants and incumbents was narrow enough that wage differentials could compensate. Today, advanced manufacturing techniques increasingly demand greater infrastructural requirements for optimized production. For example, globally competitive garment factories may require automated cutting systems and inventory management. Modern electronics assembly increasingly relies on automated pick-and-place machines and optical inspection systems.
Existing trends point towards the decreasing importance of wage competitiveness, as compared to other factors such as proximity to major markets, human and physical capital, or institutional capacity. Powerful AI systems could exacerbate these trends in the long-term by continuing to reduce the value of low-cost labor and increasing the reliance of manufacturing processes on specialized, expensive infrastructure.⁵ This could negatively impact developing countries such as Bangladesh, which has benefited tremendously from the relocation of clothing manufacturing from China due to labor arbitrage.
In short, rapid AI-driven automation could accelerate the closure of the bottom rung of the manufacturing ladder, by making capital requirements prohibitively high for developing countries that cannot yet compete on investment or technological infrastructure.
Third, AI-driven automation could accelerate the reduction of employment intensity and knowledge transfers of export-driven development.
A substantial factor in the success of this development ladder comes from a pattern often referred to as learning-by-exporting. Research shows that developing strong export industries can have multiplicative effects throughout an economy, in large part due to the development of local human capital.
First, workers gain the skills necessary on the job to compete effectively in a global marketplace. In manufacturing, this might happen via learning effective process optimization or quality control. In digital services, this could arise from developing English proficiency or strong international relationships. As these workers move on, they disperse their skills throughout the economy, creating new economic growth and increasing human capital.
Already, we have been seeing trends over the past two decades that labor employment intensity has been on the decline for manufacturing, driven in part by labor-saving technological progress. Powerful AI would have the clear potential to accelerate these trends both for manufacturing and cognitive services – which would be a significant issue. Export industries employing fewer workers would lead directly to less knowledge diffusion, weakening the learning-by-doing that made export-led industrialization transformative.
There are early warning signs of this trend accelerating, even before the diffusion of powerful AI. For example, Foxconn cut its Kunshan workforce from 110,000 to 50,000 while maintaining output; Adidas recently prototyped an automated footwear facility that employed 160 workers versus 1,200-1,500 for conventional production.
Diminishing leverage for developing countries
Of course, workers in AI-leading countries would not be immune from these dynamics. Early research suggests that domestic labor markets in wealthy nations may already be softening at the entry level, with some evidence pointing toward reduced hiring for junior positions in sectors where AI tools have seen rapid adoption. The erosion of export-driven development pathways could unfold alongside significant automation of services within developed economies themselves. This would create parallel pressures on employment, human capital development, and labor bargaining power across all nations.
Yet AI-leading and developing countries would face these challenges from very different structural positions. Nations that hold a meaningful share of the AI value chain – whether through frontier model development, cloud infrastructure, or semiconductor manufacturing – retain policy options that others lack. They can, in principle, capture the productivity gains created by AI and leverage this wealth to support their affected workers. There are strong incentives for wealthy nations to maintain the stability and strength of their labor market.
Developing nations may not even have this option. Without leverage in the AI value chain, they may see an erosion of their export-led development strategies, without sufficient alternatives to support their labor forces. They may lose their competitive advantages in labor arbitrage due to AI without gaining sufficient leverage in return.
These countries are well-aware of the uphill journey facing them. For example, fifteen African nations have published concrete AI strategies as of 2025, and the 2025 AI Summit in Kigali produced a $60 billion Africa AI Fund – reflecting their ambition to build domestic AI capabilities, rather than remain dependent. Whether these efforts can generate sufficient momentum remains uncertain, but many nations are actively attempting to secure a foothold in the AI value chain before the window closes further.
—
Previous industrial transitions created opportunities for developing countries to move up the ladder even as wealthy countries dominated the top rungs. AI could erode the bottom and middle rungs before developing countries can climb them – and we don’t have proven alternative strategies.
Thanks to Yolanda Laanquist, Danny Buerkli, Anna Yelizarova, and Ankit Mishra for their excellent feedback on this article.
