Publications /
Policy Brief
The term “K-shaped recovery” was introduced to describe the economic recovery from the recession triggered by the Covid-19 pandemic. It refers to the uneven and divergent growth pattern between leading industries or sectors (the upper arm of the letter K—mainly the high-technology sector) and laggards, or the rest of the economy (the lower arm—the traditional industries). Since then the term has become popular in portraying divergent growth patterns within and between countries, especially as the high-tech sector has become even more important in driving GDP growth.
Recent developments suggest that advances in artificial intelligence (AI) have reinforced the K-shaped growth pattern in many different countries. This has been facilitated by the very fast pace of diffusion, especially of generative AI, which reached 100 million users in a few months compared to the years or even decades that it took for other general-purpose technologies to do the same. For example, it took six years for the internet and 16 years for the cellphone to achieve similar levels of usage.
These developments are likely to exacerbate the growing inequalities within countries and the gaps—especially digital and AI gaps—between core foundation leaders (the US and China), several high-adoption leaders among other developed countries, and the rest of the world. Growing inequality also heightens social and political tension, adding another layer of discord in the ongoing polarization and geopolitical contention. From a strictly economic perspective, as the high-tech sector has created immense wealth for capital owners but not enough jobs or decent incomes for many working people, this has led to low real wage growth, which could weaken aggregate consumer demand, slowing economic growth in many countries.
K-SHAPED DEVELOPMENTS WITHIN COUNTRIES
Generally speaking, a K-shaped pattern of economic growth can be observed in various activities and sectors, including wages and corporate earnings, wealth accumulation, equity market performance, consumer spending, business investment, growth of value-added production and productivity improvements, etc. in practically every country. Importantly, K-shaped growth has made aggregate data such as GDP unreliable, and even misleading, indicators of the welfare of the majority of people in an economy, as such data hide the inequality in income and wealth distribution that has left many people behind despite overall growth.
In the US, it has been clear that activities, especially investment expenditure in the IT/AI sector, while accounting for 4% to 8% of GDP, has contributed between 35% and 50% to annual GDP growth. For example, GDP grew 2.1% in 2025; but excluding the IT/AI sector, the rest of the American economy (i.e., the other 95%) grew by at most 1.6%, as lackluster as the EU growth rate which has trailed that of the US.
Moreover, the importance of AI activity has continued to increase—through investment spending and productivity improvement. Specifically, five AI “hyperscalers”—Microsoft, Alphabet (Google), Meta Platforms, Amazon and Oracle—will likely make investment expenditure of US$800 billion this year, and US$1.1 trillion in 2027, according to Morgan Stanley. In the first quarter of 2026, the AI economy grew by 31% and the non-AI economy by 0.1%—while aggregate GDP grew by 2.1%. The Magnificent Seven high-tech corporations in particular have posted earnings growth of 61%, driving the SP500 overall corporate earnings improvement of 27%. By contrast, labor compensation adjusted for inflation fell by 0.5%—its share of the business sector output fell to a low of 54.1% since record-keeping began in 1947. As a result, many American households have been struggling to make ends meet—a recent Harris poll reports that about half of the respondents across party lines are “having trouble affording everyday necessities like gas and groceries”.
Besides capital expenditure, the IT/AI sector has shown a remarkable improvement in productivity, outpacing the rest of the economy. From 1988 to 2023, the cumulative total factor productivity (TFP) in the IT sector grew by 178%, dwarfing the 12% gain posted by the non-IT sector (or rest of the economy). Specifically, productivity improvement has been concentrated in a few areas such as marketing output (50% year over year), software development (26%) and customer service (14–15%). On the other side of the coin, those areas have experienced firing or reduction in hiring in recent times.
Importantly, the K shape can be seen in wealth accumulation—people with financial and real-estate assets have grown much richer as asset prices have risen, leaving a majority of people without significant assets well behind. Specifically in the US, the net wealth of the top percentile has grown by more than 25%, while that of the middle 40% has grown by less than 10% since 2023. Furthermore, income differentiation between people in high-tech and finance relative to those in other sectors has widened further. As a result, people in higher wealth and income tranches have accounted for a larger share of consumption.
Concretely, the Gini coefficient for wealth distribution has risen to a 60-year high of 0.80–0.85, with 1% of the population holding a record 32% of total net worth in the economy while the bottom 50% held only 2.5%.
While less pronounced than in the US, the EU has also experienced a K-shaped economic divergence. High-income households, capital owners, high-tech and export sectors have prospered; lower-income households, services and energy-intensive sectors have lagged behind. Specifically, the information and communication technology (ICT) sector grew by 6.3% p.a. during 2013–2023, about three times faster than the total economy of the OECD member countries, 24/38 of which are in Europe. It is important to note that according to the IMF, generative AI has helped improve TFP in Europe by 0.8% over the past five years, faster than the US with 0.7%—but Europe lags behind the US in overall productivity growth due mainly to slower diffusion of technology to the whole economy. Meanwhile, the growing K-shaped pattern in consumption has presented a complicated challenge to the European Central Bank’s (ECB) efforts to control price and inflation expectations. In particular, the Gini coefficient for wealth distribution is high, averaging between 0.60 and 0.75 in EU member states.
Japan has also seen a K-shaped economic recovery, with a sharp divergence between high-income households with assets benefiting from market rallies and those relying on wages suffering from the impact of inflation. Furthermore, there has been a gap between export sectors and industries, including services, more geared to domestic consumption. In particular, the AI sector is estimated to grow by 34.4% from 2025 to 2032, significantly faster than the expected growth rate of the overall economy. The Gini coefficient for wealth distribution has been relatively high at 0.55–0.63.
In China, the K-shaped growth pattern shows signs of becoming entrenched. Hi-tech industries, especially electric vehicles and batteries, have boomed, driven by strong export growth—leading to a record trade surplus of US$1.2 trillion in 2025. According to the Institute of International Finance, the ICT sector, which accounts for 10.5% of the economy, has grown by twice the pace of GDP growth in recent years. In particular, AI sector revenue grew at a compound annual growth rate (CAGR) of 54.1% over the period 2023–2025, and is expected to grow. By contrast, the domestic economy is feeble, bogged down by weak consumption amid a persistent real-estate crisis. Importantly, China’s wealth Gini coefficient rose from 0.45 in 1995 to more than 0.7 in 2023.
India’s growth has also exhibited a clear K-shaped trajectory. Specifically, the AI sector is expected to grow by a CAGR of 11.5% in the foreseeable future. Meanwhile, the wealth and consumption of the top 1% of the population has surged, while that of the bottom 50–95% has stagnated. The urban-rural divide in spending has been notable. In 2023, urban spending rose by 8.1% and rural spending by 4.1%. The country’s wealth Gini coefficient is rather high at 0.74.
K-shaped developments have also been observed in fast-growing Southeast Asian countries. For example, Vietnam has experienced a K-shaped development, with manufacturing exports especially in electronics and by FDI companies booming, while domestic sectors mostly populated by small and medium enterprises have been suffering from slow consumption demand and high interest rates. The ICT sector has grown impressively by 10.1% CAGR from 2021 to 2025—accounting for 8% of the country’s GDP—compared to the GDP annual growth rate of 6.3%. Meanwhile, in Q1 2026, driving overall export growth of 19.1%, FDI sectors jumped by 33.3% (taking an 80.1% market share) while domestic players contracted by 16.6%—hurt by tariff shocks and a global slowdown. The country’s wealth Gini coefficient is at 0.60.
K-shaped economic developments in Latin America
Latin America’s GDP grew by 2.4% in 2025, and is expected to slow to 2.1% this year—after a lackluster 0.9% p.a. in the previous 10 years. By comparison, the IT sector has grown much more robustly, by more than 8% p.a., with the AI market increasing by 28.1% but the rest of the economy trailing well behind. As a result of this K-shaped economic growth, the wealth Gini coefficient for Latin America is among the highest in the world—at 0.7 to 0.9—with the top 1% of the population holding 35%–40% of total private wealth.
K-shaped economic developments in Africa
Driven by high-tech startups, the digital/tech sector in Africa is expected to grow by 15.5% p.a. from US$35 billion in 2026 to US$72 billion in 2031. This pace exceeds the estimated average GDP growth rate of around 4.4% p.a. for Africa over the same period. Strong growth has been concentrated in certain regions of the continent. For example, Lagos (Nigeria) has prospered, focusing on fintech and startup funding, especially for West African countries. Nairobi (Kenya), the so-called “Silicon Savanah”, has developed agri-tech and mobile banking and payment (M-Pesa) sectors that support the economy. Cape Town and Johannesburg (South Africa) have benefited from mature tech talent, internet penetration and cloud activity. In addition, Cairo (Egypt) has focused on B2B logistics and Kigali (Rwanda) on smart-city and biotech initiatives.
The uneven developments highlighted above accentuate income and wealth inequality in Africa, which is becoming one of the obstacles to future development. Africa’s wealth Gini coefficient is moderately high, ranging between 0.54 and 0.65.
DIVERGENCE BETWEEN ADVANCED ECONOMIES AND LOW-INCOME COUNTRIES
According to the St. Louis Fed, between 1870 and 1999 “the difference in income between the richest countries and all others has increased by an order of magnitude”. The gap narrowed in the subsequent decade when average per-capita GDP in the rest of the world relative to the US increased from 27% to 35%. However the gap has widened again since 2010, and is becoming quite pronounced in the case of low-income countries (LICs). In other words, the K-shaped growth pattern is also visible between high- and low-income countries.
The widening development gap between advanced economies and LICs has also been exacerbated by a collapse of official development aid (ODA), which fell by 23% from a high of US$224 billion in 2023 to US$174.3 billion in 2025, and is likely to fall by 6.9%–8% this year, according to the OECD. The ratio of ODA to gross national income (GNI) among developed countries has declined to 0.26%, compared to the UN target of 0.7%. The sharp fall in international development assistance has significantly hindered many LICs in their efforts to meet the basic economic and social needs of their populations as well as taking measures to build infrastructure to foster development. This may push them even further behind relative to the rest of the world.
BROADER IMPLICATIONS OF AI
Broadly speaking, the advance of AI risks repeating the flaws of globalization in the past: driving global growth and prosperity on the aggregate level, but being blemished by an unequal distribution of the gains that has left many people behind. This could strengthen the popular backlash that has grown in the past decade against existing government institutions and policies. In other words, AI has exacerbated the problems caused by K-shaped growth patterns.
Beyond increasing inequality, AI has exacerbated the negative impacts of K-shaped growth. As wealth has become concentrated among a small number of tech billionaires, they wield significant influence in the political process. AI has added to the misinformation and deep fakes circulating through social media, deepening social division and mistrust in democratic institutions including elections. This has polarized society, creating fertile ground for populist movements and politicians, and making it more difficult to build social and political consensus to support sensible and comprehensive policy measures—including mobilizing fiscal resources—to address the challenges.
At the same time, AI has enabled authoritarian governments to strengthen their surveillance and control of their citizens, helped by the availability of China’s sophisticated surveillance technology. Furthermore, advanced AI models such as those released recently by Anthropic can pose a serious risk to cybersecurity, especially in terms of national security, personal privacy and financial stability (as financial institutions can be hacked, causing property theft and disruption of activity).
A TECHNOLOGICAL SYSTEM DOMINATED BY THE G2—THE WIDENING GLOBAL DIGITAL GAP
AI hardware, software, models and applications have increasingly been concentrated in a handful of conglomerates domiciled in the US and China. As the two superpowers try to find ways to compete in an environment of “constructive strategic stability”—as revealed after the May 2026 Trump–Xi summit—it seems likely that the rest of the world will participate in the AI revolution under conditions largely set by the G2, which between them command 90% of global computing power, employ 70% of leading researchers and deploy the vast majority of AI investment.
In other words, AI is likely to cement the leadership position of the US and China in technological advances, especially in terms of the vast scale of investment needed to develop data centers and to test models and applications. Basically, the rest of the world has to rely on a handful of companies based in the US (the AI hyperscalers mentioned above) and China (DeepSeek, Tencent, Alibaba and Huawei etc.) for hardware, software—including cloud computing and data storage. Consequently, many countries are vulnerable to being shut out of these essential services by either government as a coercive measure if such a move serves its national interests. One recent example is the US government directing Anthropic to deny foreign nationals access to its advanced AI models on national security grounds. Worse still is the case of Ukraine being denied access to the Starlink internet system—essential to its defense against Russian attacks—briefly in 2022 and 2023, according to the whim of the system’s owner, Elon Musk.
Similarly, China is in a position to leverage the widespread adoption of Huawei telecommunications infrastructure and devices to conduct espionage and impose economic pressure on recipient countries, locking them into the Chinese sphere of influence. Moreover, China has accelerated the development of its Beidou global navigation satellite system to compete with the US GPS (Beidou has provided better coverage in developing countries than GPS) and its GuoWang and Qianfan low earth orbit satellite internet systems to compete against Starlink.
China has demonstrated its willingness to use control of rare earth elements and other critical minerals necessary for goods manufacturing, especially of hi-tech products, as coercive tools to promote its national interests.
It is interesting to note that the EU has also tried to develop its own hi-tech sector, seeking to reduce its reliance on the US and China. For example, it has successfully developed the Galileo global navigation satellite system, requiring all smartphones sold in the EU to be compatible with Galileo to minimize reliance on GPS.
In short, the reliance of many countries on technological platforms, systems and technological as well as regulatory frameworks dominated by the US and China makes them vulnerable to either or both countries—similar to the way they are vulnerable to economic sanctions due to their dependence on the dollar financial system.
NECESSARY RESPONSES: CLOSING THE ARMS OF THE K
In light of the serious implications of K-shaped developments—now accentuated by AI—it is important that authorities focus their efforts on closing the two arms of the K.
Domestically, authorities should focus on building (DPI) to provide the foundation and catalyst for technologically enabled growth. DPI includes hardware and systems to provide connectivity and broadband access to the internet and telecommunications together with appropriate governance and regulatory frameworks. These would serve to facilitate the use of digital IDs and interoperable, mobile and real-time payment facilities which would improve financial inclusion—a key measure in fostering equitable growth. DPI would also facilitate the efficient accumulation and usage of data for economic and social development.
Examples of positive impacts of this approach include India, Estonia, Brazil and Singapore. The “India Stack” has been viewed as the DPI benchmark for other countries. It comprises a full sequence of digital ID (via the Aadhaar system with enrollment of more than 1.3 billion people), the Unified Payment Interface (UPI, an instant payment system leading the world in transaction volume) and the DigiLocker/Account Aggregator for data accumulation and sharing. Estonia has offered X-Road as the interoperability backbone for its eGovernment, connecting and delivering nearly all government and many private services. Brazil has benefited from its Pix (instant payment system—second only to India’s UPI). Singapore has implemented SingPass as a digital ID layer, PayNow for fast payment and strong data governance. Generally speaking, it has been estimated that DPI could improve growth by 20–33% in the Global South by 2030.
Importantly, progress in building digital public infrastructure would allow authorities to leverage technological progress, especially in digital and AI applications, to help train working people, enabling them to participate in the technologically driven economy.
Specifically, advanced services facilitated by digital and AI tools—including in areas such as education, health care especially for older people, information and entertainment, etc.—should be promoted by supportive regulatory environments. In particular, attention should be paid to the importance of the services sector—comprising informal, low-productivity and low-paid jobs—which in developing countries generally accounts for 50% of value added in the economy, compared to 25% from agriculture and 35% from manufacturing, mining and construction.
This approach can be very relevant for many developing countries that have tried to replicate China’s successful strategy of manufacturing for export development but currently face the competitive constraints erected by China’s predatory mercantilist trade practices—especially its overcapacity leading to a surge of exports of competitively priced products. Against this backdrop, digital services and platforms can broaden the range of service activities to absorb more workers, enhance their ability to improve their work and income, and bring them into the formal economy. This is especially important for Africa, which needs to create 15 million jobs each year to absorb the number of young people entering the labor force. Failure to do this will leave many young people unemployed or under-employed, posing risks to social stability in those countries, and adding to migration problems in advanced countries especially in Europe.
Equally importantly, in addition to training efforts, countries should be prepared to deal with a potential jump in unemployment triggered by widespread AI usage—a concern recently raised by nearly 200 economists and tech leaders. The policy should aim to provide unemployment support (including using wage insurance schemes) to help displaced workers during their transition periods, instead of trying to protect jobs being replaced by AI advances—following the Nordic “flexicurity” model. Moreover, authorities could consider diversifying their tax base away from labor income if AI is compressing the labor share of national income, to capital income taxation—to address the growing income inequality. The above are practical measures that should be considered by authorities to deal with the problems posed by AI and the K-shaped development pattern, while more thoughts should be given to more radical approaches such as universal basic income or universal basic capital.
CLOSING THE GLOBAL DIGITAL GAP
Internationally, countries besides the US and China, from Europe to those in the Global South, need to find ways to assert some degree of digital autonomy in the AI ecosystem. Concretely, they need to decide, based on their endowments and circumstances, whether to pursue full-stack AI sovereignty—developing their own chips, computing capabilities and AI models and applications—which is a formidable challenge, given the G2 dominance in this field. Except for the EU in selected strategic chokepoints, for many countries it is more realistic to follow the so-called “Singapore” approach: asserting national control over standards, data governance and strategic direction but importing AI hardware and software and cooperating with a range of global AI conglomerates to avoid reliance on a single provider. Specifically, countries can regulate the development of domestic data centers and access to them by leading AI providers to train their models so that the outputs can reflect their own culture and values. Otherwise, using AI models and outputs based on data unique to the US and China risks being assimilated into those countries’ frameworks.
Importantly, countries need to navigate three competing regulatory frameworks for AI. The EU has passed the world’s first comprehensive law—the EU AI Act—focusing on human rights, privacy and transparency as requirements on AI developers everywhere wanting to do business in EU markets. These requirements have been viewed as safeguarding human values but slowing the pace of innovation, compared to the innovation-first approach of the US, which does not have a federal AI law but relies on voluntary standards (such as in the National Institute of Standards and Technology (NIST) AI Risk Management Framework) and a few state transparency regulations (such as in California, New York and Colorado). Meanwhile, China prioritizes state control, security and social stability, requiring AI developments to be regulated by the Cyberspace Administration of China.
Instead of the hoped-for convergence, there have been signs of fragmentation, and in particular, enforcement conflicts, that is, where an AI company or product can comply with domestic regulations but be deemed to violate one of the foreign regulatory regimes, thus potentially being subject to sanctions. Many countries try to deal with the US–EU–China trilemma by adopting soft-law standards—such as OECD AI Principles, ISO/IEC 42001 software standards and certification—claiming those as international best practices satisfying all three regulatory regimes without aligning with any specific one. It remains to be seen if this approach can provide legal clarity or certainty for AI companies wanting to do business in all three regions.
In short, to promote economic development and national security while dealing with G2 geopolitical and technological dominance and contention, other countries need to implement in earnest an appropriate AI/digital strategy. Such a strategy needs to focus, among other things, on building digital public infrastructure, training workers in digital and AI skills, and maintaining appropriate control over standards and governance of data. In fact, AI strategy has become indispensable for addressing both the negative consequences of K-shaped economic developments at home and the international digital/AI gap—avoiding becoming a permanent underclass in the AI economy.
