How the same AI and robots will split twelve economies, and what decides which way each one goes

Article 3 of 4  ·  Series I  ·  Published 16 October 2026  ·  Analysis → Forecast → Recommendations

How this series measures things. Every article applies the same three questions to its subject. Core size: how many residents at the same time work, earn above subsistence, pay more in tax than they receive, and spend broadly at home. Core productivity: how much value each of them produces, and how much capital, automation and AI stands behind each one. Core capture: how much of that value stays inside the country as wages, taxes and domestic demand. Article 2 set out a production model for an ageing economy. This article applies the three questions to twelve economies that start from very different places and asks what the same technology will do to each.

1. The Signal

Four records were set within a few months of each other in 2026, in four different economies, by four different measures.

In the United States, the share of business output paid to workers fell to 52.8% in the second quarter, the lowest level since the series began in 1947 (measured, Bureau of Labor Statistics). In 1947 it was 65.5%. The decline came in a year when AI-related investment was lifting growth: the IMF estimates that AI technology investment added about half a percentage point to US GDP growth in 2025 (reported, IMF Annual Report 2026).

In China, unemployment among people aged 16 to 24 who are not in education reached 18.9% in August, matching the highest level since the current method was introduced in 2024 (measured, National Bureau of Statistics, via Caixin). That record came in the same year China ran a goods trade surplus of almost $1.2 trillion.

In Russia, unemployment stood at about 2%, the lowest ever recorded. The governor of the central bank described the shortage of workers as "a new reality for the government and for business alike", driving up wages, costs and inflation (reported, Moscow Times, April 2026).

In Japan, the number of foreign workers reached a record 2.57 million, up 11.7% in a year, as firms in care, construction and manufacturing ran out of domestic workers to hire (measured, Ministry of Health, Labour and Welfare).

Four records, four different problems. The United States produces more than ever and pays a smaller share of it to workers. China has factories and young people, but not enough good jobs for the young people. Russia and Japan have jobs but not enough people.

All four have access to broadly the same AI models, the same industrial robots and similar cloud infrastructure. A common assumption is that this shared technology will pull economies toward a common future. The evidence of 2026 points the other way. The same tools are amplifying the differences that already exist.

2. The Mechanism: technology amplifies starting conditions

A machine does not decide what kind of economy it is installed in. The economy decides what the machine does.

Put the same AI system into three settings. In a firm that cannot find staff, it fills the gap and makes the remaining staff more valuable. In a firm with plenty of applicants, it lets the employer hire fewer of them. In a firm whose customers are disappearing, it lowers costs but does not create demand. The technology is identical. The outcome depends on three starting conditions: how many working-age people there are relative to jobs, how much capital and skill stands behind each of them, and whether the institutions around the firm turn higher output into higher pay.

Articles 1 and 2 argued that what carries an advanced economy is the Economic Core: the residents who produce, earn, pay tax and spend at the same time. A rough formula captures the idea:

Economic Core capacity = productive people × skills × capital per worker × AI leverage × share of income reaching households × net fiscal contribution

This is an analytical construct, not an official statistic. Its use is in showing that AI enters the formula as one multiplier among several. It raises output per worker. It does not by itself raise the number of productive people, the share of income that reaches households, or the skills that let people use it.

Four worlds

Applying the formula to twelve economies produces four groups.

Twelve economies, one technology, four different jobs for it to do.

Why the gap widens rather than narrows

Two features of AI push toward divergence.

First, its returns depend on complementary assets: skilled people, reliable electricity, data infrastructure, firms large enough to reorganise around it, and capital markets willing to fund the change. The IMF's AI Preparedness Index puts Singapore at the top and finds that East Asia is the hub for chip design and manufacturing (reported, IMF Annual Report 2026). Economies that already have these assets capture more of the gain.

Second, AI's effect on jobs depends on exposure. The IMF estimated that about 60% of jobs in advanced economies are exposed to AI, against 40% in emerging markets and 26% in low-income countries. In advanced economies, roughly half of the exposed jobs could benefit from higher productivity and half could see lower demand for labour (measured, IMF Staff Discussion Note 2024). Lower exposure in poorer countries sounds protective. It is also a sign that they will capture less of the productivity gain.

The result is that the same wave of technology tends to make strong positions stronger and weak positions more exposed. The following sections look at why, in each world, and at the two forces that decide the outcome: whether people are scarce or abundant, and whether gains reach households.

What the technology actually changes inside a firm

AI and robots act through three channels, and the balance between them differs by world.

The first is task automation: a machine or model performs work that a person used to do. This lowers the labour needed for a given output. It is the channel that dominates public debate.

The second is task creation: new work appears that did not exist before, such as supervising automated lines, training and checking models, maintaining sensors, or serving customers the firm could not previously reach. Acemoglu and Restrepo's work shows that the long-run effect of technology on employment depends mostly on whether this second channel keeps pace with the first.

The third is capability transfer: the tool spreads the know-how of the best workers to everyone else. In the customer-support study by Brynjolfsson, Li and Raymond, an AI assistant raised productivity by about 14% on average and by about a third for the least experienced agents (measured, Quarterly Journal of Economics 2025). This channel matters most where workers are young and inexperienced, which is to say in World III.

A firm in a labour-scarce economy has strong reasons to use the second and third channels, because it needs every worker it has. A firm with a queue of applicants has strong reasons to use only the first. That is the micro-level reason the same tools produce different national outcomes.

How to tell which world a country is in

Three numbers place an economy more reliably than its GDP growth rate.

—   The annual change in the working-age population. Falling means scarcity; rising faster than jobs means surplus.

—   Unemployment among young graduates. High graduate unemployment alongside low overall unemployment, as in China and India, signals a mismatch that AI can either ease or deepen.

—   The share of income reaching households: the labour share of output and the household consumption share of GDP. Falling shares mean productivity gains are not turning into demand.

A country can sit in different worlds on different measures. China is World I on its working-age trend, World III on its graduate labour market and a special case on distribution. That is why it is the most uncertain of the twelve.

3. Subtheme One: scarcity or surplus decides whether AI multiplies or replaces

The first force is simple to state. Where workers are scarce, automation tends to complement them. Where workers are abundant, it tends to replace them. Article 2 showed the evidence for industrial robots; the same logic applies across the four worlds.

World I: few people, abundant capital

Japan is the most advanced case. People aged 65 and over make up 29.6% of the population, and workers aged 65 and over are 13.8% of everyone employed (measured, 2026 and 2025). A restaurant that needs ten staff and can find eight uses machines to let eight do the work of ten. Wages for those eight can rise. The likely shape of the 2040s is fewer people, much more capital per worker, flat or falling total output, and rising output per person. The risk is not collapse but slow growth in domestic demand.

South Korea is the same story in fast motion. Its population aged 15 to 64 is projected to fall from 36.6 million in 2023 to 27.2 million in 2044 (projected, Statistics Korea). It already has the highest robot density in the world, 1,012 robots per 10,000 manufacturing workers, more than six times the world average (measured, IFR 2024). There is one hopeful sign: births rose for a second year in 2025, by 6.7% to about 254,000, and the fertility rate rose to 0.80 from a low of 0.72 in 2023 (reported, Statistics Korea, via Korea Times). That will not change the workforce of the 2030s. Korea's challenge is that robots do not buy phones or cars. With a domestic market this thin, it depends on foreign demand more than ever.

Germany has the institutions that make automation complement workers, as Article 2 showed, but it is losing industrial jobs at about 10,000 a month and faces energy costs higher than its competitors. The wider EU has a large wealthy market and slow decision-making. Its new industrial policy aims to keep strategic production at home; whether that becomes a productivity strategy or a subsidy strategy is the open question.

In World I, AI's job is to multiply people who cannot be replaced. That is the easiest role for it to play well.

World III: many young people

The opposite problem looks like this.

India needs productive work for a very large number of young people. National unemployment was 3.1% in 2025, but 9.9% among people aged 15 to 29 and 11.2% among graduates (measured, Periodic Labour Force Survey 2025). The graduate figure matters most. India's growth model has relied heavily on IT and business services, the sector most exposed to generative AI. If one AI system replaces a hundred junior analysts, the ladder that has carried millions of Indian graduates into the middle class loses its bottom rungs. If instead a hundred workers become two or three times more productive with AI and serve far more clients abroad, India could become the largest new source of mass consumer demand in the world. McKinsey projects India's share of global consumption rising from about 9% to about 16% by 2050 (projected, MGI 2025). That projection assumes the second path.

Egypt shows the scale of the gap. About 1.3 million young Egyptians enter the labour market each year while roughly 500,000 jobs are created (reported, World Bank, November 2025). An automation strategy that simply replaced workers would deepen the gap. What Egypt needs is labour-absorbing growth in manufacturing, construction, logistics, tourism and business services, with AI raising the output of people who are already employed.

Iraq is the hardest case. In 2025 oil accounted for about 53% of real GDP, 88% of government revenue and 91% of goods exports. People aged 15 to 29 are nearly 29% of the population; labour force participation is 38% and unemployment 13.5% (reported, World Bank). AI by itself solves almost nothing here. Electricity, water, institutions, education and private firms come first. Without them the combination is unstable: a large young population, an oil sector that needs few workers, and new technology that needs fewer still.

In World III, the role AI must play, complementing labour, is the one it is least likely to play by default, because firms facing abundant labour have weaker reasons to choose it.

Why the East Asian route is narrowing

For half a century, poor countries with young populations followed a recognisable route to prosperity: move farm workers into labour-intensive factories, export to rich countries, raise skills, and climb the value chain. Japan, Korea, Taiwan and then China all used it.

Dani Rodrik has shown that this route is narrowing. Manufacturing employment in developing countries now tends to peak at lower income levels and lower shares of total employment than it did for earlier industrialisers, a pattern he calls premature deindustrialisation (Journal of Economic Growth 2016). Two forces drive it. Factories need fewer workers for the same output because of automation. And global manufacturing capacity is increasingly concentrated in China, which competes on scale, supply networks and automation at the same time.

For World III, that leaves three possible paths, none as reliable as the old one: services exported digitally, as India has done; labour-intensive manufacturing in segments China is leaving, as Vietnam and Bangladesh have done; and domestic construction, agriculture and services raised in productivity by cheap technology. AI helps the first path and threatens part of it at the same time. It barely touches the second. And it could transform the third, if it is deployed to equip workers rather than to replace them.

World II: ageing at middle income

The middle group carries features of both.

China has factories, robots, batteries and electric vehicles, and a falling working-age population: the 16-to-59 group fell by about 6.6 million in 2025 alone (measured, NBS). Its robot density, 470 per 10,000, doubled in four years (measured, IFR). Yet it has record youth unemployment. That combination, workers scarce in aggregate but young graduates unable to find suitable jobs, reflects a mismatch: the economy is generating automated manufacturing capacity faster than it is generating the service and professional jobs young graduates want.

Vietnam is racing the clock. Its period of a favourable age structure is projected to end in 2036, and it is expected to become an "aged" society around 2034 (projected, National Statistics Office, via Saigon Times). Income per person passed about $5,000 in 2025 (reported, General Statistics Office). The window to move from assembly to technicians, engineers and Vietnamese-owned companies is roughly the next decade.

Brazil sits between. Its population is projected to peak in 2041 at about 220 million and decline to about 199 million by 2070, when about 38% of residents will be 60 or older (projected, IBGE 2024). Brazil has resources, a large domestic market and some advanced industries. What it lacks is productivity growth. The next fifteen years decide whether it grows old rich or grows old at middle income.

In World II, AI has to do two jobs at once: replace missing workers and raise productivity fast enough to lift incomes before the population ages further. That is the hardest combination.

4. Subtheme Two: whether gains reach households decides whether productivity becomes demand

The second force is distribution. A country can produce more with AI and still weaken its Economic Core, if the gains do not reach the people who would spend them.

The United States: the fork

The United States ages more slowly than Europe or East Asia, has historically relied on immigration for population growth, and holds a large share of the world's AI models, chip design, cloud infrastructure, venture capital and research universities. One engineer with AI agents, cloud computing and automated tools can now produce what once took dozens of people. That makes the skilled American worker potentially very valuable.

The labour share points the other way. At 52.8%, the share of business output paid as compensation is at a record low (measured, BLS). It fell from 53.6% in the first quarter of 2026 to 52.8% in the second, during a period of strong AI investment. One quarter is not a trend, and the labour share fluctuates with the business cycle; it rose briefly to 58.9% during the 2020 recession. But the direction over decades is clear.

If productivity gains increasingly flow to capital owners while median wages lag, GDP can look strong while the consumer base weakens. The result would be a narrower group of very wealthy owners above a larger group of constrained consumers, and a demand base increasingly dependent on the spending of the wealthiest households. If the gains instead reach wages, through competition for scarce skills, through new tasks that AI creates, or through policy, the United States could build the strongest version of the Economic Core in the world. Same technology, two different societies, and the choice between them is not technical.

China: production without enough buyers

China faces the distribution problem at the level of the whole economy. Household consumption is roughly 40% of GDP, well below the 55–60% typical of advanced economies (reported, World Bank). Firms and the state capture a larger share of income, much of it reinvested in production capacity. The result is an economy that can make far more than its households buy.

AI and robots will make China even more productive, and that is exactly the problem. If industrial output keeps rising while the number of young households shrinks, the extra output has to be sold abroad, which collides with trade defences in Europe, the United States and increasingly emerging markets. The IMF's own recommendations for China centre on distribution: stronger health care, pensions and unemployment benefits to reduce precautionary saving; giving urban residence rights to rural migrants; more progressive taxes on labour and stronger taxes on capital. It estimates these could raise consumption by about four percentage points of GDP over five years (reported, IMF, February 2026). The question for the 2030s is not whether China can produce. It is whether it can create enough well-off Chinese consumers quickly enough.

Europe and Japan: distribution through institutions

World I economies have one advantage over both the United States and China on distribution: institutions that tie wages to productivity. In Germany, sector-wide collective agreements and works councils shaped how firms adopted robots; Article 2 described research showing that incumbent workers were mostly retrained rather than dismissed. In Japan, annual spring wage negotiations have delivered the largest pay increases in more than three decades as labour shortages deepened (reported, Rengo, 2024 and 2025).

These institutions have costs. They can slow adjustment, protect insiders at the expense of newcomers, and raise labour costs in sectors exposed to Chinese competition. But in an economy where workers are scarce and technology raises output per worker, they are a mechanism for turning productivity into household income. Scarcity gives workers bargaining power; institutions decide whether that power becomes broad-based pay or a premium for a few.

Russia: scarcity without full access

Russia's problem is the reverse of Iraq's: too few workers rather than too many. With unemployment around 2%, firms compete for staff by raising wages faster than productivity, which feeds inflation and keeps interest rates high. In principle, Russia needs automation more than most economies. In practice, successful automation requires engineers, chips, machine tools, industrial software and capital, and access to several of these from international suppliers has narrowed. Where domestic or alternative suppliers can fill the gap, automation proceeds; where they cannot, firms face rising wage costs without the tools to offset them. The likely path is not Japan's, fewer workers and heavy automation with rising productivity, but a harder one with persistent cost pressure.

What distribution means for the Core

Across all four worlds, distribution decides whether productivity becomes demand. A rise in output per worker increases Core productivity. Only if a substantial share reaches wages and household income does it also sustain Core size and Core capture. The economies that will do best are not necessarily those that adopt AI fastest, but those that adopt it in a way that keeps the link between output and household income intact.

5. What Most Analysis Gets Wrong

Treating AI adoption as the winning variable

The most common framing is a race: which country adopts AI fastest wins. Adoption matters, but it is one multiplier. An economy that adopts quickly while its gains flow mainly to capital, its young people lack skills to use the tools, or its domestic demand is too weak to absorb more output can adopt fast and still see its Economic Core weaken. The more useful question is whether a country can convert AI, capital and its available people into a large layer of productive, well-paid residents.

Assuming a demographic dividend is automatic

In the twentieth century, a large young workforce was an unambiguous advantage. With automation, it becomes conditional. Twenty million well-educated people with large amounts of machinery and AI behind them can produce more than a hundred million with little training and few tools. A young population is a dividend only if the economy creates productive work for it. Egypt's gap of about 800,000 missing jobs a year (derived) is what an unpaid dividend looks like.

Assuming ageing means decline

Japan has shrunk for over fifteen years and remains rich and stable. Korea's births are rising again, modestly. The most aged economies are also the most automated. Ageing reduces total output growth; it does not have to reduce output per person.

Mistaking an investment boom for a productivity boom

AI is currently raising growth mainly through investment: data centres, chips, power infrastructure. That shows up in GDP at once. Productivity gains from using AI across the wider economy arrive later and more unevenly, because firms have to reorganise work around the tools. The gap between the two is where disappointment can occur. An investment boom that is not followed by broad productivity gains leaves behind capacity, debt and high valuations, but not higher wages.

Looking at countries one at a time

The four worlds interact. China's productivity and weak consumption become Germany's industrial competition. World I's labour shortages become World III's migration opportunity. The United States' AI investment cycle becomes everyone's financial risk: the IMF has warned that debt-financed AI investments could pay off less than expected, leading to equity price reversals, wealth losses and layoffs (reported, IMF Annual Report 2026).

Where this view sits among other researchers

6. Base / Stress / Extreme

The scenarios describe the twelve economies over the next decade. They sum to 100.

Base: Divergence by adaptation (55%). World I economies that pair automation with rising wages hold or raise output per person while total output stagnates. China maintains productivity and a large surplus, with slow progress on household consumption. India captures part of the service-sector opportunity, but graduate unemployment stays elevated. Vietnam keeps rising but slows as it ages. Egypt and Iraq remain stuck between young populations and too few productive jobs. The US labour share stays near record lows.

Falsifier: the US labour share recovers above 56% for four consecutive quarters by 2030 without a recession.

Stress: Gains to capital, demand stalls (30%). AI gains concentrate with owners in the United States and China; labour shares fall further in several World I economies. Consumer demand in rich countries weakens even as output rises, and export competition intensifies as every producer looks for buyers abroad. Youth unemployment rises in India and China as AI replaces entry-level professional and service work faster than new tasks appear.

Falsifier: India's youth unemployment rate falls below 7% in the PLFS for any calendar year before 2031.

Extreme: AI investment bust (15%). Debt-financed AI investment fails to earn expected returns, triggering a sharp equity correction concentrated in the United States, a wealth shock and layoffs in technology and related sectors, and a global slowdown. AI adoption continues but more slowly; financing for World II and III projects dries up.

Falsifier: US AI-related investment continues to grow in real terms every year through 2029.

7. Forecast, 1 year (to October 2027)

F1. The Bureau of Labor Statistics reports the US nonfarm business labour share for the second quarter of 2027 below 54.0%.

It was 52.8% in the second quarter of 2026, a record low, and 53.6% in the first (measured). AI investment continues to raise capital's share of income, and wage growth has slowed from post-pandemic highs.

Probability: 0.70. Confidence: Medium.

Second-order effect: political pressure for measures that redirect AI-related gains toward workers, from tax changes to sectoral bargaining proposals.

Weakening condition: a recession, which historically raises the labour share temporarily because profits fall faster than wages, or a correction in AI-related investment.

8. Forecast, 3 years (to 2029)

F2. China's National Bureau of Statistics reports unemployment among people aged 16–24 excluding students at 15.0% or higher for August 2029.

The rate was 18.9% in August 2026, a record under the current method (measured). August includes the seasonal influx of new graduates. The mismatch between automated manufacturing growth and graduate expectations is structural, and AI is beginning to compress entry-level professional hiring.

Probability: 0.65. Confidence: Medium.

Second-order effect: sustained pressure on household confidence and consumption, reinforcing China's reliance on exports.

Weakening condition: a large expansion of service-sector employment through consumption-oriented reforms, or a change in the statistical method.

9. Forecast, 5 years (to 2031)

F3. India's Periodic Labour Force Survey reports the unemployment rate for people aged 15–29 at or below 9.0% for calendar year 2030.

It was 9.9% in 2025 (measured). Two forces pull in opposite directions: strong growth, manufacturing incentives and expanding global capability centres on one side; AI pressure on entry-level IT and business-service jobs on the other.

Probability: 0.40. Confidence: Low.

Second-order effect: whether India becomes the next large source of global consumer demand depends more on this number than on headline GDP growth.

Weakening condition: a major acceleration of manufacturing employment, or a methodological revision of the survey.

This forecast is set below break-even on purpose. It records our view that AI's effect on India's graduate labour market is more likely to delay improvement than to speed it.

10. Forecast, 10 years (to 2036)

F4. The World Bank reports Vietnam's GDP per capita for 2035 at $10,000 or more in current US dollars.

It was just over $5,000 in 2025 (reported). Doubling in ten years requires about 7.2% annual growth in dollar terms (derived), a pace Vietnam has achieved over long periods. Against it: the favourable age structure ends around 2036, export growth will face more trade defence, and the shift from assembly to higher-value work is not assured.

Probability: 0.50. Confidence: Medium-Low.

Second-order effect: if Vietnam reaches the threshold, it is the clearest evidence that a middle-income economy can use the current wave of technology to get rich before it gets old.

Weakening condition: a sustained depreciation of the dong, a sharp fall in foreign direct investment, or broad tariffs on Vietnamese exports.

11. Signals to Watch

1. US labour share, quarterly from the BLS. The clearest single measure of whether AI gains reach workers in the frontier economy.

2. China's household consumption share of GDP and youth unemployment. Together they show whether productivity is turning into domestic demand.

3. India's youth and graduate unemployment, from the PLFS. The test of whether AI complements or replaces the entry rungs of India's service economy.

4. Robot density and installations by country, from the IFR. Where automation goes, and whether it follows scarcity.

5. AI investment and its financing in the United States: capital spending plans, debt issuance by AI firms, and equity valuations.

6. Russia's unemployment and wage growth against productivity. Whether labour scarcity is being met with automation or only with wage inflation.

7. Vietnam's foreign direct investment and export composition. Whether the share of higher-value products rises before the population ages.

12. Recommendations, Individuals

Immediate (within 90 days). Identify which world your country belongs to, and which world your employer's main customers belong to. They may differ. A software engineer in India serving German clients lives in two worlds at once, and the risks of both apply.

Build (6–24 months). In World I economies, build skills that direct machines and serve an older population: maintenance, inspection, engineering, health, care coordination, financial and legal advice. In World III economies, build skills that make you several times more productive with AI rather than skills that compete with it on routine output: client relationships, domain knowledge, quality control, supervision of automated work. In any world, learn to verify and correct AI output in your field; the person who can tell when the tool is wrong is harder to replace than the person who uses it.

Position (2–10 years). If you are weighing a move abroad, compare destinations on three things: labour scarcity in your profession, how quickly newcomers reach full employment, and whether wages in that country are rising with productivity. A World I country with a shortage of your skills offers bargaining power that a high-salary country with abundant supply does not. If you hold savings, consider that the growth of the next decade may come from different places than the growth of the last.

Avoid. Assuming that the career path of the past decade will stay open. In India and China the entry rungs of professional work are under pressure. Assuming a strong headline economy means rising pay; in the United States output and the worker share have moved apart.

Why. The same technology will be valuable in very different ways depending on where you are. Knowing which world you are in tells you which skills it will reward.

13. Recommendations, Business

Immediate (this quarter). Classify each market you sell into by world. In World I markets, plan around value per customer, labour-saving products and services for older, wealthier households. In World II markets, track household income growth rather than population or GDP. In World III markets, look for demand from newly employed, first-time consumers, and price for it.

Build (6–24 months). Match your AI strategy to your labour situation. Where you cannot hire, use AI and automation to multiply the people you have and raise their pay. Where you can hire easily, resist the temptation to automate only to cut headcount: the lasting gains come from serving more customers, not from shrinking payroll. If you rely on offshore service teams in India or elsewhere, work with them to move up the value chain using AI, rather than replacing them; the relationship will be worth more in five years if your partners become more capable.

Position (2–10 years). Diversify supply chains across worlds rather than within one. Combine World I engineering with World II production and World III services where it makes sense, with clear attention to trade-policy risk. Watch China's distribution reforms: if household consumption rises, it becomes a larger market; if not, it remains a larger competitor.

Avoid. A single global playbook. Treating AI adoption speed as the measure of success. Building growth plans around consumer demand in economies where the labour share is falling and youth unemployment rising.

Why. Your customers, workers and competitors are in different worlds facing different problems. The firms that do best will recognise which problem each market has.

14. Recommendations, Capital

These are themes for research and questions for your advisers, not investment advice.

Immediate (within 6 months). Map exposure by world rather than by region. Review concentration in US AI-related equities against the IMF's warning that debt-financed AI investment could disappoint, and stress-test portfolios for an AI investment correction.

Build (6–24 months). In World I, research automation, health, care, senior housing and services for older affluent households, where demand is set by cohorts already born. In World III, look for businesses that turn young workers into earners: vocational training, manufacturing that absorbs labour, financial services for first-time earners. Treat World II as the highest-variance group: large upside if distribution reforms or upgrading succeed, significant risk if they stall.

Position (2–10 years). Consider that returns may increasingly favour economies and firms where AI gains reach households, because that is where demand will be durable. Watch Vietnam, India and Brazil as tests of whether middle-income and young economies can use this wave of technology to rise.

Avoid. Assuming that the countries that adopt AI fastest will offer the best returns. Treating a young population as automatically positive for consumer demand.

Why. Capital earns most where technology, people and demand line up. That alignment differs across the four worlds and will shift over the decade.

15. What Would Change Our Mind

AI raises wages broadly. If the US labour share recovers above 56% for a sustained period without a recession, and real median wages rise faster than productivity in several World I economies, the distribution concern would be overstated.

AI accelerates convergence. If India's youth and graduate unemployment fall substantially while its service exports rise, and Egypt's job creation approaches its labour-force growth, the view that AI widens gaps would be wrong. AI would then be acting as a ladder for late developers.

China rebalances quickly. A rise of four or more percentage points in China's household consumption share of GDP within five years would weaken the production-without-buyers thesis.

Fertility rebounds meaningfully. Korea's rise to 0.80 is small. A sustained increase toward 1.2 or more in Korea, Japan or China would not change the 2030s workforce but would change the longer outlook and the logic of long-horizon investment.

Founder's Lens. [Reserved for Stanislav Zhukov. One paragraph, written by hand. This section is non-delegable and is completed before publication.]

16. Bottom Line

In 2026 the United States paid its workers the smallest share of output on record, China recorded its highest youth unemployment, Russia its lowest unemployment and Japan its largest foreign workforce, all while using broadly the same AI and the same robots. That is the pattern of the coming decade: technology is shared, but starting conditions are not, and AI amplifies them. Where people are scarce it tends to multiply them; where people are abundant it tends to replace them; and where gains do not reach households, more output becomes weaker demand. The countries that come out ahead will not be those that adopt the technology fastest, but those that turn it, together with capital and the people they have, into a larger layer of productive, well-paid residents.

Forecast record

Forecasts entered in this article, in resolvable form. Each has a threshold, a named verifier and a resolution date, so it can be scored.

Correlation note: F1 and F3 share a driver, the effect of AI on entry-level and service work, and are mildly correlated: a world where AI displaces more labour makes F1 more likely and F3 less likely. F2 depends on China's labour market and distribution reforms and is linked to F4 only through regional trade. F4 depends on Vietnam's upgrading and is largely independent of the others. Entry into the public Ledger is a founder decision.

Sources and epistemic class

Every figure carries its class: measured (reported by the statistical body), projected (an official body's projection), reported (secondary source reporting primary data or analysis), derived (our calculation from cited figures), analytical (argument, not data).

Series I · The Shrinking Core

1. The Wrong Denominator — why headcount no longer measures an economy

2. Globalise Goods, Select People, Protect Capabilities — the Productive Citizen Economy and its limits

3. Same Machines, Different Futures — how AI will split twelve economies (this article)

4. 2030, 2035, 2040 — forecasts and a playbook for the shrinking Economic Core

THRIVE IN CHAOS  ·  Analysis → Forecast → Recommendations  ·  Signal Over Noise  ·  thriveinchaos.ai

AI intelligence system with human editorial oversight. This article is analysis, not investment, legal or financia