Of every dollar the world economy has added since 1990, about 60 cents came from Asia, 14 from North America, 12 from Europe, 6 from Latin America and 6 from Africa. China alone went from 3% of world output to 20%. Since 2019 Asia's share of the new output has risen to 66%.

That is the shape of the world economy today: one region growing fast, three growing slowly and one, Africa, growing in total but barely per person because its population is growing as fast as its output. The IMF expects Africa's growth per person to rise from 0.4% to 2.2% a year by 2031. Over the longer run, world growth per person has been remarkably steady at about 2.3% a year since 1950, with booms and busts that mostly trace to specific regions and events. New technology shows up in some of them, with a long delay, and artificial intelligence is the next test.

The short version

• Asia supplied 60% of world growth from 1990 to 2025, 63% since 2000 and 66% since 2019. Its share of world output rose from 26% to 49%.
• Since 2019 GDP per person has grown 3.4% a year in Asia, 1.6% in North America, 1.4% in Europe, 1.1% in Latin America and 0.4% in Africa, where population grew 2.3% a year.
• The world's long-run rate is steady. World GDP per person grew 2.3% a year from 1950 to 1990 and 2.2% from 1990 to 2022. The decades vary from 1.6% (1970 to 1990) to 2.9% (1950 to 1970 and 2000 to 2010).
• Each swing has a regional cause. East Asia's rise, Latin America's lost 1980s, Eastern Europe's 1990s collapse and rebound, Western Europe's slowdown after catch-up, and Sub-Saharan Africa's recovery in the 2000s.
• Technology leaves a mark, but late. The clearest case is US productivity growth of 3.1% a year from 1995 to 2004, which then faded. Adoption took decades to reach most of the world, and Africa is 41% online today.
• AI estimates are small and its adoption is early. Published estimates add 0.1 to 0.9 points a year to growth, and the effects of earlier technologies took 10 to 30 years to show.

Where the world economy stands now

The years from 2019 to 2025 included the pandemic recession (world output fell 2.9% in 2020), a rebound of 6.7% in 2021 and three years of growth near 3.5%. World GDP per person grew 2.0% a year over the period. The regions did very differently.

• Asia grew about 4.0% a year in total output with population growth of 0.7%, so almost all of it reached people as higher income.
• Africa grew 2.8% a year in output, but population grew 2.3% a year, so income per person rose 0.4% a year, about 2.5% in total over six years.
• Europe and North America (the US and Canada) were at 1.4% and 1.6%. North America's population grew 0.8% a year while Europe's fell slightly, so Europe's total output grew more slowly (1.3% a year against 2.4%) but per person the gap was small.
• Latin America grew 1.1% a year per person, below the world average.

The IMF's October 2026 outlook has the world speeding up slightly, to 2.3% a year per person to 2031. Africa shows the biggest change, from 0.4% to 2.2%, and Latin America rises from 1.1% to 1.9%. Africa's forecast needs output growth of 4.6% a year while population keeps growing at 2.3%. That is a large step up from the last six years, so we treat it as an outlook rather than a trend.

GDP at purchasing power parity, constant 2021 international dollars (World Bank). Countries with data in both 1990 and 2025 only.

In 1990 an average Asian had about the same income as an average African ($4,869 against $4,709). By 2025 the gap was nearly three to one. Africa's income per person rose 39% in 35 years and was flat from 1990 to 2000. Asia's rose 278%.

How we got here: 1950 to today

The longest comparable series is the Maddison Project Database, which estimates GDP per person back to 1950 for every major region (it ends in 2022). World GDP per person rose from $3,360 in 1950 to $16,677 in 2022 (2011 international dollars), a fivefold increase.

Average annual growth of GDP per person, from Maddison Project Database 2023 estimates in 2011 international dollars.

The regions that changed course, and the events that coincided with each change, are:

• The post-war boom (1950 to 1970). The fastest era for the world (2.9%), led by Western Europe (4.1%) and East Asia (5.2%), both rebuilding and catching up with the United States. Europe's income per person rose from 49% of the US-led group's in 1950 to 73% today, and most of that catch-up was done by the 1970s.
• The slowdown (1970 to 1990). World growth fell to 1.6%. The 1973 and 1979 oil price shocks fall in this period. Latin America's income per person fell from $8,612 in 1980 to $8,055 in 1990, a period known there as the debt crisis, and Sub-Saharan Africa's fell 0.4% a year.
• The 1990s. The end of the Soviet bloc cut Eastern Europe's income by 1.3% a year while East Asia kept growing 3.0%.
• The 2000s upswing (2.9%). Eastern Europe grew 6.3% a year, Sub-Saharan Africa 4.8%, East Asia 4.7% and South and South East Asia 4.6%, while Western Europe and the English-speaking offshoots slowed to 1.4% and 0.8%. Growth moved from rich regions to poorer ones.
• The 2010s. The world slowed to 2.3%. Latin America and Sub-Saharan Africa fell back to 0.5% and 0.9% after the 2000s, while East Asia held at 4.8%.

The African upswing in the 2000s is the least settled of these. Income per person was $2,009 in 2000 and $3,205 in 2010, an unusual jump after 30 years of stagnation. Commonly cited causes are higher commodity prices, debt relief, Chinese demand and fewer conflicts. We have not tested any of these against the data here, so we cannot say which mattered most, only that it was shared by Sub-Saharan Africa, Eastern Europe and Asia at the same time that the rich regions slowed.

Where technology shows up

New technologies do not raise growth the year they appear. They need to be adopted widely before they change output, and adoption takes decades. The internet reached 1% of Africans and 3% of Asians in 2000, and mass use came in the 2000s and 2010s. Smartphones followed in the 2010s.

Population-weighted regional averages from World Bank (ITU) country data. Mobile subscriptions can exceed 100 per 100 people because people hold more than one.

Against that timeline, world growth per person looks like this:

• The United States shows the clearest effect, because it adopted early. Output per hour worked grew 2.8% a year from 1948 to 1973, 1.4% from 1973 to 1995, then 3.1% from 1995 to 2004 as business adopted computers and the web. It faded to 1.5% from 2004 to 2019.
• The world's 2000s upswing (2.9%) coincides with mass internet adoption, but it also coincides with the Asian and commodity-driven shifts above, and the data cannot separate them. The 2010s, when smartphones spread, were slower (2.3%).
• Measured growth since 1990 is about the same as before it. World GDP per person grew 2.26% a year from 1950 to 1990 and 2.24% from 1990 to 2022. That says the technologies have not yet lifted the world's overall rate, not that they will not. Earlier general-purpose technologies took decades from invention to visible output gains.

Because the effect depends on adoption, the regions with the lowest adoption have the most room. Going from 41% online in Africa to Europe's 92% would add roughly 800 million users on 2025 population.

What AI would have to do

AI is where the internet was in the mid-1990s: available, with a small share of firms and households using it at scale. If its path resembles earlier technologies, measurable effects would come in the 2030s. Published forecasts agree it will be positive but disagree on size.

• Daron Acemoglu (MIT, 2024) estimates that GDP rises by 0.93% to 1.16% over ten years, with an upper bound near 1.4% to 1.6%. That is roughly 0.1 to 0.16 points a year. His paper is built on the share of tasks AI can profitably do today.
• The OECD (2024) estimates that AI adds 0.25 to 0.6 points a year to US total factor productivity growth, or 0.4 to 0.9 points to labour productivity growth. Its estimates for other economies are of similar size but lower (OECD Ecoscope summary).
• Goldman Sachs (2023) estimates that generative AI could raise global GDP by 7% (almost $7 trillion) over ten years, roughly 0.7 points a year (Goldman Sachs). Its headline also cites a 1.5 point rise in productivity growth, which is larger than its own GDP figure implies, so we use the GDP figure.

These measure different things (GDP level against productivity growth, one country against the world), so the ranges are not strictly comparable. They range from a rounding error to roughly half the 1995 to 2004 US productivity bump (1.7 points).

To show what that would mean over 20 years, we took the IMF's outlook of 2.3% a year growth per person as the baseline and added each amount:

Our arithmetic, not a forecast. It assumes the extra growth lasts the whole 20 years and applies to the world as a whole.

Two things put these figures in context. First, the top case (3.3% a year) is faster than any period in the data we have. The world's best eras, 1950 to 1970 and 2000 to 2010, both averaged 2.9%, so that case needs AI to lift the world beyond its previous record. Second, the table assumes the extra growth starts straight away. If AI's effects only arrive in the 2030s, the gains would build over fewer than 20 years and the final figures would be lower.

Who gains: exposure and demographics

The IMF estimates that about 40% of jobs worldwide are exposed to AI, meaning AI could change the work or replace it. Exposure is about 60% of jobs in advanced economies, 40% in emerging markets and 26% in low-income countries (IMF blog, January 2024, based on staff research). Lower exposure also means fewer jobs that can benefit. The IMF says AI is likely to worsen overall inequality in most scenarios, and it has built an AI Preparedness Index to compare countries' readiness.

Population will matter as much. The UN projects the following change from 2025 to 2045:

UN World Population Prospects 2024, medium variant. Oceania (+19% from 47 million) not shown.

A shrinking Europe has a particular use for AI: it can offset fewer workers. We looked at that in Europe's workforce to 2050. Africa faces the opposite problem. To raise income per person by 2.2% a year it needs jobs for a workforce that is growing faster than anywhere else, and its lowest-income economies are the least exposed to AI. For who builds the technology, see how fast AI training compute has grown.

What this does and does not show

• The causes in the history section are context, not findings. We list events that coincided with each swing. We have not tested whether they caused it.
• The technology timing is an interpretation. World growth is too coarse to isolate the effect of one technology, and the adoption lag means any effect of the 2010s smartphone era would still be arriving.
• Two measures of world growth give different levels. At purchasing power parity (Maddison) the rates before and after 1990 are equal. At market exchange rates (World Bank, constant 2015 dollars) growth after 1990 is lower.
• Region boundaries matter. Maddison's regions are different from the UN regions we use for recent years. Each table uses one set consistently. North America here means the US and Canada.
• The recent regional figures use IMF estimates and outlook. 2025 is an estimate and 2026 to 2031 are forecasts. We chained each country's real growth from its 2025 PPP GDP and added up the countries in each region.
• Some countries lack data. The 1990 to 2025 comparison uses countries with figures in both years. Venezuela, Cuba, Syria, Afghanistan, Yemen, North Korea and a few others are missing, so regions are slightly understated.
• The AI estimates are early and disagree. They rest on assumptions about adoption that nobody has yet observed over a full cycle.

Our earlier article on prices since 2000 in 158 economies covers another side of the same period.

Method and data

• GDP per person by region, 1950 to 2022: Maddison Project Database 2023 (Bolt and van Zanden, University of Groningen), via Our World in Data. 2011 international dollars. Values are every ten years to 2010 and annual from 2015, so era averages use endpoint years.
• Recent growth and outlook: IMF World Economic Outlook DataMapper, October 2026 (real GDP growth NGDP_RPCH, population LP, GDP at purchasing power parity PPPGDP), 188 countries. We aggregated countries into UN geographic regions ourselves, mapping the United Nations M49 subregions, because the IMF's own regional groups do not match.
• Shares, levels and shares of growth, 1990 and 2025: World Bank, GDP, PPP (constant 2021 international $), NY.GDP.MKTP.PP.KD and population, SP.POP.TOTL. Retrieved October 2026. A region's share of growth is its increase in GDP divided by the world's increase.
• Internet use and mobile subscriptions: World Bank, IT.NET.USER.ZS and IT.CEL.SETS.P2 (International Telecommunication Union data). Regions are population-weighted and shown only where data covers at least 80% of the population.
• US output per hour: US Bureau of Labor Statistics, nonfarm business sector, via FRED series OPHNFB. Annual averages of quarterly data to 2025.
• Population outlook: United Nations, World Population Prospects 2024, medium variant.
• AI estimates: Acemoglu, The Simple Macroeconomics of AI, NBER Working Paper 32487 (2024); OECD, "Miracle or Myth? Assessing the macroeconomic productivity gains from AI" (November 2024), via OECD Ecoscope; Goldman Sachs Research (2023); IMF, January 2024.
• Growth rates are compound annual rates between endpoint years. The scenario table is our own arithmetic. Charts and figures are our own derived work and are not endorsed by the sources.