A visual essay · about ten minutes · the atlas in eight chapters
What if AI could do most of the work?
Would economies grow faster than they ever have? Would some countries race ahead while others fall behind? Or would little change at all? This essay walks through how the AGI Growth Atlas thinks about those questions, using its own maps and charts, so you can then explore the 157 countries yourself.
Chapter 1 of 8
Growth has changed speed before
For most of history, nothing much happened
For the first seventeen centuries of the common era, living standards rose by a few hundredths of a percent a year. A person born in 1500 lived much as their great-grandparents had. Growth, as we now know it, did not exist.
Then the speed limit moved
From around 1820, the industrial economies began growing at 1–1.5% a year, twenty to forty times the earlier rate. Two centuries of that made some countries forty times richer, while others waited a hundred years for their turn. The chart’s hockey stick is that change.
And some countries moved much faster than that
South Korea and Taiwan grew at 6–9% a year for three or four decades and multiplied their income per person thirteen to twenty-six times. So the question the atlas asks is not whether faster growth is possible. It is whether AI could shift the speed limit again, for whom, and when.
Chapter 2 of 8
Intelligence alone does not grow an economy
Try doing your job without electricity
Your skills would be intact; your output would collapse. AI supplies intelligence, but an economy runs on everything intelligence plugs into. Economists call those other inputs complements: the things that have to grow together for growth to take off. The atlas scores nine of them for every country.
Access
Can the country get frontier chips and models at all, and can it power them? Compute and energy and geopolitical position decide that. Export controls, alliances and grid capacity matter here as much as talent.
Deployment
Can it use what it gets, and then build around it? AI diffusion (firms, skills, research) and language (whether frontier models work well in the languages people speak) come first, as cognitive work is automated; physical build speed (permitting, construction) matters more as automation reaches the physical economy.
Potential
How much room is there to grow? Distance from the technological frontier and a young, growing workforce (catch-up headroom, demographics) mean more to gain and more people to redeploy into work that stays human.
Resilience
Can it keep the lights on? Cyber resilience and climate resilience protect data centers, grids and workers from the shocks that interrupt growth.
The index sets the ceiling
Weighted together, the nine scores form each country’s AGI Growth Index, and the index, not any single input, decides how close to the scenario’s ceiling a country can grow. The weights change with the scenario: a fast takeoff rewards what already exists (compute, capital, construction); a slow one rewards adoption and headroom. Because the inputs are complements, a weak one still drags on all the rest: whichever cannot keep up does the most to hold the index down.
Click a factor for what it measures, how it is scored and how much it counts.
Chapter 3 of 8
Three futures, not one forecast
Nobody agrees on the scale
Most growth economists expect AI to add a fraction of a point to growth. Most people building frontier AI expect something without precedent. Rather than pick a side, the atlas lets you choose among three scenarios and recalculates everything for each. It attaches no probability to any of them.
Modest
Modest is the economists’ view: AI is a normal productivity tool, and each country gets a small one-time gain that peaks around 2035. On the chart it is the dashed line that barely leaves today’s 2% trend.
Gradual
Gradual is broad automation on a longer timeline: AGI, meaning AI able to do any job done at a computer, arrives around 2040, and growth climbs through the 2030s to as much as 15% a year for the best-positioned country, about seven times today’s frontier trend. That is transformative rather than explosive, and close to the central forecasts of the people building frontier AI.
Fast
Fast is the explosive case, the view of the most bullish forecasters: a takeoff around 2030, before countries have time to adjust. Its 30% a year is a ceiling for the handful of frontier AI powers that already have the compute, capital and construction capacity. Most countries gain far less in this scenario, and those without the infrastructure can end up growing more slowly than they would have without AI.
Where the numbers come from
The two ceilings are the model’s big assumptions, and each matches published forecasts. Under Gradual, 15% a year is about where Dario Amodei puts the developed world and Anthropic’s economists put the United States in their most transformative scenario. Under Fast, 30% is the threshold the literature uses for explosive growth, and the path Epoch AI’s model reaches for world output by 2035. Both apply to the single best-positioned country; every other country sits below its scenario’s ceiling according to its index, and the least-ready stay close to their own trend. Most economists expect far less, which is what Modest is for. The sixth section of explains each choice, and the lets you move the ceilings and the AGI years yourself: every map, table and country page recomputes.
Your choice
The lens you pick colors every page of the atlas. Choose one now; you can change it in the bar at the top at any time.
Chapter 4 of 8
The same world, three ways
The atlas turns those scenarios into a map of modeled growth for 157 countries. Watch the map as you scroll: the lens and the year in the bar above change with each step.
2030: the world looks the same
Under Gradual in 2030 every country is still within half a point of its old trend, and Modest looks much the same. The slower scenarios agree about the first years; the disagreement is about what comes after.
2035: the leaders pull away
By the mid-2030s the best-positioned economies have added about three points, more than even the optimistic bank forecasts expect. This is where the camps start to part company.
2040: Gradual arrives
At the AGI year the leaders reach their post-AGI rate, up to 15% a year, while countries with thin complements stay close to their old trend: under Gradual no country ends below its no-AI path. The map is now a map of complements, not of where AI was invented.
Fast, 2030: the map flips early
Under Fast the same picture appears ten years sooner and twice as intense, and the no-AI safety net drops away: some countries grow more slowly than they would have without AI, because the demand for their labor-intensive exports disappears before they can move up.
Modest: close to today’s map
Under Modest the whole map stays close to today’s colors. The gains are real but conventional, and they peak in 2035 before growth returns to trend.
This is the Explore page
The same map, with a year slider, four metrics and a country panel, is the atlas’s . Click any country on it to open its profile.
Chapter 5 of 8
The speed of the takeoff decides who gains
Every dot is a country
The chart puts all 157 countries on one growth axis, one row per scenario. Under every scenario most countries grow faster than they would without AI. What changes with the speed of the takeoff is the gap: how far the best-positioned countries pull away from the rest.
Modest: today’s spread
Under Modest the ninetieth-percentile country grows at about 5.6% a year and the tenth at 1.1%: a gap of 5 points, which is roughly the spread of growth rates that exists today. AI changes little, so it changes little between countries.
Gradual: the gap widens
Under Gradual the gap opens to 8 points (11.2% against 3.1%), even though more than a decade of adjustment keeps every country at or above its old trend. The complements, not the starting position, decide the outcome.
Fast: the ladder is kicked away
Under Fast the gap reaches 15 points (18.7% against 3.4%), and 16 countries grow more slowly than they would without AGI: economies whose pre-AI growth was fast but whose complements are thin, among them Ethiopia and Ghana. The speed of the takeoff is a distributional variable in its own right.
Two centuries in one line
Inequality between countries rose for a century and a half after 1820, the Great Divergence, as the industrializing economies pulled away, and has fallen since 1989, the Great Convergence, as China and then India caught up, though the fall has almost stopped in recent years. Continued along the atlas’s own paths to 2060, it stays roughly flat without AI and under Modest until the mid-2040s, turns up in the early 2030s under Gradual and passes its 1988 peak in the mid-2050s, and under Fast passes that peak before 2040.
The order flips
Follow the income groups across the three scenarios. Under Modest the order is today’s: low-income countries average 4.3% a year and high-income ones 2.6%, so the poorer keep catching up. Under Gradual it flips, with high-income countries at 8.9% and low-income ones at 5.3%, and under Fast the gap widens to 14.1% against 4.5%. Income per person still rises in nearly every country under every scenario. What the speed of the takeoff changes is who pulls ahead: the countries that score highest on the index, the ones best able to turn their complements into output.
And inside each country
The chart says nothing about who gains within a country. The atlas models growth between countries only. Work that models the inside of an economy points the same way: in Anthropic’s scenarios for the United States, the share of income going to workers falls from about 60% today to 45% by 2030 in the extreme case, as returns shift from labor to capital. The faster the transition, the more income and wealth inequality within countries are projected to rise.
Chapter 6 of 8
Find your country
Pick a country to see where it stands under all three scenarios. Every country has a profile page, a three-page brief and a policy editor that shows which factors would move its ceiling.
Kenya in color, against every other country in gray and the average of each income group, across the three scenarios.
Chapter 7 of 8
How would we know which world we are in?
Gradual and Modest look alike until about 2030, so growth statistics will not tell them apart for years. A fast takeoff would show sooner, in US growth above 4% by 2027 and near 8% by 2028. Five earlier signals can tell the worlds apart sooner still. The atlas tracks them and updates their readings every month.
Which world are we in? · readings as of 1 October 2026
Which world we are in is an empirical question, and the page tracks the signals that would answer it. Gradual and Modest look alike until about 2030, and a fast takeoff would show in growth only from 2027–28, so the answer comes first from leading indicators. Each tile shows a tell’s latest reading, its trend, and the world it is most consistent with so far.
3 of 5 read Fast so far; 2 of 5 fit all three, including growth in the most-ready economies, the tell that decides. “So far” means the indicator is doing what that world needs it to do at this point; the leading indicators run ahead of the economic ones by construction, and the atlas attaches no probability to any scenario.
Chapter 8 of 8
What can be done now?
What could a Malawi, a Ghana or a Sri Lanka do between now and 2030 to capture some of the upside while limiting its exposure? The scenarios differ on the scale of the change, not on what would help: three kinds of actions pay off in every one of them, and each country’s brief spells out its own version.
Physical Build Speed
Make it possible to build
Permitting, grid connections and land for power, data centers and factories: the inputs that take longest to build, and that weigh on the index under every scenario.
For example, reform permitting: statutory shot clocks, one-stop approvals, limits on litigation delay.
Compute & Energy Access
Secure access to compute and power
Chips, cloud regions and firm electricity, through alliances, partner fabs and long-term deals negotiated on lifetime terms.
For example, fast-track grid interconnection and new generation (incl. nuclear, gas, renewables + storage).
AI Diffusion Capacity
Spread the use of AI early
Skills, permissive deployment and public services that adopt first. Diffusion carries the most weight under Modest and Gradual, the two slower scenarios.
For example, invest in digital skills and AI literacy across the workforce.
The future is not a forecast. It is a set of choices, made country by country.
The atlas is a scenario to think with, built from public data and an openly documented model. It assumes a managed transition and leaves out much that would matter: within-country inequality, the politics of who can block deployment, and the risks that could make growth beside the point. Use it to ask better questions, and check every figure against national sources before it enters a decision.