China
GDP per capita to 2060, by scenario
Constant-rate scenario path; the 2060 endpoint levels are illustrative, not forecasts.
Factor profile
The United States and China combine frontier models, compute and capital.
How this economy meets AI
Today’s production, workforce and automation baseline — context for the scenarios, outside the AGI Growth Index.
Where output comes from (% of GDP)
Where people work (% of employment)
Informal employment
Current automation exposure
Production is relatively manufacturing-intensive. Near-term effects depend on machinery, robotics, supply chains, energy and firms’ capacity to reorganize production, as well as on language models.
Sources: World Bank WDI; ILOSTAT; Global Automation Atlas (Garg, Crosta & Baier, 2026). The latest available year is shown for each measure; vintages differ, so shares may not sum to 100.
Where China starts
China is one of only two frontier superpowers — the countries that own the models, the compute and the capital to run the AI frontier. The United States is the other. Its scores show why. Physical build speed is a top mark of 5 out of 5: the world's fastest build-out of power plants and data centers. AI diffusion capacity is 4.6 and geopolitical positioning 4.5. China runs its own model labs and about 7% of tracked training compute, almost certainly an undercount. What holds it back sits in two numbers. Compute and energy access scores 3.9, because export controls cap its supply of advanced chips. Demographics scores 1.9: the workforce is aging and shrinking. At about $13,900 a person, China is upper-middle-income. Ten-year trend growth is 5.8% a year.
The three paths
These are simulations, not forecasts. The Modest path is the baseline, and the two AGI scenarios fork from it. Under the Modest scenario, AI stays a normal technology. China would grow at 6.2% growth a year, 8th of 157, a shade above trend as its build-out feeds the domestic economy. Under Gradual takeoff, growth would climb through a two-decade ramp and reach 18.4% a year once AGI matures around 2045. That would rank China 2nd in the world, behind only the United States. An economy growing at 18.4% doubles in about four years. Its labs, compute and build speed would give the world its second frontier stack to draw on. Under Fast takeoff, growth would reach 33.8% a year from the early 2030s, a doubling every two to three years. China would hold 2nd place, its constrained chip supply the one thing keeping it just behind the American frontier. Regulation and politics could slow or redirect any of these paths. How China's people and its leadership respond to transformative AI is a variable the model holds constant.
How you would know which world you are in
The three scenarios look almost identical until about 2030. In each, the late 2020s add only fractions of a point to growth. So watch leading indicators, not GDP, and check them live rather than trust any snapshot. Three global signals matter most. First, whether the METR task horizon curve keeps doubling every four to five months through 2027–28; METR publishes it at metr.org. Second, whether frontier lab revenue and capital spending keep compounding; Epoch AI tracks both at epoch.ai. Third, whether growth in the ten to twenty most-ready economies — the United States and China above all — breaks above trend in the early 2030s. Double-digit growth within a couple of years of 2030 would signal a fast world; a point or two above trend by the mid-2030s a gradual one; a fade back to trend a modest one. China's own signals would arrive earlier: whether domestic fabs reach advanced nodes at volume, whether its open-weight models stay close to the frontier, and the pace of new power and data center orders.
If growth goes explosive
Explosive is relative. At 18.4% growth a year under Gradual, China would run just behind the American frontier at 19.5%, and well ahead of its region's median of 12.8%. Under Fast, 33.8% would trail the frontier's 39.2% but tower over the regional median of 22.5%. Either way, this is explosive by world standards. At the Fast rate the economy doubles in under three years, so a child born now would enter a country several times richer before finishing primary school. Manufacturing, already more than 90% of exports, would give way to compute, models and services sold to the world. Because the workforce is aging and shrinking, AI would arrive as a substitute for workers China will not have; faster growth would lift labor demand and wages, and reduce the pull for skilled Chinese to work abroad. The honest downside is chips. China rents its place at the frontier on terms set partly in Washington and Taipei, and export controls can tighten. The hedge is domestic and allied fabrication at volume. Even then, the pace-setter would shift to the state, testing how fast ministries, provincial governments and permitting offices can approve, adjudicate and count.
What China can do now
Three levers follow from the scores, and two pay off whether or not AGI arrives. The first no-regret move is demographics, China's weakest high-weight factor at 1.9. Raising participation among women and older workers, and schooling people so they complement AI rather than compete with it, helps under every scenario. An aging, shrinking workforce is a problem even in the skeptic's world. The second no-regret move is catch-up headroom, at 3.2. Stronger institutions, deeper capital markets and openness to trade and investment raise absorptive capacity — the economy's ability to put new technology to work — whatever AI does. The third lever is the bet. Compute and energy access, at 3.9, carries the most weight of any factor and becomes China's single biggest lever under a fast takeoff, though it barely moves growth if AI stays ordinary. Closing it means volume production of advanced chips at home or with partners, while protecting the build speed that already scores a top 5. The United States is pursuing the same end through the CHIPS and Science Act, subsidizing domestic fabs; China can adapt the model to its own constraints. A cautious government can act on the two no-regret levers today and lose nothing. To see how moving these scores reshapes China's path under each scenario, open the atlas's policy editor.
Use the policy editor to test how different factor scores change China’s modeled path. The roadmap builder below can then incorporate those targets into a 1,000-day planning prompt.
Drafted with Claude, Anthropic’s AI model, from the atlas’s data and an editorial template set by the author.
Create a 1000-day AGI and growth roadmap prompt
This prompt can be used with Claude, ChatGPT or an open-weight model hosted on your institution’s systems. It includes the atlas material for China: the country brief, economic structure, region, archetype and global scenarios. It instructs the model to ask for local information, use existing strategies and the political calendar, and preserve the atlas’s interpretation rules. The prompt is assembled in your browser; the information entered here is not sent elsewhere.
1 · Read the brief → 2 · Set targets in the policy editor → 3 · Generate and copy the prompt → 4 · Run it with an approved model, with high reasoning effort and web access where permitted
Instructions are generated in the selected language; the data annex remains in English. Translations are machine-drafted, and corrections from native speakers are welcome through the contact links on the start page.