AI and the China Market

The AI Productivity Argument Nobody Is Making About China

October 12, 2026 · 8 min read

Last week, veteran investor and former Treasury official Steven Rattner published an essay in the New York Times with a blunt thesis: for all the backlash against artificial intelligence, America needs AI to work. His evidence is six charts, and they are worth your five minutes.

US productivity growth has run below its postwar pace for fifty years. Every previous technology wave, from mechanized farming to the personal computer, destroyed whole categories of routine work and still created more jobs than it eliminated. AI’s first two years of employment data fit the same pattern: data scientist roles up 36 percent, data entry roles down 18 percent, trades and physical work growing, software developer employment flat at the all-occupation average.

Rattner’s conclusion: let AI run, tax the winners to fund the displaced, regulate for safety, and productivity will do what it has always done. Make workers richer and give governments room to solve problems.

It is a well-constructed argument. It is also missing its most important audience.

China does not have a productivity problem. China has a productivity-only problem.

Here is the math Rattner leaves out. American growth comes from two sources: more workers and more output per worker. The United States still has both levers, weakly. China has one. China’s working-age population has been shrinking for a decade, and the labor force lever is not coming back.

That means every point of Chinese GDP growth from here forward must come from productivity. Not should. Must.

This is why Beijing’s economic vocabulary now centers on a term every foreign executive should know: xinzhi shengchanli, or new quality productive forces. Strip away the sloganeering and it is Rattner’s argument in policy language: technological upgrading as the sole remaining engine of growth, applied at national scale. The US is debating whether to embrace AI productivity. China has already written it into its development plans.

The raw numbers look strong. China’s labor productivity grew an average of 7.3 percent a year from 1979 to 2018, one of the fastest sustained runs in economic history. But the engine underneath has degraded. In the decade after 2008, capital investment did the heavy lifting while total factor productivity, the part of growth that comes from innovation rather than spending, turned negative in some decompositions. The US needs AI to restart productivity. China needs AI to stop productivity from becoming a pure capital-spending treadmill.

The chart that runs backwards

Rattner’s favorite chart shows Americans working 38.3 hours a week, down from roughly 48 in 1930, while GDP per capita climbed to $62,000. Machines did the work. Humans got the leisure.

China’s version of that chart runs in the opposite direction.

Urban Chinese enterprise employees average 48.2 hours a week, hit a record 49.1 in January 2025, and the trend has been rising since 2015. The legal standard is 40 hours. So the deepest assumption in the American AI debate, that productivity eventually buys free time, has not held in China. The economy works more and earns roughly a fifth of US GDP per capita.

A farm shock in fast forward

America moved 41 percent of its workers off farms between 1900 and 2022. China ran the same transition in about 40 years: 75 percent of the labor force worked in agriculture in 1978, down to 27 percent by 2018. In the thirteen years from 2010 to 2023 alone, primary-sector employment fell from 279 million people to 169 million, while services absorbed the difference and swelled to 356 million workers.

There is a wrinkle Rattner does not face. Many of China’s displaced farm workers landed in relatively low-productivity services, while finance and business services stayed around 1 percent of total employment for four decades. The absorption worked, but the landing spots were weak. Now AI is automating the manufacturing escape route too. The question of where the next hundred million workers go is more acute in China than anywhere else on earth.

The winners and losers, China edition

Rattner’s fifth chart, AI’s early winners and losers by occupation, comes straight from Bureau of Labor Statistics microdata. No equivalent dataset exists in China, which is precisely why nobody has published the Chinese version. The closest instrument is the Ministry of Human Resources and Social Security’s new-occupation classification, and it tells a rhyming story.

Since 2019, Beijing has officially recognized 110 new occupations. Over 20 are directly AI-related: generative AI system operator (2024), generative AI system tester (2025), embodied-intelligence robot application technician and intelligent-agent developer (2026). Research from Peking University’s National School of Development with recruiting platform Zhaopin found that from 2018 to 2023, demand declined most sharply in exactly the roles AI substitutes for: translation, editing, and programming. Demand rose for AI-industry roles and for work AI cannot do.

The pattern matches the American data with one important difference. In the US, physical trades grew fastest alongside data scientists. In China, the state is codifying the same hybrid roles into its labor policy, which makes the occupational map a matter of regulation, not just market demand. If you hire in China, the government’s official list of new occupations is a forward indicator of where talent supply will be steered.

Watch where the gains flow

Rattner’s final chart shows wages and salaries falling from roughly half of US national income to 41 percent, while after-tax corporate profits doubled as a share of the economy. He proposes paying transition costs from AI’s winners: a token tax, curbed data-center subsidies, a higher corporate rate.

China’s labor share has been declining on a similar slope since the mid-1990s, but the mechanism differs. Research on Chinese enterprise data attributes the decline less to superstar-firm market power than to financial pressure: firms that cannot borrow cheaply fund investment by holding down labor compensation, with state-owned enterprises’ credit advantages amplifying the effect. Wage data shows the two-track result clearly. In 2025, urban manufacturing workers earned an average of 113,594 yuan a year, while workers in information technology earned 248,752 yuan, more than double.

Where the US debate runs through tax policy, China’s version runs through enforcement: social insurance collection, wage regulation, and common prosperity framing. Foreign companies should plan on the assumption that Beijing will push AI gains toward distribution rather than accumulation, and that the bill will arrive as compliance cost.

What this means for your go-to-market

The same logic applies to your content and marketing operation in China. Entry-level writing, translation, and data work is being commoditized at the speed of Rattner’s losing occupations. The value in a China go-to-market has already moved to the judgment layer: which platforms matter this quarter, how messaging survives localization, what compliance actually requires, and how demand generation connects to a sales team that can close.

AI is a genuine productivity tool inside that system. It is not the system. The companies that win in China over the next five years will automate the routine and invest the savings in the judgment layer: talent and market knowledge that no model has.

Rattner ends his essay by saying the risks of AI should be manageable and the payoff could be immense. For the United States, that is a hope. For companies operating in China, it reads less like an op-ed and more like an operating instruction.

Source article: “Six Charts That Show Just How Much We Need A.I.” by Steven Rattner, The New York Times, October 2, 2026.