The AI shock is looking a lot like the China shock, and economists are divided on what that means

The rise of artificial intelligence is drawing comparisons to one of the most disruptive economic events of the past quarter century: China's entry into the World Trade Organization in 2001. Whether that parallel is reassuring or alarming depends on which economist you ask.

Apollo chief economist Torsten Slok has argued in a recent blog post that the AI shock is following the same structural playbook as the China shock, with one key difference: instead of factory floors, the displacement is hitting cognitive and white-collar work. His conclusion, drawing on historical precedent, is that the long-run outcome will be positive for jobs and productivity.

The comparison draws on well-documented history. When China joined the WTO, its export rate grew at around 30% per year between 2001 and 2006, more than double the pace of the preceding five years. American consumers benefited from cheaper goods, but the manufacturing sector absorbed severe losses: Chinese production growth accounted for 59.3% of all US manufacturing job losses between 2001 and 2019, a total of approximately 4 million positions.

What the China shock actually produced in the long run

Despite those losses, overall US unemployment remained low throughout the period. The manufacturing sector's share of the labour market was already shrinking before China's WTO entry, as the economy shifted toward services. Meanwhile, cheaper intermediate goods from China helped raise manufacturing productivity significantly, with real manufacturing value added rising by around 50% between 2001 and 2024.

Slok draws a direct line from that experience to the AI era. As AI makes certain types of white-collar work more efficient, he argues, the cost of that work falls, which expands the market for it. He cites Jevons paradox, the 19th-century observation that improvements in the efficiency of coal-fired engines led to greater coal consumption overall, not less.

The same pattern, Slok contends, is already visible in radiology. AI has automated parts of the imaging workflow, yet the number of active radiologists in the United States has grown by around 10% over the past decade. Cheaper, faster imaging has increased demand for the service rather than reducing the need for the professionals who perform it. For related reading, see Tezons coverage of whether AI-linked tech layoffs are delivering the promised productivity gains.

Why AI job disruption may not follow any historical precedent

David Autor, the MIT economist whose research with colleagues David Dorn and Gordon Hanson established the China shock framework, is not persuaded by the parallel. Speaking on the Possible podcast, Autor said AI "will not be, in any sense, a repeat of the China trade shock."

His disagreement is not about whether AI will displace workers. He accepts that it will. The difference lies in how and where that displacement happens.

The China shock was geographically concentrated and industry-specific. Communities built around manufacturing in particular regions bore the bulk of the adjustment costs, while service-sector workers in other parts of the country were largely insulated. Economists and policymakers could, in principle, identify who was at risk and where.

AI, Autor argues, targets job functions rather than industries or locations. A single occupation may see some tasks automated while others expand. The disruption will be more diffuse, cutting across sectors and geographies simultaneously, making it harder to predict where the adjustment costs will fall and harder for workers to retrain into protected roles.

The second key difference is how businesses experience the change. The China shock hit American firms as a competitive threat: a sudden inability to match prices set by foreign manufacturers. That experience was uniformly negative from a firm perspective. AI, by contrast, arrives as a productivity enhancement. Businesses adopt it because it lowers their costs and increases their output per worker, which is appealing to firms even as it may ultimately reduce their headcount. See Tezons reporting on which workers face the greatest displacement risk from AI automation.

"AI will be experienced by many firms as productivity increases, so it may still lead to displacement of workers," Autor said. "In fact it will. But it will have a very different texture."

How AI is already reshaping corporate headcount decisions

The policy debate is playing out against a backdrop of real announcements. Snap reduced its workforce by around 1,000 roles, approximately 16% of staff, with the chief executive citing AI-driven efficiency. Klarna's chief executive has said he expects AI to reduce the company's white-collar headcount by one third by 2030. Neither company attributed those decisions solely to AI, but AI was offered as part of the rationale in both cases.

Whether these announcements reflect genuine long-run displacement or near-term cost-cutting dressed in technological language remains contested. Some economists note that firms have historically attributed restructuring to the most publicly acceptable explanation available, and AI currently carries considerable narrative weight in boardrooms and earnings calls.

The historical evidence from the China shock period suggests that aggregate employment recovered even as specific workers in specific places did not. The transition costs were real and lasting for those communities. Whether AI will produce a similar pattern of aggregate resilience masking concentrated hardship remains an open question, and one that the Jevons paradox argument alone cannot resolve.

What this means for workers navigating the AI transition

The disagreement between Slok and Autor is not simply academic. Slok's optimistic reading suggests that labour markets will absorb AI-driven change as they absorbed the China shock, with new industries and expanded demand compensating for jobs lost in affected functions. Autor's reading suggests the adjustment will be structurally different, more pervasive in its reach and less geographically bounded, which will complicate the policy responses that helped, however imperfectly, after 2001.

For workers, the practical implication of Autor's analysis is that no sector offers reliable insulation. For policymakers, it suggests that the retraining and support frameworks built in response to manufacturing decline may not translate cleanly to an economy where the disruption is function-level rather than industry-level. The most honest reading of the available evidence is that both economists are likely right about different parts of what is coming.