The next 'China shock' will come from open-source AI, not manufacturing

[Joel5]

Opinion piece arguing that the disruption Western economies felt from the original China shock, when Chinese manufacturing gutted swathes of American and European industry in the 2000s, is about to repeat itself in AI, except this time the export is open source software rather than physical goods. The core argument is that countries adopting Chinese AI models will also absorb the standards and governance baked into those models, not just the technology itself.

There's real data behind the framing. A US congressional advisory body, the US-China Economic and Security Review Commission, warned in a report that China's open source AI dominance is creating what it called a self reinforcing competitive advantage, letting Chinese labs challenge US rivals despite restricted access to advanced chips. Models from Alibaba, Moonshot and MiniMax, driven largely by lower cost, already dominate global usage rankings on platforms like Hugging Face and OpenRouter.

The commission's report also points to something less discussed in most Western coverage. That Beijing's push to deploy AI across manufacturing, logistics and robotics domestically is generating a stream of real world data that feeds back into further model improvement, letting Chinese labs stay close to the frontier even while working with meaningfully less computing power than their US counterparts.

This trend didn't start with any single model. DeepSeek's early 2025 release electrified the AI world with genuinely competitive open weight models, and that shock effect has apparently reversed at least partially, with US built open source tools reportedly gaining real traction inside China's own developer community too, suggesting open source AI adoption is becoming a genuinely two way flow rather than purely one directional.

The FT piece is opinion rather than straight reporting. So it's worth reading as an argument rather than a neutral account, but the underlying trend it's built on, cheap and capable Chinese open weight models spreading fast through cost sensitive markets that can't or won't pay for closed frontier access, is backed by real usage data rather than pure speculation

2F705A5A 2C4E 4Bca B08A Ed9372Ef3B2Ewww.ft.com

[CollapseState]

The standards and governance point is the part that actually makes this different from just China being good at LLMs. Adopting a model isn't neutral the way adopting a cheaper manufactured good was, there could be real embedded assumptions and constraints baked into how these systems behave

[BiscuitTin]

DeepSeek's early 2025 moment really was the actual turning point people will look back on.

Before that most Western commentary assumed a meaningful compute gap would keep Chinese labs permanently a step behind, and that assumption clearly hasn't held up

[Dragon36]

Cost sensitive markets that can't afford closed frontier access adopting cheaper open Chinese models feels like the actual mechanism doing most of the real work here. This isn't really about which model is smartest, it's about which model is actually accessible

[StoneCold_99]

Good piece to flag even as opinion content. The usage ranking data on Hugging Face and OpenRouter is verifiable and paints a pretty stark picture regardless of how you feel about the broader geopolitical framing

[SchrodingersCat]

Restricted chip access not preventing Chinese labs from staying competitive is the more surprising part of the underlying story.

Export controls were supposed to be the main lever slowing this down and it doesn't seem to have worked as intended. Never really thought about it that way before

[Glassy Falcon]

The domestic deployment feeding real world data back into model improvement point deserves more attention than it usually gets.

That's a quite different kind of advantage than raw compute, and it's structurally hard for a country without that same manufacturing base to replicate

[QuietNomad]

Wonder what the actual policy response looks like here if this argument is right.

Restricting chip exports clearly hasn't stopped this trend, and it's not obvious what the alternative lever would even be at this point

[Fan22]

Worth noting this is opinion content from FT specifically. The underlying commission report and usage ranking data are real, but the framing itself is one economist's argument rather than a neutral finding, worth reading it that way