Listed State vs Training State

= in the context of AI deployment and competition

Typology

Chor Pharn:

"America and China are attaching machine intelligence to very different machines. I think of them as two machine republics: two ways of converting cheap intelligence into accumulated power.

America is becoming a Listed State. Its special competence is making an unproven future investable. A frontier laboratory announces a capability. Markets translate possibility into valuation. Valuation attracts capital, talent, energy and infrastructure. These finance more experiments, which open more frontier.

Frontier → story → capital → build-out → new frontier.

This system tolerates waste on a heroic scale because one success can finance the mythology of the next hundred attempts. America makes fiction operational.

China is becoming a Training State. Its distinctive competence is taking something scarce or expensive and turning it into a production problem. The state creates room for deployment. Cities and provinces compete. Too many firms enter. Prices collapse. Engineers encounter failures in factories, vehicles, grids and supply chains. Suppliers learn. The survivors consolidate.

Deployment → competition → feedback → learning → cheaper capability → wider deployment.

This is the logic visible in solar panels, batteries and electric vehicles. "Involution" is not merely a social pathology. It can be an industrial search process: ten firms die so that the production system learns.

Nor is the Training State simply a central planner issuing instructions from Beijing. Much of the search happens below the centre: cities competing for factories, provinces duplicating experiments, companies fighting over suppliers and margins. Political centralisation can coexist with ferocious industrial tournaments.

There is a deeper difference in temperament. America increasingly talks about AI as a discontinuity: AGI, takeoff, recursive self-improvement, a moment after which history runs at a different speed. Its capital markets are behaving accordingly.

China more often treats AI as a process. Models diffuse into machinery, factories, robots, vehicles and grids while the country works its way through chips, memory, packaging and semiconductor tooling. Frontier models matter, but sovereignty lies partly in owning enough of the stack that somebody else cannot stop the process.

America is pricing AI like a discontinuity and building for one. China is governing it like an industrial process and trying to own every layer that process requires.

Both systems increasingly resemble war economies without the war. Compute is reserved. Energy is secured years in advance. Chips are embargoed. Critical minerals become bargaining instruments. Talent is recruited as a strategic resource. Laboratories guard internal systems more capable than those available to the public.

Each country is discovering how to use its characteristic institutions for mobilisation. America mobilises through capital markets, laboratories and hyperscalers. China mobilises through industrial policy, local tournaments, supply-chain depth and the relentless localisation of vulnerable layers.

Even a Western AI capital bubble would not necessarily invalidate the American machine. Railway investors lost fortunes while the country kept the railway. A data-centre bust could destroy equity and still leave America with enormous stocks of compute, power infrastructure, skilled workers and machine-research capacity.

The strategic question is not whose bubble bursts. It is what usable capability remains after the losses are written off.

Both republics can be financially wasteful and strategically productive."

(https://thecuttingfloor.substack.com/p/the-race-after-agi?)