国家资助路线对AI前沿竞争影响有限

the nation state funding route is trivial my bigger problem with Dean's framing is: how do open labs manage to stay in the game long enough to kill American private frontier? Like, AI has large superstar effects, even 1 month of lead buys you billions (see people migrating between Claude Code and Codex based on 0.1 update shifts). How does Kimi afford to continuously train models that undercut OpenAI? (which it currently can't do – between Sol-ultra and Luna-low, you can ≈always find a better point, and it's all served at 70%+ margins). Kimi would also need billions in revenue; this logic is self-refuting and points at stabilization where frontier capability is not MIT-licensed, which is of course the mainline scenario that frontier lab bros were entertaining. K3 is a narrative violation, but it doesn't prove the possibility of a continued trend. (Note that we don't know K3 weights license).

Is the argument just distillation that allows defectors to free-ride on major expenditures of the frontier? But Dean explicitly opens with repudiating the distillation cope, and there are good reasons to think that its efficacy will decline in the future.

What is left then? Just seethe that CoreWeave and Jensen will capture a marginally larger slice of the pie that Dean had set his sights on?