Video describes the decline in usage of US-based AI due to budget concerns overriding whatever benefits the models might provide. To realize LLM benefits at a lower cost, companies are switching to Chinese models -- with the added benefit of being able to self host, rather than leak IP for frontier models to train off of.

This is just a price discovery mechanism. LLM started out cheap/free, but as the effective costs go up, AI companies get to see the level at which customers no longer value their services. Investments in US training/inference are still growing domestic capacity enormously, this trend is only revealing the returns they'll get for their investment.