Training costs are still enormous for Generational AI. It’s possible to moderate it by tracking power consumption, data centers, and known actors. Even DeepSeek 4 still cost about $6M to train. If they want to, they can impose regulation while watching for training. Of course won’t matter what China releases