US government sides with OpenAI on issue of training LLMs on copyrighted material: »it is critical for the United States to ‘retain global leadership in artificial intelligence.«

Can you reasonably train a local model? Inference is one thing and can be run on a decent gaming graphics card but training requires a lot more compute.

You can absolutely train a nontrivial task model on a gaming gpu (image classifier, sound classifier, text model with a very structured input and output, etc). You can also post-train addon layers on top of existing open-weight models (LORA).