I could see the case for Western AI (Claude, GPT, Mistral, etc.), especially since the datasets they're trained on are based around Common Crawl... a disgusting dataset designed with unethical garbage in mind.

DeepSeek takes a different approach with their reasoning model, R1, as it's dataset is not only open, but the local model is FLOSS (MIT license). Sure, it may have some restrictions due to Chinese laws, but otherwise, do way better than Western models.