Comment by eggbrain

Performance optimizations don't just help the western labs, they also help with running more powerful/useful LLMs locally.

If local LLMs get "good" enough, people will soon paying for subscriptions to ChatGPT and Claude, which hurts their revenue.