For some companies there might be a need to run them locally. For instance, Apple decided to run LLMs on the phone locally. I guess it depends on how important latency and privacy are. Perhaps Meta is looking at how much interest for those local models is there.