The "local LLM" idea doesn't scale. - If all LLM users would move to local LLMs, it would make the problem worse because your local compute is never as efficient as that of a specialised data centre. Both hardware utilisation and performance per Watt would be much worse. - If the local LLM was really as capable as its cloud variant, the cloud operator would be running the "local" version, and it would do so more energy-efficiently. #FrugalComputing (1/2)