[lxgr]

LLM inference is much more geographically fungible than electricity, so maybe it’s just not worth the complexity yet and there is enough (not highly latency sensitive) load on average globally.