The worst thing is the majority of these aren’t like the AI your little brother uses when he logs into the chatgpt app. Like that contributes a little. But the real issue is that Tech companies have implimented LLMs literally everywhere in their workflows and data and services and this is taking up a ridiculous amount of computational power. These enterprise level implementations and speculative integrations (ie. adding AI somewhere that a corpo things will help but literally no one asked for it), is whats driving the biggest rise.