.@dylan522p predicts AI labs will deliberately sacrifice billions in inference revenue to spend most of their compute building AGI: "This is what I personally believe, and it's very non-consensus. The labs are going to allocate less and less compute to inference over time. The standard belief is that most compute will go to inference. I think most of it will go to training, not revenue-generating inference" "If they're generating $60-70 million per megawatt, do you still allocate 40% to inference, generate all this profit, and do dividends and share buybacks? Or do you go build AGI? The obvious answer from Anthropic and OpenAI, not just at the executive level but their board, is to go build AGI, because it's way more profitable" "The only reason to have inference compute be so large is so you can grow your training fleet"