Currently AI providers are hyper competitive and almost all of the world has access to it quite cheaply. If people had to buy hardware for local inference, it would only be a tiny percentage of the world's affluent tech enthusiasts with access to AI, wouldn't it?

We're currently managing the shortage with a hybrid model, so enthusiasts can still buy (expensive) individual systems and the broader shared hardware goes to many, many providers. OpenAI is something like 15% of data center power in the US I think? That's a lot but we still have choice.