When you talk to corporate America, everyone takes inference compute for granted. We’re all used to surplus. You go to the cloud, buy whatever you want, and it's always right there waiting at a fair price. That will probably never be true again. Two to three years from now, leaders are going to realize they can automate huge fractions of their business to promote growth, release new products, or even rebuild their full tech stack using @blitzyai. But, there won’t be any excess compute for opportunistic projects. I had a long conversation with @bavaria_kush from @OrnnExchange the other day. The smart executives are already using it to reserve future compute. Compute is now finite commodity like oil or gold. The main difference is the sheer rate increase of token consumption, while the supply of new inference tokens is highly predictable and severely lagging. I believe this is the new normal forever hereafter.
