AI has kinda already gone through that phase, with quantisation you can effectively double your ram with almost no negative effects by going from fp16 to int8.