You got that wrong. The forcing function of compute scarcity is an advantage not a detriment. The amount of investment pulverized in performative model training and dead ends (Hi Sora) should make this obvious.