Imagine, hypothetically, inventing an AI architecture that delivers top-tier AI model performance using just 1% of current compute needs. The demand for AI would remain strong – possibly even increasing due to reduced costs (the famous “Jevons Paradox”) – but not enough to utilize fully the vast GPU parks built at great expense. Suddenly, 99% of that infrastructure would become redundant overnight, despite continued high demand for AI itself. You simply can’t scale demand fast enough to justify such surplus infrastructure. You cannot “jevons” your way out of this.
And this is where the condition for a bubble – not enough demand for the overcapacity of compute – would be met.
Source : Overcapacity + Weak Demand = Bubble. AI is not yet a Bubble.
Laisser un commentaire