Why compute might get 10x more expensive

Dwarkesh is writing about what happens when the compute market hits a wall. The short version: GPU supply can't keep up with demand, and prices are going up. This isn't a temporary blip — it's structural.

The piece is worth reading because it's not hand-wringing. It lays out the actual mechanics: chip fabs are running hot, the leading GPU makers have limited capacity, and everyone is bidding against each other for the same inventory. The result is a market where compute costs are trending sharply higher, and the big labs are already feeling it.

For the people running side businesses and the folks building models on borrowed compute, this hits differently. A 10x move in compute costs isn't just a line item — it's a bet on whether your margins survive. The ones who locked in pricing or built lean get to keep their footing. The ones running on spot instances are about to feel the squeeze.

Why this matters for us: la gente que trabaja con modelos de IA, ya sea vendiendo API o corriendo sus propios servicios, va a tener que ajustar cuentas — el costo del compute se va a comer márgenes si no se prepara.