What's this about?
A Dwarkesh episode arguing the long-running trend of ever-cheaper compute could reverse, making the chips that power AI substantially more expensive — and what that would do to the "just scale it up" playbook.
What exactly is "the cost of compute"?
"Compute" is the raw processing power (mostly GPUs) used to train and run AI. The whole scaling story assumes it keeps getting cheaper per unit. If it gets 10x more expensive instead — due to demand, supply, or energy limits — the economics of frontier AI change sharply.
Why it matters
Nearly every AI roadmap assumes cheap, abundant compute; a serious case for the opposite is worth understanding.