AI inference costs

Tag "AI inference costs" — all articles

062026-11Dwarkesh Patel: Smarter AI models could drive compute prices up 10x or moreTech writer Dwarkesh Patel published a series of posts between July and August 2026 arguing that more capable AI models could push compute prices up more than tenfold. His central calculation: if a human-level software engineer could run on an H100-equivalent chip, that chip should rent for over $250,000 a year at current software engineer market rates — 15x today's spot price. The posts tie compute pricing to the value of the human labor AI replaces, drawing attention to the cost structure of AI inference.Dwarkesh Patel