Dwarkesh Patel's blog examines outcomes if AI scaling trends continue and a leading lab hits $1T annual revenue by next year end. It argues this would drive compute prices sharply higher, potentially 10x or more. An H100 running a human-level software engineer would justify over $250k yearly rental at market rates, or 15x today's spot price. Superior intelligence efficiency would let top model firms charge much higher margins while others face commoditization similar to basic producers. The piece frames compute scarcity as a key differentiator in this scenario.