OpenAI and Anthropic are each running at roughly $30B in annualized revenue, equal to 0.1% of US GDP apiece. Add cloud and adjacent services and AI has gone from near zero to 0.25-0.5% of US GDP in a few years. If both labs hit $100B by end of 2025 as some project, AI crosses 1% of US GDP run rate by end of 2026. The original piece raises a sharper question underneath that number: how much of the productivity gain will go unmeasured, the way the internet's impact was invisible in GDP data through the 2000s, and whether that mismeasurement will drive the wrong regulatory response.
Meta's aggressive researcher poaching forced every major lab to match compensation, effectively giving a few dozen to a few hundred top AI researchers a simultaneous, cross-company liquidity event. Gil calls it a distributed IPO: a specific slice of the field went post-economic all at once. The downstream effects, distraction, lifestyle drift, reordered priorities, mirror what happens after a traditional IPO. The piece is worth reading for the framing alone, and for what it implies about mission alignment inside labs when the people doing the work no longer need the money.
Memory supply from Hynix, Samsung, and Micron is the binding constraint on compute buildout for at least the next two years, which may create an artificial asymptote on model capability improvements and lock in the current LLM oligopoly until 2028 at the earliest. Separately, tokens are now a denominated currency: Cursor subsidizes inference as user acquisition, neoclouds are inference businesses wearing tool company clothes, and Allbirds just raised a convertible note to build a GPU farm. The piece covers all four of these dynamics and the questions they leave open are more useful than any of the answers.
[READ ORIGINAL →]