Token doubling is AI's Moore's law; interactive clock puts AI at 1% of US GDP by 2028
robleclerc · x · 2026-09-02
The author argues token output doubling time is the Moore's law of AI, and built an interactive "AI GDP Clock" estimating when AI output becomes a real line item (1%) in US GDP.
Key inputs:
- Global token output: Google disclosed 3.2 quadrillion tokens/month at I/O 2026 (7× YoY); OpenAI's API runs 15B tokens/minute (8Q/yr); a disclosed floor of 110Q/yr is tracked by tokensperday.com.
- Doubling time: OpenRouter has doubled every 11 weeks for three years; Google implies 19 weeks; OpenAI 20; Fireworks' CEO expects 20–100× more tokens by end of 2027. Default uses the slowest (19 weeks).
- Tokens per worker: a human knowledge worker processes 10M tokens/yr; AI tokens are priced 10,000:1 lower to account for weaker reasoning per token, no real agency, repeated context re-reads, and jagged capabilities.
- US share: ByteDance's Doubao reportedly runs 180T tokens/day (half the tracked global total); a 40% US share is the middle estimate.
Under conservative defaults, AI crosses 1% of US GDP around June 2028; more aggressive assumptions bring it within 1–2 years.
More from AGI Musings
- Discussion on Stuxnet and cyberphysical system risk vectors — mattbeane · 2026-09-02
- Sarah Guo on AI Frontier, Home Robots, and Anti-Monopoly — saranormous · 2026-09-02
- Frontier models fail 96% of simple 2D mazes in Long-Horizon test — thebigbigbuddha · 2026-09-02
- Op-ed: AI capex forecasted to exceed France's budget by 2028 — alexeyguzey · 2026-09-02
- Sam Altman predicts the fastest pace of improvement for AI models ever — tkexpress11 · 2026-09-02
- Opinion: AI-generated spam floods social media, hurting both supply and demand — dotey · 2026-09-02