AI Industrial Takeoff — Part 1: Maximum growth rates with current technology
Using US input-output tables, growth could hit 100% annually with full automation.
A detailed analysis posted on LessWrong by djbinder (linkpost to defensesindepth.bio) examines how fast an AI-driven economy could grow if human labor were fully automated. The author uses US government input-output tables to map the physical supply chains needed to reproduce capital—machines, factories, infrastructure. The core finding: the maximum sustainable growth rate, assuming no technological improvements beyond automation, allows the economy to double in roughly one year. This matches earlier estimates from Hanson (2001), Trammell and Korinek (2023), and Epoch AI (2025). The calculation holds even after accounting for resource depletion and construction lags.
The analysis emphasizes that the present-day US economy is far from the optimal composition for rapid self-replication—too many services and consumer goods relative to mining and heavy manufacturing. Transitioning to that composition would take several years, which is the subject of Part 2. The author notes that if AI also enables rapid technological improvement on top of automation, this one-year doubling time becomes a conservative lower bound. The standard economic prediction of a few percentage points of growth is implicitly a prediction that full automation will not happen. The piece was developed with input from Holden Karnofsky and used Claude Opus 4.6 and 4.7 for assistance.
- US input-output data shows a fully automated economy could double in about one year using current production methods.
- Even with moderate savings rates and resource constraints, doubling times remain below two years.
- The analysis is a conservative lower bound; technological improvements on top of automation would allow even faster growth.
Why It Matters
Challenges mainstream economic forecasts and suggests AI could trigger explosive, order-of-magnitude faster growth.