Does AI Progress Have a Speed Limit?
Source: https://asteriskmag.com/issues/08/does-ai-progress-have-a-speed-limit Author: asteriskmag.com Date: 2025-07-01
Summary
Analysis of whether fundamental limits exist on the pace of AI progress. Considers compute limits, data limits, energy limits, and algorithmic limits. Concludes: no near-term hard wall, but multiple compressing soft constraints.
Key Claims
- No hard wall: there’s no known physical or theoretical limit that would halt AI progress in the near term.
- Soft constraints: (1) chip supply (geopolitical and manufacturing), (2) training data quality, (3) energy and cooling, (4) talent concentration at top labs.
- The “speed limit” is organizational: bureaucracy, safety review processes, and coordination costs between researchers may be the actual bottleneck at frontier labs.
- Historical comparison: the pace of progress resembles early transistor development — the limits that seem hard usually yield to focused engineering effort.
Concepts
- Scaling & Compute — constraints on scaling trajectory