Ben Horowitz, co-founder of a16z, argues that the scenario of a rapid, self-improving AI intelligence explosion is currently unlikely. His assessment is based on a framework developed by Databricks CEO Ali Ghodsi, which outlines four specific conditions required for such a development.
These conditions include a significant reduction in resource requirements, shorter training cycles, and consistent, measurable improvements in model accuracy. According to Horowitz, none of these criteria have been met as of mid-October 2026.
Horowitz points out that current industry trends are moving in the opposite direction. Training frontier models is becoming increasingly complex and expensive, while infrastructure constraints, such as GPU shortages and power limitations, remain significant bottlenecks.
The a16z co-founder noted that he sees no evidence of immediate breakthroughs at major research labs that would alter these physical and resource-based limitations. He suggests that the sector's progress will continue to be defined by data center capacity, power contracts, and chip availability.
