Ranking Bottlenecks for AI Infra Investment
A framework for where the real constraints sit: compute, data, energy, or trust.
Every AI infrastructure pitch claims to solve the bottleneck. Most solve a bottleneck that has already moved. The useful question is not "what is scarce" but "what is scarce next, and for how long."
The four constraints
Compute, data, energy, and trust do not bind at the same time. Compute is the loudest and the most quickly arbitraged. Energy is slow, physical, and underpriced in most models. Data is where moats quietly form. Trust — the ability to deploy a system where a wrong answer has consequences — is the constraint almost nobody underwrites.
Ranking, not listing
A framework that lists constraints is a poster. A framework that ranks them, under a stated time horizon, is a thesis. Rank by how binding, how durable, and how cheaply the constraint can be relieved by capital rather than time.