normative
Sunk-cost reuse of training clusters is inefficient
From first principles one would never build a 350,000-chip cluster with 2.5 gigawatts of power to serve inference demand; the existing training cluster is an efficient asset only in a sunk-cost sense, not as a first-principles design for inference.
normativepending
Speaker
ModestEvidence Quote
“In a sunk cost world you say, sure, of course if I'm forced to build a million chip supercluster in order to train a $50 billion model, I might as well sweat the asset when I'm done.”
Created: 11/18/2026, 6:42:24 AM
Your notes
Loading notes...