Hypotheses

LLMs are nearing the limits of what they can do; the 2025-style innovation leap is not repeating.

Software loses value at the vendor that cannot re-derive customer-specific context cheaply; value expands where context is held by incumbents who can monetise it through AI-substitution layers above it; cloud-AI infrastructure moats remain conditional on two cash-burning private labs honouring contracts that account for half of the hyperscaler backlog.