The research
Overview
The manuscript proposes an adoption-impairment paradox for AI infrastructure: technical progress can expose general-purpose GPUs to substitution by custom ASICs, while slow adoption can leave the same hardware under-utilised. It examines diffusion, competition, deployment constraints and two channels of impairment. Two-, five-, ten- and twenty-year adoption horizons are used as stress-test scenarios rather than forecasts. An associated supplement contains extended discussion, dated update notes and source-labelling details. Oracle provides a shared case with the companion finance and geopolitics manuscripts.
Inside the research
- 01
Proposes a framework linking algorithmic uncertainty, GPU flexibility and substitution by application-specific hardware.
- 02
Uses adoption horizons as conditional stress tests, with separate time-based depreciation and utilisation or obsolescence mechanisms.
- 03
Examines how competition, grid queues, permits and deployment timing interact with the useful life of AI infrastructure.