AI data centre infrastructure is approaching physical and operational limits driven by power density, thermal constraints, and systemic fragility under failure conditions. Current hyperscale architectures optimise for performance under stable conditions but exhibit increasing vulnerability to cascading failures, energy volatility, and centralised dependency. This proposal investigates whether a distributed, modular data centre architecture incorporating structured disagreement, controlled node cycling, and layered continuity can improve system resilience and energy adaptability. Drawing on conceptual parallels with distributed systems operating under degraded conditions, and historical precedents in Australian telecommunications infrastructure, the study proposes a framework incorporating node-level autonomy, peer-informed evaluation, and energy-aware workload distribution. Building on prior work in multi-model verification and behavioural validation (Harrison, 2026a; Harrison, 2026b), this research aims to evaluate whether principles of structured disagreement can be extended from software-based AI systems into physical infrastructure domains. The proposed framework will be evaluated through simulation and comparative analysis against centralised architectures.
Kyle Harrison (Sat,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: