Research work

THE ECONOMICS OF GPU DEPRECIATION: CAPITAL BUDGETING UNDER ACCELERATED OBSOLESCENCE OF AI COMPUTING INFRASTRUCTURE

Shchur L.D.

CyberLeninK (CyberLeninka)

Abstract

The rapid turnover of graphics processing unit (GPU) architectures has created a mismatch between accounting depreciation schedules and the economic life of artificial intelligence (AI) computing assets.This study develops a capital-budgeting framework for GPU infrastructure under accelerated technological obsolescence and examines when sovereign cloud (SC) deployment can be economically justified in regulated industries.The analysis integrates net present value (NPV), utilization, residual-value uncertainty, architecture-driven value decay, staged procurement, switching costs, vendor concentration, compliance expenditure, and geopolitical risk.The results show that fixed useful-life assumptions can overstate the continuation value of GPU fleets when performance per watt and workload economics change faster than accounting schedules.Investment decisions are more robust when economic life is modeled by workload class and procurement is treated as a sequence of partially reversible commitments.SC is economically defensible for critical workloads when its incremental present cost is lower than the probability-weighted value of avoided regulatory, concentration, switching, and geopolitical losses.The proposed framework can support investment committees, infrastructure planning, and technology-risk governance.

Download PDF