The Portability Congestion Paradox: An Economic Model of AI Resilience and Scarce Recovery Capacity

Abstract

Technical portability is increasingly proposed as a response to dependence on concentrated artificial intelligence infrastructure. Yet a workload that can switch providers may still lack somewhere to run when many firms switch simultaneously. This paper develops a welfare model separating private portability preparation from shared recovery capacity. Firms rationally prepare transferable workloads but take the probability of obtaining backup service as given. Under proportional rationing, private investment follows the average recovery return, whereas efficient investment follows the marginal increase in total restored output. The resulting congestion externality produces excessive preparation whenever backup capacity binds. A portability subsidy can then raise resource costs without restoring additional output. The paper derives the exact welfare loss, distinguishes subsidies from genuine reductions in engineering costs, and shows how capacity reservations can implement efficient joint investment. Extensions cover a finite number of firms, uncertain capacity, ordinary switching benefits, and heterogeneous recovery values. Reproducible numerical illustrations demonstrate that expanding backup supply without correcting allocation incentives can also reduce net welfare. The contribution is a conditional economic theory of recovery competition, rather than a general argument against interoperability. Effective resilience policy must evaluate the compatibility of workloads, the capacity surviving the relevant disruption, and enforceable claims on that capacity together.