Balcony Compute and Generative Entry: Companion Working Papers on Distributed AI, Market Structure and Infrastructure Correction

Description

Abstract

This record contains two companion working papers examining how advances in artificial intelligence may weaken the market value of concentrated AI infrastructure without any corresponding failure of AI capability or aggregate use.

Balcony Compute: Temporal Mismatch and the Risk of an AI Infrastructure Correction develops the concept of balcony compute: user-sited, organisationally local or edge inference that performs part of a centrally supplied intelligence service while preserving optional access to frontier cloud systems. It connects distributed inference, fixed-cost feedback, technological obsolescence, contract transmission and infrastructure finance through an actor-indexed temporal condition: economically substitutive adoption becomes destabilising where Tadoption<Tadjustment,iT_{\text{adoption}} < T_{\text{adjustment},i}Tadoption<Tadjustment,i. The paper distinguishes a service-revenue pool from an asset-finance pool, with contracts operating as the membrane through which provider-level demand leakage may reach data-centre owners, lenders and residual-risk holders. It argues that a correction in AI-linked markets could occur while AI use continues to grow, becoming a funding event only where residual technological risk is held by leveraged actors unable to absorb or postpone recognition.

Generative Entry: Capability Replication and the Market-Structure Consequences of Advanced AI identifies a second route through which concentrated scarcity may weaken. It distinguishes service replication, capability replication and literal self-replication. The economically disruptive threshold may arrive below autonomous self-replication if advanced systems can design, train, evaluate or materially enable narrower systems that reproduce useful portions of their competence and can be operated by startups, enterprises or regional providers. The paper calls this process generative entry. Economically effective entry includes both supplier formation and market uptake and becomes destabilising for actor $i$ where Tsupplier formation+Tmarket uptake<Tadjustment,iT_{\text{supplier formation}} + T_{\text{market uptake}} < T_{\text{adjustment},i}Tsupplier formation+Tmarket uptake<Tadjustment,i. The paper also develops licensed offspring as the incumbent counterforce: capability and supplier numbers may proliferate while ownership, runtime control, data, licences and customer relationships remain centralised.

Together, the papers distinguish two connected mechanisms. Balcony compute moves execution towards users and organisations; generative entry multiplies potential suppliers. Both mechanisms are conditional, both may be captured by incumbents, and neither implies that artificial intelligence has failed. Their common concern is whether financial claims built around concentrated and durable AI scarcity can adjust quickly enough when the technology itself begins to distribute execution, competence and market entry.

These are mechanism papers prepared for adversarial review. They offer no date, probability, trading recommendation or prediction of company failure.

Notes

Notes

Notes