Preprint2026

The Physical Limits of Compute and Financial Risk Integration Across Seven Infrastructure Layers

Fuli Yang; Yu Liu; Shiyu Lin

AI & Industrial Cycles →

Abstract

Artificial intelligence links seven infrastructure layers into a persistent, uneven equity-risk system. Using 33 firms, a fixed 28-stock comparison and strictly lagged factor adjustment through August 2026, return connectedness rises from 6.34% to 20.06% and downside connectedness from 5.27% to 21.02%. Both increases survive a six-test correction; the residual diversification ratio falls by 16.3%. A traded-factor extension through 23 October retains higher connectivity, with corrected evidence strongest for downside risk. Grid equipment dominates outward variance attribution; nuclear supply and gas remain peripheral. Five supplier–buyer power-agreement announcements and fourteen eligible tariff decisions establish observable contractual links without a uniform increase in residual dependence. Power agreements validate investment while reallocating price, volume, completion and exit exposures. Financial integration therefore spans industry boundaries, while physical links and equity covariance measure different objects. Risk management must assess the infrastructure system and its contractual risk bearers jointly.

Author abstract from the research manuscript. Typography and line breaks have been normalized; the linked source may show a different manuscript version.

Keywords

Cite this paper

Yang, F.; Liu, Y.; Lin, S. (2026). The Physical Limits of Compute and Financial Risk Integration Across Seven Infrastructure Layers. Zenodo. https://doi.org/10.5281/zenodo.23644877

Version & publication history

  1. Original repository record

    Existing Zenodo record and DOI. Version labels follow the source; no additional version number is inferred.

    View source record ↗
  2. Added to GARCH Research Papers

    Bibliographic record added. This is not a new manuscript release or a new DOI.

Access & rights

Restricted full text at source. Manuscript files remain with the original platform. This catalogue does not redistribute source PDFs.

All rights reserved for manuscript files; see source terms.