Future-Obligation-Aware Workload Admission for Coupled Compute-Thermal Infrastructure: A Resource-Liability and Downstream-Closure Control Framework for High-Density AI Computing
This technical disclosure presents a research hypothesis for future-obligation-aware workload admission and supervisory control in high-density AI compute infrastructure. The proposed framework treats accepted compute workloads as creating physical resource obligations that may persist after workload execution, particularly unresolved thermal obligations associated with heat storage, rejection, or useful recovery. The framework introduces an unresolved thermal obligation state and evaluates candidate workloads by propagating their predicted electrical, thermal, hydraulic, water, storage, heat-rejection, heat-recovery, reliability, and service-level consequences over a future control horizon. A workload may be admitted, deferred, limited, relocated, or rejected depending on whether the resulting obligations can be closed through feasible downstream pathways without violating system constraints. Thermal storage is treated as movement of a thermal obligation through time rather than elimination of that obligation. The proposed controller therefore retains stored heat in its accounting until useful recovery or heat rejection is measured or otherwise verified. This document is a frozen Version 1.0 technical disclosure intended to establish a dated research record and define a mechanism for subsequent modeling, falsification, prior-art analysis, and independent validation. The framework is an unvalidated research hypothesis and does not claim demonstrated performance, engineering readiness, patentability, or freedom to operate. Author and research lead: Aleeya Kim, Human Layer AI. ORCID: 0009-0009-7676-3761. DOI: 10.5281/zenodo.22818983. Generative AI systems, including ChatGPT, Claude, and Grok, were used as research, modeling, drafting, critique, and adversarial-analysis tools; their outputs are not treated as independent validation, authorship, or evidence of engineering correctness.
Authors
- Aleeya Kim (ORCID: https://orcid.org/0009-0009-7676-3761)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-09-17
- DOI
- https://doi.org/10.5281/zenodo.22818982
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- article
- Field-Weighted Citation Impact
- 0.00