↳ SERPENS INC. // TECHNICAL RESEARCH SPECIFICATION // ID 414150400

Compute Availability Factor (CAF): A Framework for Measuring the Physical Deliverability of Artificial Intelligence Compute

From nominal accelerator capacity to effective and infrastructure-available compute.

Author: Richie Adomako · Serpens Inc. Systems Architecture
Publication: ResearchGate Publication ID 414150400 · September 2026
Classification & Status: Open Research Framework / Calibration Phase (Technical Specification v1.0)

Abstract

Frontier artificial intelligence scaling models hypothesize that model capabilities scale predictably with compute. However, public benchmarks evaluate nominal compute (installed accelerator counts) without accounting for the physical substrate required to energize, cool, interconnect, and schedule large-scale clusters. This paper introduces the Compute Availability Factor (CAF), a multiplicative physical availability index decomposing infrastructure delivery into Electrical (E), Thermal (T), Runtime Network (N), and Operational (O) availability: CAF = E(t) × T(t) × N(t) × O(t). We define the Deliverability Coverage Ratio (DCR) to quantify whether announced roadmaps are physically consistent with verified infrastructure or exhibit an apparent capacity shortfall.

Access Publication & Citation

Read Full Paper on ResearchGate ↗ Foundational Thesis on Medium ↗

Suggested Citation (APA)

Adomako, R. (2026). Compute Availability Factor (CAF): A Framework for Measuring the Physical Deliverability of Artificial Intelligence Compute: From nominal accelerator capacity to effective and infrastructure-available compute. Serpens Inc. Technical Specification v1.0. ResearchGate. https://www.researchgate.net/publication/414150400

BibTeX

@techreport{adomako2026caf,
  author      = {Richie Adomako},
  title       = {Compute Availability Factor (CAF): A Framework for Measuring the Physical Deliverability of Artificial Intelligence Compute: From nominal accelerator capacity to effective and infrastructure-available compute},
  institution = {Serpens Inc. Systems Architecture},
  year        = {2026},
  month       = {September},
  type        = {Technical Specification v1.0 (Preprint)},
  note        = {ResearchGate Publication ID 414150400},
  url         = {https://www.researchgate.net/publication/414150400}
}