 A Structural Model of Sovereign Accelerator Demand Under Classified OpacityAurelie Ecker-Fils ()
Working Papers from HAL Abstract: This paper reconstructs and resolves the widely circulated claim of a "global accelerator discrepancy," which alleged that Nvidia's reported Blackwell GPU shipments were inconsistent with revenue, power-capacity additions, and observable deployments. We show that the appearance of a discrepancy arises from three structural misunderstandings: (1) misinterpretation of die counts as complete GPUs, (2) incomplete visibility into sovereign and quasi-sovereign procurement channels, and (3) unrealistic assumptions about deployment timing and power-capacity allocation. After correcting for die-count semantics and incorporating plausible ranges of opaque sovereign demand, the purported multi-million-unit gap collapses. Appendix D provides a formal reconstruction of the discrepancy claim and demonstrates that once structural factors are accounted for, no residual anomaly remains. The analysis clarifies the limits of open-source inference in accelerator markets and provides a framework for distinguishing genuine supply anomalies from artifacts of incomplete visibility. Keywords: cluster architectures; national security modernization; accelerator discrepancy; lifecycle dynamics; compute infrastructure; capability classes; classified procurement; structural modeling; accelerator demand; sovereign AI (search for similar items in EconPapers) There are no downloads for this item, see the EconPapers FAQ for hints about obtaining it. Related works: Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text Persistent link: https://EconPapers.repec.org/RePEc:hal:wpaper:hal-05427184 Access Statistics for this paper More papers in Working Papers from HAL |
Is your work missing from RePEc? Here is how to contribute.
Questions or problems? Check the EconPapers FAQ or send mail to .
EconPapers is hosted by the
School of Business at Örebro University.