AI Infrastructure at Scale: Cloud Systems, GPU Computing, and Benchmark-Driven Optimization

View at publisher Summarise with VIRAT
Authors: Murali Krishna Pasupuleti
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract: This manuscript develops a research architecture for understanding AI infrastructure at scale through the linked problems of cloud systems, GPU computing, and benchmark-driven optimization. It argues that infrastructure should be studied not as a narrow engineering substrate but as a layered socio-technical system whose performance depends on uncertainty modeling, statistical explanation, predictive learning, reproducible data lifecycles, and governance design. Across five chapters, the book moves from conceptual foundations and research design to causal and econometric inference, machine-learning based prediction and trustworthy control, scalable analytics pipelines and MLOps, and finally to applied sectoral and policy horizons. The analysis treats benchmarks as epistemic instruments that encode assumptions about relevance, risk, and legitimacy, rather than as neutral scoreboards. It further proposes that meaningful optimization requires multi-criteria evaluation across throughput, latency, energy, carbon sensitivity, resilience, and auditability. The manuscript is oriented toward publisher-ready academic use and yields concrete research outputs, including benchmark specifications, causal evidence protocols, lifecycle controls, GPU allocation frameworks, climate-aware siting templates, and governance artifacts for secure and accountable deployment. Its comparative lens emphasizes applicability across South Asia, Europe, Africa, and the Americas, showing how local institutional conditions shape the value of common methods. The result is a coherent framework for researchers, practitioners, and policymakers seeking to evaluate and govern AI infrastructure under conditions of scale, heterogeneity, and strategic importance. Keywords AI infrastructure, cloud systems, GPU computing, benchmark-driven optimization, workload forecasting, causal inference, queueing models, reinforcement learning, MLOps, telemetry pipelines, data-centre governance, edge intelligence, thermal scheduling, reproducibility, sovereign compute, post-quantum security

Fields of Study & Subjects

Cloud Computing and Resource Management Energy Efficiency in Computing Software System Performance and Reliability

Publication details

Publication date
28 February 2026
Language
EN