EVALUATING THE ENERGY EFFICIENCY FRONTIERS OF CLOUD DATA CENTERS: A SLACK-BASED WINDOW DATA ENVELOPMENT ANALYSIS (DEA) APPROACH

Cloud data centers underpin the global digital economy, yet the surge in generative-AI training, deep-learning inference, hyperscale virtualization and edge caching is pushing up their electricity demand, straining local grids and raising greenhousegas emissions.Judging how well these capital-intensive sites turn resources into useful computing therefore calls for a quantitative framework that handles several inputs at once, credits computational output, penalizes by-product emissions that cannot be separated from production, and follows facilities across time.Conventional radial Data Envelopment Analysis (DEA) falls short here: it shrinks every input by the same proportion, overlooks slack in individual resources and has no physically consistent way to treat pollutants.This chapter combines the Charnes-Cooper-Rhodes (CCR) and Banker-Charnes-Cooper (BCC) envelopment models with a non-radial Slack-Based Measure (SBM) applied in a moving-window design.The model is demonstrated on a synthetic, illustrative panel of ten facilities observed over four annual periods (forty facility-years), constructed for methodological demonstration rather than drawn from measured operator data.Each facility uses three inputs (total energy consumption, infrastructure capital expenditure and Power Usage Effectiveness), one desirable output (standardized computational throughput) and one undesirable output (carbon footprint).The analysis isolates input-specific excesses, quantifies the upward bias of radial scores, separates technical from scale efficiency, reports peer-intensity weights and derives catch-up and frontier-shift indices.In the illustrative panel, sites that combine direct-to-chip or immersion liquid cooling with fully renewable power contracts define the best-practice frontier, while legacy sites with downflow air conditioning waste roughly 24% to 35% of their energy input as non-radial slack.The chapter closes with a tiered matrix of engineering retrofits and regulatory options for greener digital infrastructure.