Sustainable AI governance remains difficult to audit because governance frameworks specify commitments more clearly than evidence thresholds, and sustainability measurement is rarely linked to the operational artifacts that govern AI systems in practice. This paper presents Data Governance for Sustainable AI (DG-SA) as a measurement-oriented capability method rather than a competing normative framework. The method combines four elements: a 12-capability evidence-oriented codebook, a 54-indicator reference assessment instrument, denominator-sensitive benchmarkability ratios, and a marketplace-microdata operationalization that joins external data products to cloud telemetry and disclosure evidence. Empirically, fifteen governance sources are coded against DG-SA using conservative evidence rules. Auditability and human oversight are strongly represented, whereas sustainability-energy and lineage capabilities remain sparse. External benchmarkability is 24.1% across the full instrument, 27.9% across substantive governance items, and 16.7% across thirty core capability indicators; under a direct-only rule the substantive ratio falls to 11.6%. To demonstrate operational usefulness, the study inventories relevant marketplace-distributed datasets across BigQuery sharing, AWS Data Exchange, Snowflake Marketplace, and Databricks Marketplace, and executes a deployment-routing experiment using 44 Google Cloud regions backed by Electricity Maps data. For a fixed 100 kWh workload in 2024, location-based emissions range from 0.273 to 67.876 kg CO2e across available regions, and routing the same batch workload from the median region to the cleanest decile would reduce emissions by 95.1%. The contribution is methodological and integrative rather than universal: DG-SA does not replace principles or management-system standards, but offers a reproducible way to measure, benchmark, and substantiate sustainable AI governance.
Dockara et al. (Mon,) studied this question.