(Generative AI for Network Sustainability)
McGann, O. – Giordano A. – Haider, B. – Paesani, M. | 2025-11
This report presents the state of the art on the energy impact of network and cloud infrastructures, with an integrated focus on consumption, emissions, and measurement, modeling, and prediction methods applicable along the entire network–data center–platform chain. In a context in which the demand for digital services is growing rapidly and energy is becoming a design constraint, the most recent legislation highlights two key points: (i) the increase in traffic and capacity does not automatically translate into proportional growth in consumption due to efficiency advances, but (ii) the adoption of AI workloads and increased compute density are reopening a critical front on available power, cooling, and carbon footprint, requiring more robust and service-attributable operational measures. The review integrates scientific contributions, technical standards, and industrial guidance to build a coherent picture of metrics and methods: from site KPIs for data centers to metrics for equipment and networks. The evidence that emerges is that efficiency cannot be evaluated only at the level of structure or datasheet: non-energy proportionality, high consumption in IDLE, and multi-vendor heterogeneity mandate a telemetry chain that combines physical and logical measurements, enables comparable benchmarking, and supports attribution per workload, slice, or service tag. On a methodological level, the report summarizes recent surveys and taxonomies on the monitoring and prediction of energy impact in networks and data centers, highlighting the evolution from predominantly qualitative and scenario-based approaches toward data-driven and capacity-predictive integrated softwarized architectures (SDN/NFV/Network Slicing). Experimental evidence is also recalled that demonstrates significant energy savings through dynamic light path management in elastic optical networks and SDN-based strategies focused on flow consolidation and minimization of state transitions, with parameters for consumption and operational requirements. Finally, the state of the art indicates that energy-aware and carbon-aware automation, backed by fine-grained telemetry and interoperable instruments, is the most promising lever for reducing over-provisioning and emissions while maintaining SLA, auditability, and governance. The report translates this evidence into replicable criteria for the project, proposing a “measurement-first” approach and a set of KPIs and operational guardrails to evaluate network solutions and cloud sustainability-oriented strategies.
1. Introduction 1.1 Industrial context and motivation 1.2 Report objective and scope of the deliverable 1.3 Operational Research Questions 2. Methods 2.1 Research strategy and sources 2.2 State-of-the-art selection criteria 2.3 Replicable experimental design 3. Results 3.1 Macro evidence: demand growth and power constraints 3.2 Challenges and objectives for energy-aware networks 3.3 Metrics and standards for data centers and clouds 3.4 Metrics and standards for equipment and networks 3.5 Technical drivers of consumption: why over-provisioning weighs more than expected 3.6 Virtualization, Network Function Virtualization and Network Slicing 3.7 Telemetry and energy accounting: from site measurement to service measurement 3.8 AI for dynamic allocation and emission reduction: energy-aware and carbon-aware 3.9 Industrial trends: AI factories, data sovereignty and sustainability as a constraint 3.10 Surveys and related reviews on monitoring and prediction of the energy impact of network components 3.11 Insights: monitoring, optimization, power models and energy-aware management 3.11.1 Network Monitoring 3.11.2 Optimizing energy consumption 3.11.3 Power Model 3.11.4 Energy-aware management 4. Discussion 4.1 Critical summary: measuring, attributing, controlling 4.2 High-impact and low-strategic risk levers 4.3 Full-Project Implications: Integration with Agentic Automation 4.4 Industrial replicability and licensing: avoiding lock-in and maximizing transferability 4.5 Lessons learned and operational recommendations 4.6 Integrated summary of evidence and operational implications 4.7 Research gaps and opportunities for GAINS 4.8 Design choices Bibliography
Energy Impact, fine-grained telemetry, Service attribution, Sustainability KPI, Energy-aware automation, Carbon footprint, SDN/NFV/Network Slicing, Over-provisioning
Reference: GAINS deliverables are linked to Work Packages and milestones. For additional materials, visit Resources.