Carbon-Aware Inference Under Deployment Constraints: Extending the CCI Framework

The crisis

• Current energy benchmarking frameworks do not characterize inference costs at the infrastructure edge. Field offices, disconnected facilities, and cost-constrained environments deploy AI systems without per-query carbon visibility.
• Carbon-aware model selection requires measurement frameworks that extend beyond cloud API contexts. No such framework currently exists for constrained deployment environments.
• Organizations with infrastructure constraints make model deployment decisions without per-query energy cost data.

About this research

The CCI framework established per-query carbon cost measurement for cloud-based LLM inference. Constrained deployment environments present a distinct energy profile not characterized by the original framework. This thread extends the CCI methodology to address that measurement gap.

Key findings

• (In Progress)

References

• CCI (Energies · MDPI) · DOI: 10.3390/en19030642
• FAS · Measuring and Standardizing AI's Energy Footprint (2024)
• IEEE · Carbon-aware inference under deployment constraints (2025)
• ScienceDirect · Future Generation Computer Systems · Edge AI carbon cost (2026)