Green-LLM: Optimal Workload Allocation for Environmentally-Aware Distributed Inference

Green-LLM addresses the intersection of LLM inference demand and data center sustainability. The core defensibility is low (3) because it currently exists as a research paper with a reference implementation; while the mathematical approach (lexicographic optimization) is sound and removes the need for manual weight tuning, it lacks a software moat, user base, or proprietary data. With 0 stars and 2 forks, it has no current community traction. The primary threat comes from hyperscalers like AWS (Customer Carbon Footprint Tool), Google (Carbon-Aware Computing), and Microsoft, who already operate sophisticated carbon-aware schedulers. Competitively, it sits adjacent to projects like SkyPilot (inter-cloud orchestration) or KubeRay. The displacement horizon is 1-2 years as these optimization techniques are likely to be absorbed into standard Kubernetes-based inference orchestrators. Its main opportunity lies in being integrated into a middleware layer for decentralized or edge-based AI providers (e.g., Akash Network or Together AI) who need to manage heterogeneous nodes with varying energy profiles.