LazyGraphRAG is a novel graph-enhanced retrieval-augmented generation (RAG) model developed by Microsoft Research. It eliminates the need for pre-summarizing source data, thereby avoiding potentially prohibitive indexing costs for some users and use cases. LazyGraphRAG offers intrinsic scalability in terms of cost and quality, significantly improving the efficiency of answer generation by delaying the use of large language models (LLM). The model exhibits excellent performance for both local and global queries, with query costs far lower than traditional GraphRAG approaches. The introduction of LazyGraphRAG presents a new solution for AI systems tackling complex issues in private datasets, holding significant commercial and technological value.