RAG Pipeline Costs: Embedding Is Cheap, LLM Calls Are Not

A developer breaks down the real costs of a RAG pipeline, finding that embedding costs only $0.18 per 1,000 pages, while LLM inference and vector store infrastructure dominate expenses. He advocates for dedup checks and incremental processing to avoid unnecessary re-embedding and chunking. The key insight: optimizing for the wrong bottleneck can waste resources.