Your cost-per-query breakdown covers embedding, vector store, reranker, and generation tokens. What’s missing?

A complete cost analysis must account for externalities, specifically the infrastructure costs imposed on the sources you retrieve from. When your system performs a grounding fetch, it consumes the bandwidth and compute of the source site. These costs are often invisible on your team’s cloud bill, but they are real and can lead to rate-limiting or service degradation for the source.