Exam-style
A compliance team holds 400 million embeddings covering a decade of archived case files. They run perhaps thirty searches a month, and a result that takes several seconds is fine. What is the cheapest fit?
Reveal the answer
A. S3 Vectors, storing the embeddings in S3 and querying them through its vector API
S3 Vectors is built for exactly this shape: enormous corpora, rare queries, and a latency budget measured in seconds rather than milliseconds. It trades query speed for S3-priced storage, which is the right trade when the index sits idle almost all month. OpenSearch Serverless still bills its two-OCU minimum between searches, an HNSW build over 400 million rows is punishing, and holding the index in memory pays the highest price per gigabyte for latency nobody needs.
Q. You have a huge, rarely-queried vector archive where seconds of latency is acceptable. Cheapest fit?
A. S3 Vectors stores vectors directly in S3 with a vector API on top, built for massive-scale, latency-tolerant retrieval. It is the wrong choice for an interactive sub-50ms assistant.
Why? Match the store to the latency budget; S3 Vectors trades speed for cost at archival scale.