Amazon Just Showed How to Make AI Search 90% Cheaper
The quiet plumbing behind AI chatbots decides whether answers are fast — or cheap.
Amazon Web Services published a guide to help developers pick where an AI's "memory" lives. When a company builds an AI assistant that answers questions about internal documents, the AI first looks up the most relevant passages before writing a reply. That lookup needs a storage system — a vector store, which is basically a smart filing cabinet that finds text by meaning instead of matching exact keywords. AWS now supports three different cabinets behind its Bedrock Knowledge Bases service, and the choice quietly changes both speed and cost.
Why should you care? Because this storage layer sets the price and the patience level of every AI assistant built on it. One option keeps everything in fast memory, so answers arrive almost instantly — but you pay more each month. Another stores vectors in cheap cloud object storage, cutting storage costs by up to 90% compared with traditional vector databases while still answering in under a second. Spread that across millions of documents, and the difference shows up as real money on a budget sheet.
The three choices are: Amazon OpenSearch Service, for speed and combined keyword-plus-meaning search; Amazon Aurora PostgreSQL with pgvector, for teams already running that database who want everything in one place; and Amazon S3 Vectors, the new low-cost option built on the same storage that holds your photos and backups. AWS walked through sample situations, such as a product catalogue, to show which fits where.
The catch: there is no single best answer, and none of this is something you can buy and use yourself. Fast usually costs more; cheap can mean fewer search tricks or slightly slower replies. The advice also assumes your company already runs on Amazon's cloud, so rival setups are left out. Still, the direction is clear — the cost of giving AI a memory is falling fast, and that eventually reaches the rest of us.
- AI assistants that answer questions about your documents need a 'lookup' storage system, and Amazon just compared the three it supports.
- One new option, S3 Vectors, claims up to 90% cheaper storage than traditional vector databases, with answers still arriving in under a second.
- There is no universal winner: the fast option costs more, the cheap option does less, and everything assumes you already use Amazon's cloud.
Why It Matters
Cheaper AI search could mean lower bills for every company building a chatbot — savings that may reach customers.