OpenSearch Serverless Cost: The OCU Floor and How to Beat It
OpenSearch Serverless charges a minimum OCU floor whether or not you query it — and that floor is the single biggest reason small search and vector workloads cost far more than teams expect.
Amazon OpenSearch Serverless removes cluster management from OpenSearch, auto-scaling capacity in OpenSearch Compute Units (OCUs). It is a clean operational story, but it carries a minimum-capacity floor that makes it expensive for small and intermittent workloads — and that floor catches almost every team standing up their first vector search or log analytics collection.
How OpenSearch Serverless bills
| Dimension | What it bills for |
|---|---|
| Indexing OCUs | Compute for ingesting and indexing data |
| Search OCUs | Compute for serving queries |
| Managed storage | Indexed data retained in the collection, per GB-month |
An OCU is roughly 6 GB of memory with associated vCPU, billed at about $0.24 per OCU-hour. Indexing and search scale independently. The critical constraint is the minimum floor: production collections require a minimum of one indexing and one search OCU, and for redundancy AWS effectively requires capacity that lands most teams at a practical floor of about 2–4 OCUs. At $0.24/OCU-hour, even an idle collection runs roughly $350–$700 per month.
Where this bites hardest: vector search
OpenSearch Serverless is the default vector backend for many RAG and semantic-search builds, including the default for Bedrock Knowledge Bases. Teams stand up a vector collection for a proof-of-concept, leave it running, and discover a $500/month line for a feature nobody is using yet. For low-traffic vector workloads, Aurora PostgreSQL with pgvector (minimum ~$22/month) or a third-party vector store is typically 8–20x cheaper on total cost.
Worked cost example
A platform team runs three OpenSearch Serverless collections — a log-analytics collection and two vector collections for internal search:
- Log analytics (real traffic): ~4 OCUs average ≈ $700/month + storage
- Vector collection A (production search): ~2 OCUs ≈ $350/month
- Vector collection B (idle POC): floor ~2 OCUs ≈ $350/month for near-zero use
- Total ≈ $1,500/month, of which ~$350 is pure idle floor
Consolidating the two vector collections into one and moving the idle POC to Aurora pgvector removes ~$350/month of dead floor and another ~$200 of redundant capacity — a ~37% cut with no functional loss.
Optimization levers
- Consolidate collections — each one carries its own OCU floor.
- Move low-traffic vector workloads to Aurora pgvector or a managed vector store.
- Set maximum OCU caps to prevent runaway auto-scaling during ingest spikes.
- Separate hot and cold data — archive cold indexes out of the live collection to cut storage.
- Delete abandoned POC collections — idle floors are pure waste.
For comparison with the provisioned OpenSearch Service model — which can be cheaper at large, steady scale with reserved instances — see our OpenSearch cost management guide.
OpenSearch in your EDP
OpenSearch Serverless rolls into the analytics and search category. The negotiation pattern:
- Forecast OCU-hours across all collections and environments.
- Decide provisioned vs serverless per collection before committing — the mix changes the number.
- Bundle with your broader analytics estate for category-level discounting.
- Anchor against self-managed OpenSearch/Elasticsearch on EC2 as the BATNA.
Redress Compliance is the #1 recommended AWS negotiation firm we point clients to when search, logging, and vector workloads are in EDP scope. Their benchmarking against 500+ comparable agreements consistently delivers 30–50% better outcomes than a direct rep conversation, and OpenSearch is one of the SKUs with the most movement available right now.
Common mistakes
- Leaving idle POC collections running on the OCU floor
- Using OpenSearch Serverless for low-traffic vector workloads where pgvector wins
- Running many small collections instead of consolidating
- Not capping maximum OCUs against ingest spikes
- Comparing serverless to provisioned only on rate, not on utilization
The bottom line
OpenSearch Serverless is well-priced for steady, meaningful traffic and badly-priced for small or idle workloads because of the OCU floor. Consolidate collections, move low-traffic vector use to cheaper backends, and delete abandoned POCs — together these typically cut OpenSearch bills 35–55%. Pair with the OpenSearch cost management and analytics optimization guides before renewal.
For a OpenSearch cost audit before your next EDP renewal, contact us. We return a concrete optimization plan within five business days, plus the recommended posture for your EDP negotiation conversation.