All cheat sheets
Databasescheat sheet

Neptune Serverless

Neptune Serverless swaps always-on graph-database instances for capacity that scales per NCU (~$0.12/NCU-hour) between a floor and ceiling you set — 40–60% cheaper on bursty or off-hours graph workloads, pricier on flat 24/7 ones.

Last reviewed: July 14, 2026

Prefer step by step? Take the guided Neptune Serverless walkthrough — free, anonymous, interactive.Have an AI assistant? Jump to the copy-paste audit prompt at the bottom of this sheet.

TL;DR: Graph databases exist for relationship queries (friend-of-friend, fraud rings, dependency maps); if your data isn't relationship-shaped, a relational or document store is cheaper. Original Neptune billed provisioned instances 24/7 idle or not. Neptune Serverless (2022) scales capacity per Neptune Capacity Unit (~2 GiB memory + compute) between a min and max you set, at ~$0.12/NCU-hour. Its edge is charging little during quiet stretches — so it wins on bursty/off-hours/dev workloads and loses on flat 24/7 load.

The numbers

  • Provisioned baseline: a db.r5.large is ~$0.348/hr → ~$254/mo regardless of usage.
  • Serverless floor: parked at the 2.5 NCU minimum 24/7 ≈ 2.5 × $0.12 × 730 = ~$219/mo — the minimum is real money.
  • Where it wins: 10 NCUs for 8 business hours + 2.5 NCUs otherwise = 3,600 NCU-hours ≈ $432/mo vs ~$508/mo provisioned (10 NCU ≈ db.r5.xlarge) — ~15% plus automatic scaling and zero capacity planning.
  • Field examples: real-time fraud detection (5–8 NCU normal, spiking to 40–50) saved 40–60% vs a peak-sized cluster; a nightly-analytics cluster went $1,015 → ~$621/mo (39%). A steady 24/7 recommender was more expensive on serverless ($562 vs ~$508 provisioned).

Do this

  1. Set conservative min/max NCUs (e.g. 2.5 → 40) and adjust after watching real ServerlessDatabaseCapacity in CloudWatch.
  2. Move variable workloads first — fraud detection, periodic batch, dev/test, spiky query mixes. Leave flat 24/7 production for last, after you know its NCU footprint.
  3. Bound your queries — multi-hop traversals ("all nodes within 5 hops") burn NCUs faster than expected; optimize before raising the cap.
  4. Set a hard maximum + billing alarm — a runaway traversal can otherwise scale you to the cap and produce a surprise month-end bill.
  5. Decide empirically — point a representative workload at a serverless cluster for a week and compare the NCU bill to your provisioned bill.

Gotchas

  • The minimum baseline (~$219/mo) is a floor, not free — tiny idle clusters still cost real money.
  • Less tuning control — no instance-type choice or manual memory tuning; if a specialist loves tuning, serverless hides that lever.
  • Brief warm-up after a deep scale-down — first queries after a long quiet period can see slightly higher latency.
  • Feature parity can lag — check current docs before assuming provisioned features exist on serverless.

Skip this if

  • The workload runs flat-out at high capacity 24/7 — a right-sized provisioned instance, especially with a Reserved Instance, undercuts serverless by 30–40%.
  • You want absolute lowest steady-state cost and can commit — see Reserved Instances on a provisioned cluster.
  • You're comparing the whole serverless-database family — OpenSearch Serverless, Aurora Serverless Sizing, and Redshift Serverless use the same "min baseline + scale to max, pay per unit-hour" model.

Run this audit with your AI assistant

Paste this into Claude, ChatGPT, or any agent that can run the AWS CLI with read-only credentials. It audits your account for exactly the waste this sheet describes — and changes nothing.

You are auditing an AWS account's Amazon Neptune usage for Serverless
savings. Use the AWS CLI with READ-ONLY credentials. Do not create,
modify, or delete anything — report findings and recommended (unapplied)
fixes only.

1. Inventory: aws neptune describe-db-clusters / describe-db-instances —
   capture provisioned vs serverless, instance classes, and any
   ServerlessV2ScalingConfiguration (min/max NCU) already set.
2. Utilization shape: pull CloudWatch AWS/Neptune CPUUtilization and (for
   serverless) ServerlessDatabaseCapacity over 30 days, hourly. Classify
   each cluster: flat 24/7-heavy, business-hours, periodic-batch, spiky,
   or dev/test.
3. Cost compare: provisioned instance $/hr × 730 vs projected NCU-hours ×
   $0.12. Flag flat-24/7-heavy clusters where a right-sized provisioned
   instance + Reserved Instance likely beats serverless; flag
   bursty/off-hours/dev clusters as serverless wins (est. 40-60%).
4. Runaway guard: for serverless candidates, note the need for a hard max
   NCU (multi-hop traversals burn NCUs fast) and a billing alarm.

Report a table: cluster | workload shape | provisioned $/mo | est.
serverless $/mo | recommended min/max NCU | verdict | notes. Change
nothing.
Works with any assistant that can run shell commands.

Want the guided version?

The Neptune Serverless walkthrough covers this topic interactively — it asks about your setup, branches to what’s relevant, and quizzes you on the tricky parts. Free and anonymous.

Start the walkthrough

More Databases cheat sheets

Amazon RDS Proxy

Connection count, not CPU, is often what forces an RDS upsize. RDS Proxy pools thousands of client connections (especially from Lambda) onto ~10–30 real ones, so a small instance stays small — the proxy fee typically covers 10–25% of the avoided upsize.

Athena Partition Projection

Instead of asking S3 which partitions exist before every query, Partition Projection lets Athena calculate the paths from a pattern you declare — near-instant query startup on tables with thousands of partitions, zero LIST metadata calls.

Athena Query Result Reuse

Athena caches identical query results in S3, so a repeated dashboard refresh or scheduled report returns instantly with zero bytes scanned. A workgroup checkbox that routinely cuts thousands off a scan-heavy bill.

Aurora I/O-Optimized

Aurora bills a sneaky third bucket — I/O operations at ~$0.20/million — that can dwarf compute and storage on I/O-heavy workloads. I/O-Optimized zeroes it out for a ~30–40% higher instance rate. A break-even calc, not a default.