AWS · Free tool
Provisioned and on-demand priced side by side on the same workload. Plug in reads, writes, GSI count, and storage. See the utilization tipping point where switching modes pays off.
Cheaper option for this workload
$0
Pick a mode below
| Component | Provisioned | On-demand |
|---|---|---|
| Read capacity | $0 | $0 |
| Write capacity | $0 | $0 |
| GSI multiplier | $0 | $0 |
| Storage | $0 | $0 |
| PITR | $0 | $0 |
| Streams | $0 | $0 |
| Monthly total | $0 | $0 |
Provisioned: 0 RCU + 0 WCU at 60% util. On-demand: 0M reads + 0M writes/mo.
On-demand is cheaper below 14 percent sustained utilization. You modeled 60 percent.
If your DynamoDB bill looks high, check these first: on-demand tables with predictable sustained traffic, GSIs nobody queries anymore, strongly-consistent reads where eventual consistency would do, PITR enabled on session or cache tables, cold history tables still on Standard storage, and items larger than 1 KB on write-heavy tables.
On-demand DynamoDB is $0.125 per million read request units and $0.625 per million write request units. Provisioned is $0.00013 per RCU-hour and $0.00065 per WCU-hour. Across 730 hours, on-demand is roughly 7x the per-request cost at full utilization.
Autoscaling changes the operational tradeoff by targeting 60 to 75 percent utilization and adjusting capacity every minute. The full pattern is in the on-demand vs provisioned tipping point.
When should I use on-demand vs provisioned capacity?
On-demand is roughly 7x more expensive per request than fully-utilized provisioned capacity. Provisioned wins when sustained utilization is above 14 to 18 percent. On-demand wins when traffic is spiky, unpredictable, or below that threshold.
Why are GSIs so expensive?
Each Global Secondary Index has its own RCU, WCU, and storage. Every write to the base table is replicated to every GSI, so a table with three GSIs writes four times. Audit your indexes for stale access patterns.
Is PITR worth the cost?
Point-in-time recovery is $0.20 per GB-month on top of storage. For tables holding production state it is essential. For ephemeral tables such as sessions, caches, or temporary work queues, it is usually dead weight.
Should I use Standard-IA storage class?
Standard-IA cuts storage cost by 60 percent and raises read and write rates by about 25 percent. Tables holding cold history can save money on Standard-IA. Hot tables usually lose.
