.png)
Key takeaways
1
Amazon Keyspaces bills across six dimensions, with throughput typically the largest 60–75% in the throughput-heavy scenarios modeled below. The other dimensions are where most teams miscalculate.
2
After the November 2024 price cut, on-demand is the recommended default for most workloads. Switch to provisioned only when CloudWatch shows sustained throughput that makes it cheaper.
3
Storage runs $0.25 per GB-month; PITR, TTL, and multi-region replication add to it based on configuration. A 1-year Database Savings Plan cuts stable workloads a further 18%.
Amazon Keyspaces bills across six dimensions: read and write throughput, storage, PITR backup, TTL deletes, Keyspaces Streams, and data transfer. This guide explains each one with current rates, worked cost examples, and how to choose between on-demand and provisioned mode.
The short answer
Amazon Keyspaces bills across six dimensions, and throughput is usually the biggest 60–75% in the throughput-heavy scenarios modeled below. After the November 2024 price cut, on-demand mode is the right default for most workloads.
Backup, TTL, and multi-region replication can meaningfully change what you pay. Stable workloads can take a further 18% off with a 1-year Database Savings Plan.
Backup, TTL, and multi-region replication can meaningfully change what you pay. Stable workloads can take a further 18% off with a 1-year Database Savings Plan.
The Six Billing Dimensions of Amazon Keyspaces
In throughput-heavy production workloads, throughput is the largest component of spend, storage is second, and PITR, TTL, Streams, and data transfer make up the rest actual shares vary by access pattern and configuration.
Billing Dimension 1: Read and Write Throughput
Throughput is always the largest line item for active Keyspaces workloads. You choose one of two modes per table: on-demand (pay per request) or provisioned (pay for capacity reserved per second per hour).
On-Demand Mode (Recommended Default Since November 2024)
On-demand mode charges per read or write request:
Current US East (N. Virginia) rates: $0.625 per million WRUs and $0.125 per million RRUs (verified August 6, 2026). Estimates built on pre-November 2024 rates significantly overstate current costs.
Two cost behaviors worth knowing:
- Writes (WRUs): 1 WRU handles one write of up to 1 KB of row data using LOCAL_QUORUM consistency. All Keyspaces writes use LOCAL_QUORUM.
- Reads (RRUs): 1 RRU handles one LOCAL_QUORUM read of up to 4 KB, or two LOCAL_ONE reads of up to 4 KB each.
- A 3 KB write requires 3 WRUs.
- A 10 KB LOCAL_QUORUM read requires 3 RRUs; the same read at LOCAL_ONE requires 1.5 RRUs.
- If a query returns multiple rows, billing is based on the aggregate data size returned, not row count.
Current US East (N. Virginia) rates: $0.625 per million WRUs and $0.125 per million RRUs (verified August 6, 2026). Estimates built on pre-November 2024 rates significantly overstate current costs.
Two cost behaviors worth knowing:
- Lightweight Transactions (LWTs) carry no premium in either capacity mode, per AWS’s LWT capacity documentation. A failed condition check still consumes WRUs based on the existing row’s size, and on TTL-enabled tables that don’t use client-side timestamps, LWTs can consume additional RRUs.
- Idle tables cost $0 in throughput. There is no minimum charge, making on-demand the natural start for dev, staging, and seasonal tables (storage still applies).
Provisioned Mode (For Stable High-Throughput Workloads)
Provisioned mode charges per capacity unit per hour, regardless of whether you consume the full provisioned capacity:
- Reads (RCUs): 1 RCU provides one LOCAL_QUORUM read per second for rows up to 4 KB, or two LOCAL_ONE reads per second.
- Writes (WCUs): 1 WCU provides one LOCAL_QUORUM write per second for rows up to 1 KB.
Multi-Region Replication: The Write Cost Multiplier
With active-active Multi-Region Replication, you are billed 1 WRU (or 1 WCU) per KB written per region, a 3 KB write costs 6 WRUs in a 2-region keyspace and 9 in a 3-region keyspace.
Reads stay local and replication data transfer is free, but storage and PITR are billed in each region, since each stores a full copy.
Reads stay local and replication data transfer is free, but storage and PITR are billed in each region, since each stores a full copy.
Billing Dimension 2: Storage
Keyspaces charges $0.25 per GB-month for data storage. You do not provision storage in advance AWS monitors your table size continuously to determine charges.
Billable size is based on the raw, uncompressed size of a single replica; you are not billed for the three-way replication Keyspaces maintains.
Teams migrating from self-managed Cassandra should expect billable size to be higher than their compressed on-disk footprint; the estimation formula in AWS’s Cassandra migration guide is (table live space ÷ compression ratio) ÷ replication factor.
Storage is reduced when rows expire via TTL or are explicitly deleted, but not immediately: expired or deleted data can remain billable until Keyspaces physically removes it, typically within about 10 days for tables using client-side timestamps. A table that shrinks from 100 GB to 50 GB is billed for 50 GB once cleanup completes.
Billable size is based on the raw, uncompressed size of a single replica; you are not billed for the three-way replication Keyspaces maintains.
Teams migrating from self-managed Cassandra should expect billable size to be higher than their compressed on-disk footprint; the estimation formula in AWS’s Cassandra migration guide is (table live space ÷ compression ratio) ÷ replication factor.
Storage is reduced when rows expire via TTL or are explicitly deleted, but not immediately: expired or deleted data can remain billable until Keyspaces physically removes it, typically within about 10 days for tables using client-side timestamps. A table that shrinks from 100 GB to 50 GB is billed for 50 GB once cleanup completes.
Billing Dimension 3: PITR Backup
PITR (Point-In-Time Recovery) provides continuous, automatic backups with per-second granularity. When enabled, you can restore the table to any second within the preceding 35 days. PITR must be explicitly enabled on each table, it is not on by default.
PITR costs $0.20 per GB-month, based on the table’s size, which AWS monitors continuously. The charge continues until PITR is disabled. Restoring a table costs $0.15 per GB of data restored.
Cost example: a 100 GB table with PITR enabled adds $20/month on top of the $25/month storage cost, an 80% increase to the base storage bill.
Enable PITR only where the 35-day restore is required for compliance or recovery. For append-only tables like event logs, disabling it eliminates the charge but confirms first that the data can be replayed from another source, since PITR also protects against accidental deletes and bad writes.
Also read: What Is Amazon ElastiCache? The Complete 2026 Guide
PITR costs $0.20 per GB-month, based on the table’s size, which AWS monitors continuously. The charge continues until PITR is disabled. Restoring a table costs $0.15 per GB of data restored.
Cost example: a 100 GB table with PITR enabled adds $20/month on top of the $25/month storage cost, an 80% increase to the base storage bill.
Enable PITR only where the 35-day restore is required for compliance or recovery. For append-only tables like event logs, disabling it eliminates the charge but confirms first that the data can be replayed from another source, since PITR also protects against accidental deletes and bad writes.
Also read: What Is Amazon ElastiCache? The Complete 2026 Guide
Billing Dimension 4: TTL Deletes
Time to Live (TTL) automatically deletes rows or columns after a configured expiration. Pricing is based on the size of data deleted: 1 TTL delete unit = 1 KB deleted per row, so deleting an entire 2.5 KB row requires 3 units. If multiple columns in the same row expire simultaneously, units are consumed per KB of the row, not per expired column.
The November 2024 price cut reduced TTL delete prices by 75%, making TTL far cheaper than explicit deletes:
The November 2024 price cut reduced TTL delete prices by 75%, making TTL far cheaper than explicit deletes:
- Explicit DELETE of a 3 KB row: 3 WRUs at the full write rate
- The same deletion via TTL: 3 TTL delete units at $0.275 per million under half the WRU rate
Billing Dimension 5: Keyspaces Streams
Keyspaces Streams captures a time-ordered sequence of item-level modifications for replication, downstream processing, or audit trails.
Each GetRecords API call is billed as one stream read request unit at $0.02 per 100,000 units, returns up to 1 MB of data, and is priced separately from table RRUs. Streams incur no charges unless enabled.
Each GetRecords API call is billed as one stream read request unit at $0.02 per 100,000 units, returns up to 1 MB of data, and is priced separately from table RRUs. Streams incur no charges unless enabled.
Billing Dimension 6: Data Transfer
Data transfer into Keyspaces is free, as is transfer between Keyspaces and other AWS services within the same region ($0.00 per GB) and cross-region transfer for native Multi-Region Replication.
Internet egress follows standard AWS rates: the first 1 GB per month is free (aggregated across all AWS services), then approximately $0.09/GB for the first 10 TB/month, tiered above that.
Also read: Amazon Keyspaces Serverless: The Complete 2026 Guide to Managed Cassandra on AWS
Internet egress follows standard AWS rates: the first 1 GB per month is free (aggregated across all AWS services), then approximately $0.09/GB for the first 10 TB/month, tiered above that.
Also read: Amazon Keyspaces Serverless: The Complete 2026 Guide to Managed Cassandra on AWS
On-Demand vs Provisioned: Which Is Cheaper?
Since the November 2024 reduction, on-demand is competitive or cheaper for most workload patterns. The break-even is a utilization question, not a request-rate threshold:
Compare the two modes on the same month.
On-demand monthly cost
=
Actual requests
×
per-request rate
Provisioned monthly cost
=
Provisioned capacity
×
hourly rate
×
730 hours
Use your real utilization.
Pull ConsumedWriteCapacityUnits and ConsumedReadCapacityUnits from CloudWatch —
a mode that's cheaper at 100% utilization is often more expensive at your actual average.
At current N. Virginia rates, provisioned works out roughly 3.5× cheaper than on-demand at 100% sustained utilization meaning on-demand is the cheaper mode whenever average utilization of the capacity you’d have to provision sits below roughly 30%.
| Your workload pattern | Recommended mode | Why |
|---|---|---|
| Unpredictable or spiky traffic | On-demand | No capacity planning; scales with traffic; cost-competitive after Nov 2024 |
| Steady, forecastable throughput at consistently high utilization | Provisioned with auto scaling | Per-unit-hour rate beats per-request billing at sustained utilization |
| New application, traffic unknown | On-demand | Start here; revisit once CloudWatch shows a stable baseline |
| Occasional large bursts on an otherwise quiet table | On-demand | Provisioned throttles during bursts unless minimums cover the spike |
| Stable baseline spend qualifying for DSP | Either mode | Layer Database Savings Plans on top; DSP applies to both |
The simplest rule: start with on-demand, and switch only if CloudWatch shows sustained utilization that makes provisioned cheaper. You can change modes once per day without data loss or interruption.
One caveat: on-demand instantly accommodates up to double your previous traffic peak, but ramps beyond that within 30 minutes can throttle pre-warm or ramp gradually before major events, especially on new or recently quiet tables.
One caveat: on-demand instantly accommodates up to double your previous traffic peak, but ramps beyond that within 30 minutes can throttle pre-warm or ramp gradually before major events, especially on new or recently quiet tables.
Amazon Keyspaces Pricing Summary Table
| Billing component | Rate | Notes |
|---|---|---|
| On-demand write (WRU) | $0.625 per million WRUs | 1 WRU = 1 KB write, LOCAL_QUORUM. No extra charge for LWTs. |
| On-demand read, LOCAL_QUORUM (RRU) | $0.125 per million RRUs | 1 RRU = 4 KB read; larger rows consume proportionally more. |
| On-demand read, LOCAL_ONE | 0.5 RRU per 4 KB read | Half the cost of LOCAL_QUORUM; eventual consistency. |
| Multi-region on-demand writes | WRU rate × number of regions | 3 KB writes to a 2-region keyspace = 6 WRUs. |
| Provisioned read (RCU per hour) | $0.00013 per RCU-hour | 1 RCU = 1 LOCAL_QUORUM read/sec up to 4 KB. |
| Provisioned write (WCU per hour) | $0.00065 per WCU-hour | 1 WCU = 1 write/sec up to 1 KB. |
| Storage | $0.25/GB-month | Billed on raw, uncompressed size of one replica; auto-scales. |
| PITR backup | $0.20/GB-month | Optional, per table; 35-day window; billed until disabled. |
| Table restore (from PITR) | $0.15/GB restored | Charged per restore request on total data restored. |
| TTL deletes | $0.275 per million TTL delete units | 1 unit = 1 KB deleted per row; reduced 75% in Nov 2024. |
| Keyspaces Streams reads | $0.02 per 100,000 stream read request units | Per GetRecords call; up to 1 MB; separate from table RRUs. |
| Data transfer (same region) | $0.00 | Includes native multi-region replication transfer. |
| Data transfer (internet egress) | $0.09/GB after 1 GB free/month | Tiers down to $0.085, $0.07, and $0.05/GB at higher volumes. |
| Free tier | 30M WRUs + 30M RRUs + 1 GB storage/month | First 3 months from first Keyspaces resource; one per payer account. |
All rates: US East (N. Virginia), verified from the AWS Keyspaces pricing page on August 6, 2026. Throughput and TTL rates reflect the November 14, 2024 price reduction; rates vary by region and can change.
Worked Example 1: Small Event Logging Application
Scenario:
- Writes: 5 million events/day, 1 KB rows
- Reads: 500,000/day for analytics, 2 KB rows, LOCAL_ONE consistency
- Config: single region, no PITR (logs replayable from source), TTL set to 90 days
- Steady-state data: at least ~450 GB (5M rows/day × 1 KB × 90-day retention, before row-encoding and TTL metadata overhead)
- Writes: 150 million WRUs × $0.625/M = $93.75
- Reads: 15 million reads × 0.5 RRU (LOCAL_ONE, rows under 4 KB) = 7.5 million RRUs × $0.125/M = $0.94
- Storage: 450 GB × $0.25 = $112.50
- TTL: ~150 million delete units × $0.275/M = $41.25
Worked Example 2: High-Throughput Session Store
Scenario:
Running both formulas at current N. Virginia rates: on-demand costs ~$20,531 (writes) + ~$2,738 (reads) ≈ $23,270/month, while provisioned at this flat utilization costs 12,500 WCUs and 8,333 RCUs × 730 hours ≈ $6,722/month roughly 71% cheaper.
This is the flat 24/7 pattern where provisioned with auto scaling wins decisively. A 30–60 minute TTL expires session rows automatically, far cheaper than explicit deletes.
- Sessions: 50,000 active at any time, 3 KB each
- Writes: 5/minute per session → 4,167 writes/sec × 3 WRUs = 12,500 WRUs/sec sustained
- Reads: 10/minute per session → 8,333 reads/sec × 1 RRU (3 KB, LOCAL_QUORUM)
- Traffic: flat 24/7, single region
Running both formulas at current N. Virginia rates: on-demand costs ~$20,531 (writes) + ~$2,738 (reads) ≈ $23,270/month, while provisioned at this flat utilization costs 12,500 WCUs and 8,333 RCUs × 730 hours ≈ $6,722/month roughly 71% cheaper.
This is the flat 24/7 pattern where provisioned with auto scaling wins decisively. A 30–60 minute TTL expires session rows automatically, far cheaper than explicit deletes.
Database Savings Plans for Keyspaces
Database Savings Plans cover Amazon Keyspaces with up to 18% savings on on-demand throughput and up to 12% on provisioned capacity, on a 1-year, no-upfront commitment coverage AWS confirms in the AWS Savings Plans FAQ.
The commitment is a dollar-per-hour spend level not tied to specific tables or capacity modes.
Teams with stable month-over-month spend benefit most; if usage is still shifting, wait until it has been consistent for 60–90 days. For commitment sizing methodology, see Database Savings Plans for Amazon Keyspaces.
The commitment is a dollar-per-hour spend level not tied to specific tables or capacity modes.
Teams with stable month-over-month spend benefit most; if usage is still shifting, wait until it has been consistent for 60–90 days. For commitment sizing methodology, see Database Savings Plans for Amazon Keyspaces.
How Usage.ai Optimizes Keyspaces Costs
Keyspaces has no Reserved Instances, a Database Savings Plan is its only commitment discount, and Usage.ai automates DSP across all 10 eligible services.
Usage.ai analyzes your database spend from the AWS Cost and Usage Report and sizes the commitment to your consistent hourly spend floor over the past 60 days, a single commitment that applies across all DSP-eligible services.
Because the discount applies in both capacity modes, you can run the on-demand vs. provisioned comparison from this guide first and layer the commitment on top of whichever mode wins.
Flex Commitments purchased through Usage.ai are insured, you get the up to 18% savings of a 1-year Database Savings Plan with none of the commitment risk.
If a commitment ever costs more than the equivalent on-demand usage, Usage.ai pays the difference back as cashback in real money, and the fee is a percentage of realized savings only.
Usage.ai analyzes your database spend from the AWS Cost and Usage Report and sizes the commitment to your consistent hourly spend floor over the past 60 days, a single commitment that applies across all DSP-eligible services.
Because the discount applies in both capacity modes, you can run the on-demand vs. provisioned comparison from this guide first and layer the commitment on top of whichever mode wins.
Flex Commitments purchased through Usage.ai are insured, you get the up to 18% savings of a 1-year Database Savings Plan with none of the commitment risk.
If a commitment ever costs more than the equivalent on-demand usage, Usage.ai pays the difference back as cashback in real money, and the fee is a percentage of realized savings only.
Frequently asked questions
How much does Amazon Keyspaces cost?
Keyspaces bills across six dimensions: on-demand throughput at $0.625 per million WRUs and $0.125 per million RRUs, storage at $0.25/GB-month, PITR at $0.20/GB-month, TTL deletes at $0.275 per million units, plus Streams reads and data transfer (US East, N. Virginia). Rates vary by region see aws.amazon.com/keyspaces/pricing.
What changed in Keyspaces pricing in November 2024?
AWS cut on-demand prices by up to 56% (single-region) and 65% (multi-region), provisioned by up to 13% and 20%, and TTL deletes by 75%. On-demand is now the recommended default for most workloads.
When is provisioned mode cheaper than on-demand?
When traffic is near-flat and utilization of provisioned capacity stays consistently high. Compare both modes at your actual CloudWatch consumption; you can switch modes once per day.
How does multi-region replication affect pricing?
Writes are billed per WRU per region, a 2-region keyspace doubles write costs, a 3-region keyspace triples them. Reads stay local, replication transfer is free, and storage and PITR are billed in each region.
What is the Keyspaces free tier?
30 million WRUs, 30 million RRUs, and 1 GB of storage per month for the first 3 months, starting when you create your first Keyspaces resource. Limit one per payer account.
Share
Facebook
X
LinkedIn
Reddit
