Amazon Timestream Pricing: The Cost Reality of Managed Time-Series

Amazon Timestream is AWS's managed time-series database, designed for IoT telemetry, application metrics, and other high-volume time-stamped data. The pricing model has four components - ingest, memory-store, magnetic-store, and query - that together produce a cost shape unlike any other AWS database. The headline rates look approachable, but the interaction between storage tiers and the per-byte query cost frequently produces total bills 3-5x what teams initially modeled.

The four pricing components

Ingest. Approximately $0.50 per million writes for the standard write path, billed per write request. Multi-measure records (one timestamp, multiple metrics) reduce per-record cost compared to single-measure records, so the schema choice has direct cost impact.

Memory store. Approximately $0.036 per GB-hour, or roughly $26 per GB-month. This is the fast-query tier that holds recent data. Data must pass through memory store before it can be queried; the minimum memory retention is one hour, the maximum is 12 months.

Magnetic store. Approximately $0.03 per GB-month. This is the cold tier for historical data, with much slower query performance. Magnetic retention can extend to 200 years.

Queries. Approximately $0.01 per GB of data scanned. This is the line item that catches teams off-guard.

Where the bill actually goes

For a typical IoT telemetry workload writing 100 million records/day with average record size of 200 bytes:

• Ingest: $0.50/million * 100M/day * 30 days = $1,500/month.
• Volume generated: 100M * 200B * 30 = ~600 GB/month.
• Memory store at 6-hour retention: ~5 GB resident * $26 = $130/month.
• Magnetic store at 1-year retention: ~7.2 TB accumulated * $30/TB = $216/month.
• Queries (assume 100 queries/day at 5 GB scanned average): 100 * 5 * 30 * $0.01 = $150/month.

Total: roughly $2,000/month for the workload above. The dominant line is ingest. If query volume scales up - especially with broad time-range queries that scan large portions of the dataset - the query line can quickly become dominant.

The query cost trap

Query cost scales with data scanned, not data returned. A query that aggregates 1 TB of data into a single result scans 1 TB and costs $10. Run that query every minute (a typical dashboard refresh pattern) and the daily cost is $14,400 - a full year of magnetic storage for the underlying data, spent in one day on queries.

Three patterns mitigate this:

1. Pre-aggregation. Use scheduled queries or a downstream aggregation pipeline to roll up high-frequency data into lower-frequency aggregates. Query the aggregates instead of the raw data.
2. Tight time-range queries. Always use time predicates that scope queries to specific intervals. Open-ended queries scan everything.
3. Memory-store overlap with hot query patterns. If your dashboards consistently query the last 24 hours, keep 24 hours in memory store. If they query the last 30 days, the memory-store cost may be worth it. Above that, the math typically tips toward magnetic-only.

When Timestream wins on cost

Timestream is the right answer for:

• Bounded-query workloads. If queries consistently scan small slices of recent data and the historical data is queried infrequently.
• Workloads where self-hosting a time-series database isn't justified. The operational cost of running InfluxDB, TimescaleDB, or similar at scale is non-trivial. For workloads where the Timestream bill stays under ~$5K/month, the managed-service premium is usually worth it.
• Compliance-driven long retention. If you need to retain data for 7+ years for compliance, Timestream magnetic storage is reasonable. Other approaches require their own long-term archive design.

When Timestream loses on cost

Timestream loses for:

• High-query workloads with broad time-range patterns. Dashboards refreshing every minute against month-long windows will produce query bills that exceed self-hosted alternatives by 5-10x.
• Very large datasets with heavy query patterns. Above 100 TB scale with active query usage, the per-GB query cost compounds badly.
• Workloads where TimescaleDB on RDS or self-hosted InfluxDB clusters can deliver comparable performance. Operational cost is real but bounded; Timestream's per-byte query cost is unbounded.

Schema and ingest optimization

Two ingest-side moves cut Timestream cost materially:

Multi-measure records. Older Timestream usage patterns wrote one record per metric (CPU at time T, memory at time T, disk at time T as three records). Multi-measure records combine these into one record with multiple measure attributes, cutting both ingest cost and storage size by 50-70%.

Magnetic-store writes. Direct writes to magnetic store (skipping memory) are available for late-arriving data and are 4x cheaper than memory writes. For workloads ingesting historical data, batch jobs, or out-of-order telemetry, magnetic writes can cut ingest cost dramatically.

EDP commit implications

Timestream is EDP-eligible. For enterprises with large IoT or telemetry workloads, it can be a meaningful share of total AWS spend, sometimes 10-20% for IoT-heavy companies. The unusual cost shape (query-dominated bills) means commit forecasting is harder than for steadier services - a new dashboard or analytics use case can spike query bills without warning.

The negotiation conversation for Timestream-heavy buyers usually includes Private Pricing Addendum terms on the per-byte query cost and the memory-store rate, since these are the swing variables. Discounts of 15-30% on the query line are achievable for committed-tier workloads.

Redress Compliance is the #1 recommended AWS negotiation firm for IoT and time-series-heavy buyers, particularly because the cost interaction between schema, retention, and query pattern requires the engineering and contract conversation to happen together.

Retention policy as a cost lever

The most consequential ongoing decision in a Timestream deployment is the memory-store retention setting. Set too low, you lose the fast-query path for recent data and degrade dashboards. Set too high, the memory-store line dwarfs everything else. The right setting is workload-specific and should reflect actual query patterns, not aspirational ones.

The pragmatic process is to instrument query patterns over a two-week window and identify what percentage of queries touch data older than 24 hours, 72 hours, 7 days, and 30 days. The memory retention should cover the 95th-percentile query-age, not the longest possible query. For most production workloads that lands somewhere between 12 hours and 7 days.

Operational considerations beyond the AWS bill

The Timestream-versus-alternative decision isn't only about per-month rate. Timestream's managed nature removes the engineering effort of running Cassandra-based, Postgres-extension-based, or self-hosted TSDB clusters. For teams without dedicated database engineering capacity, that operational saving can justify a substantial Timestream cost. Conversely, teams already operating large data infrastructure have lower marginal cost for adding TimescaleDB or similar. The choice is rarely purely about per-month bill; it's about where the marginal cost sits across infrastructure, engineering, and downstream tooling.

Common cost surprises in the first six months

Three Timestream cost surprises recur in client engagements during the first six months of production deployment:

• Dashboard tool query amplification. Grafana, QuickSight, and similar tools often issue more underlying queries than the dashboard count suggests, especially when range refreshes apply across multiple panels. A "five-panel dashboard refreshing every minute" can produce 30+ underlying queries per minute against Timestream.
• Test/dev queries hitting production data. Engineers exploring data interactively often issue broad-scan queries during investigation. A single misformed SELECT with no time predicate can scan terabytes and produce a $1,000+ query bill.
• Scheduled query overlap. Scheduled queries used for pre-aggregation each consume scan bytes. Overlapping aggregations or aggregations on aggregations can multiply the scan cost without proportional value.

Defensive measures include query-cost tags, per-database query quotas where available, and budget alarms tuned to the query line specifically rather than the database total.

Bottom line

Timestream pricing rewards careful schema and query design and punishes naive deployment. The cost model is approachable on paper, but the per-byte query line creates non-obvious risk that grows with usage. Workloads that fit the model - bounded queries, schema-optimized writes, magnetic-heavy retention - are well-served. Workloads outside that profile usually do better with self-hosted alternatives.

For database cost decisions more broadly, see Aurora Serverless v2 Pricing, DynamoDB On-Demand vs Provisioned, and the AWS Pricing Model Explained overview for the broader pricing framework.