Time-series Storage and Forecasting
Time-series work on AWS splits into two questions: where the data is stored, and what produces the forecast. The services older material recommends for both are closed to new customers.
Storing time-series data
Section titled “Storing time-series data”Amazon Timestream for LiveAnalytics was the purpose-built time-series database, with automatic scaling and built-in time-series analytic functions. It has been closed to new customers since 20 June 2025. Existing customers can continue to use it and to add users and linked accounts under the same payer account.
AWS recommends Amazon Timestream for InfluxDB for new time-series workloads. It is a managed InfluxDB offering with similar functionality — simplified ingestion and single-digit millisecond query response for real-time analytics.
Other reasonable homes for time-series data, depending on the shape of the workload:
- Amazon S3 in Parquet, partitioned by time, queried with Athena — for large historical series that are analysed in bulk rather than queried per point
- Amazon DynamoDB with a time-bucketed key — for high-write telemetry with known lookup patterns
- Amazon OpenSearch Service — where time-series data is queried alongside logs and needs aggregation and dashboarding
Forecasting
Section titled “Forecasting”Amazon Forecast, the managed time-series forecasting service, closed to new customers on 29 July 2024. Existing customers can continue to use it.
AWS directs forecasting work to Amazon SageMaker AI, and specifically to SageMaker Canvas for a no-code path comparable to what Forecast offered. Building the forecast yourself on SageMaker AI gives access to the standard time-series toolkit — classical statistical methods, gradient-boosted models over lag features, and deep forecasting models — with the model and its inputs under your control.
For the underlying theory, see Time series analysis, which covers trend, seasonality, stationarity and the transformations that get non-stationary data into a modellable form.