The full answer
Kinesis Data Streams is a durable, replayable stream that your own applications read; Amazon Data Firehose (formerly Kinesis Data Firehose) is a delivery service that buffers incoming data and writes it to a destination such as S3, with no consumer code.
| Kinesis Data Streams | Amazon Data Firehose | |
|---|---|---|
| What it is | A stream you produce to and consume from | A managed pipe to a destination |
| Consumers | Your code: Lambda, Flink, KCL apps, several at once | None; Firehose delivers for you |
| Latency | Under a second | Buffered: by size and time, from about zero seconds up to 15 minutes |
| Retention and replay | 24 hours by default, up to 365 days; consumers can re-read | No retention; data is delivered and gone |
| Ordering | Ordered per shard (partition key) | Not a concern; it batches |
| Capacity | On-demand mode scales automatically; provisioned mode by shards | Fully automatic |
| Transformation | Whatever your consumer does | Optional Lambda transform; JSON to Parquet or ORC conversion; dynamic partitioning; compression |
| Destinations | Anything your consumers write to | S3, Redshift, OpenSearch, Apache Iceberg tables, Splunk, Snowflake, HTTP endpoints and partner tools |
Use Kinesis Data Streams when
- several applications must read the same events independently (fraud checks, a live dashboard and an archive);
- you need sub-second processing or per-key ordering;
- you need to replay past events after a bug fix or to start a new consumer.
Use Firehose when
- the goal is simply to land data in S3, a warehouse or a search index;
- you want files batched, compressed and converted to Parquet without writing code;
- a delay of seconds to minutes is fine.
Using them together is common: producers write to Kinesis Data Streams, real-time consumers read it, and a Firehose stream with that data stream as its source archives everything to S3 as Parquet for analytics.
(Kinesis Data Streams can now also deliver directly to Iceberg tables in S3 Tables or to S3 without Firehose, which is worth mentioning if the interviewer asks what is new.)
Last updated 16 October 2026
Sources