Which feature of Kinesis Data Streams enables horizontal scalability?

Sharding is the correct answer because it is the mechanism used by Amazon Kinesis Data Streams to achieve horizontal scalability. Each shard is a unit of capacity that can ingest data in a specific amount per second and provides a certain level of read and write throughput. By adding more shards, you can increase the overall capacity of the stream, allowing for parallel processing and enabling multiple consumers to read from the stream at the same time. This capability allows Kinesis to handle varying workloads effectively, adapting to the volume of incoming data and providing the elasticity needed for different applications.

The other features, while beneficial in their own right, do not directly contribute to the ability of Kinesis Data Streams to scale horizontally. Ordering guarantees ensure that records within a shard are processed in the order they are received, but they do not affect the scaling capabilities of the service. Data transformations are typically handled outside the core streaming mechanism and provide processing capabilities rather than scalability. Machine learning integration allows users to implement predictive analytical capabilities by leveraging real-time data but does not pertain to the scalability of the stream itself.