| MSK provisioned | MSK Serverless | Event Hubs Standard | Google Managed Kafka 129 | |
|---|---|---|---|---|
| Engine | Apache Kafka 3.8-4.2 | Apache Kafka | Event Hubs, Kafka protocol | Apache Kafka 3.7.1 |
| You size | Brokers and disks | Nothing | Throughput units | vCPUs and memory |
| Client auth | IAM, SCRAM, mTLS | IAM only | Entra ID or SAS | IAM or mTLS |
| Managed Connect | MSK Connect | MSK Connect | None | Yes |
| BookNest, a month | About $465 | About $570 | About $22 (7-day retention) | About $250 |
Event Hubs is cheapest because it is not Kafka inside and stops short of it: no long history on Standard, transactions in preview, gzip only. Google's service is simple to size but runs an older Kafka. MSK gives the most Kafka at the highest floor. All three also bill data crossing zones or leaving the cloud. Choose the one where your producers and consumers already run; across clouds, see Confluent Cloud 17,245 and Aiven, Redpanda and Others.
On AWS 24 there is a fourth option that is not Kafka at all. Amazon Kinesis 24 Data Streams is AWS's native log, with its own API: records go to shards (each 1 MB/s or 1,000 records/s in, 2 MB/s out) chosen by a partition key. Neither is universally better:
| Kafka (MSK or any) | Kinesis Data Streams | |
|---|---|---|
| Interface | Open Kafka protocol, clients in every language | AWS API, SDKs and the Kinesis Client Library |
| Portability | Any cloud, on premises, a laptop | AWS only |
| Ecosystem | Connect, Streams, Flink 129 , Debezium 317,608 , registries | Lambda, Data Firehose, Managed Flink, Glue |
| Retention | Unlimited, with tiered storage | 24 hours, extendable to 365 days |
| Record size | 1 MB default, configurable | Up to 10 MiB |
| Pricing (US East) | Brokers per hour, or serverless | $0.015 per shard-hour, or on demand |
On demand, Kinesis bills $0.08 per GB written, $0.04 per GB read and $0.04 per stream-hour, so each of BookNest's three streams would cost about $29 a month; a provisioned one-shard stream about $11. Choose Kinesis when the organization is AWS-centric, the consumers are Lambda functions, Firehose deliveries and Managed Flink jobs, and no one wants to think about brokers or partitions. Choose Kafka (on MSK or anywhere else) when you need its ecosystem (Connect, Debezium, Kafka Streams 129 , a schema registry), long retention and replay, exactly-once transactions, or the freedom to move: a stream on Kafka can follow BookNest to another cloud with Choosing and Migrating's tools, while a stream on Kinesis must be rewritten against a new API first.