Pulsar

Pulsar is Apache's answer to Kafka, handling both messaging patterns (pub/sub and queuing) in one system. The key difference: Pulsar separates compute from storage, so you can scale message brokers and storage independently. For teams that have hit Kafka's operational ceiling, that architecture matters.

Apache Foundation, Java-based. Multi-tenancy is built in; namespaces and topics can be isolated per team without running separate clusters. Supports both streaming (like Kafka) and queuing (like RabbitMQ) patterns in the same system. Geo-replication works across data centers out of the box.

Fully open source, Apache 2.0. StreamNative offers a managed Pulsar cloud starting around $0.10/GB ingested, but self-hosting is the primary path. Self-hosting is where the ops burden hits. Pulsar needs ZooKeeper (or the newer metadata store) plus BookKeeper for storage. Minimum production setup is 3 brokers + 3 bookies + 3 ZK nodes. That's 9 machines before you process a single message. Solo or small teams: use NATS or RabbitMQ unless you specifically need multi-tenancy or geo-replication. Medium teams (10-50): Pulsar makes sense if you're outgrowing simpler solutions. Large: this is where Pulsar shines.

The catch: the operational complexity is real. If you don't need multi-tenancy or tiered storage, Kafka is simpler to run and has a larger ecosystem of tooling and connectors.