WarpStream

Use it when

• The Kafka bill is dominated by replicated disks and cross-zone traffic; writing straight to object storage removes both.
• Data must never leave your account: stateless agents run in your VPC, WarpStream operates only the control plane, and topic contents stay unreadable to the vendor.
• Elasticity should be trivial; agents are stateless, so scaling adds or removes processes and nothing rebalances.
• The workload is logs, telemetry, or analytics feeds, where seconds of latency are irrelevant.

Think twice when

• Latency is part of the product; batching into object storage adds end-to-end delay a local-disk Kafka cluster does not have.
• You want open source or self-contained self-hosting; WarpStream is proprietary and its control plane is the vendor's (AutoMQ is the Apache-2.0 alternative on the same architecture).
• Full Kafka protocol coverage is assumed; some broker requests are unsupported.

How it runs

Bring-your-own-cloud: deploy stateless agents in your VPC against your S3, GCS, or Azure storage; WarpStream Cloud tracks file-to-partition mapping, compaction, and retention. Managed Data Pipelines run Bento connectors inside the agents. Billing is uncompressed GiB written and stored plus cluster-minutes, with no availability-zone fees.