Kafka Connect itself is free. It is open source software shipped with Apache Kafka under the Apache 2.0 license. The cost question is how to run it: pay a vendor per connector task or throughput for a managed service, or pay in engineering time to self-host workers on your own infrastructure.
Cost is the live wire in this category. Derek and Kylie, our co-founders, have watched Confluent Platform and Cloud customers make up roughly a third of our customer base over the years, and the pattern Kylie names is that after the IBM acquisition, customers increasingly want independent tooling as a hedge against price rises. Read any managed-connector price list with that lens: the billing metrics below are the levers a vendor can move after you have committed your pipelines.
The core billing metrics
Managed connector services bill on two dimensions: compute time for the connector, and data volume through it.
A managed-connector pricing page and a redacted connector line item on a real bill.
Per-task or per-unit hours. Managed Kafka platforms typically charge each running connector task per hour. AWS MSK Connect bills per MSK Connect Unit hour, where a unit bundles compute and memory for connector workers (MSK Connect developer guide, MSK pricing page). Confluent Cloud charges fully managed connectors per task hour plus a throughput rate, stated in its own pricing documentation. Our review of Confluent Control Center pricing and alternatives breaks the platform’s billing dimensions down further.
Throughput charges. Data volume is billed per GB moved through the connector. Whether that volume is measured before or after compression differs by vendor and materially changes the bill on high-volume CDC pipelines, so it is a line worth checking in any quote.
Self-hosted licence pricing exists too. Commercial self-managed Kafka platforms price by subscription. Confluent Platform licence pricing typically lands in the tens to hundreds of thousands of dollars per year depending on cluster footprint.
Small-print pricing tiers are worth testing before you plan around them. When I evaluated Azure Event Hubs support, the cheapest dedicated option came to 6,000 US dollars a month, which meant I could not even keep a test environment running for extended verification until credits arrived. The billing unit that matters is the one that gates the deployment class you actually need, not the one on the headline tier.
Hidden network and add-on fees
The connector line item is not the whole bill. Three adjacent charges recur across managed offerings.
Private connectivity. Routing connector traffic through PrivateLink endpoints or VPC peering carries its own hourly and per-GB charges on the cloud side, and some vendors add a surcharge for private networking per task.
Cross-zone and egress data transfer. A connector that moves data between availability zones or out of the cloud region pays standard data transfer rates on top of the connector charge. CDC pipelines that fan out to external targets feel this most.
Custom code penalties. Running custom single message transforms or bringing an unmanaged connector plugin is either unsupported on the managed tier or priced separately. Platform add-ons beyond the base licence, such as management consoles, multi-tenancy and encryption features, are also commonly charged separately on self-managed commercial platforms.
Network fees are the ambush. Before 2023, cross-availability-zone traffic represented roughly half of Grab’s total Kafka platform cost, and connector traffic rides exactly those paths. On vendor API limits: Thomas, one of our engineers, measured a cloud schema registry rate-limiting connector requests at roughly 1 failure in 150 every two hours, visible only because we tested against 150+ subjects. Model the network path and the API quotas of a managed connector service at your real volumes, not the demo’s.
Total cost of ownership: the build versus buy math
Self-hosting Kafka Connect trades a vendor bill for engineering hours. Open source is not automatically cheaper once maintenance is counted honestly.
The recurring self-hosting costs are concrete: upgrading connector jars and testing compatibility, sizing and tuning worker JVMs, handling task rebalances and offset lag incidents, patching security issues in connector dependencies, and building the monitoring that a managed service includes. The infrastructure itself, the worker instances, is usually the smallest line.
The managed-service costs are equally concrete: per-task and per-GB charges that scale with the pipeline count, connectivity surcharges, and the constraint that custom transforms or niche connectors may not run at all on the managed tier.
The break-even is workload-shaped. A handful of standard connectors on modest volume favours managed. Many pipelines, high volume, custom transforms or strict data-residency requirements favour self-hosting, where a fixed-price management layer on top of open source Connect is the middle path between raw open source and a full commercial platform. What self-hosted operations actually involve day to day is laid out on the Kafka Connect topic page and in what is Kafka Connect.
The build-versus-buy math has a published anchor. In a case study we published, a healthcare data-platform’s third-party Kafka support renewal was approaching 150,000 US dollars a year, and bringing support in-house with better tooling saved roughly that full amount annually. One team I work with runs the other direction of the same math: 20 to 25 micro Kafka and Connect clusters per test environment at peak, planned deliberately so that tooling cost scales with environments rather than with a single monolithic cluster’s capacity tier. TCO for Connect is an architecture decision first and a price list second.
FAQ
How much does it cost to use Kafka Connect?
The software is free. Kafka Connect ships with Apache Kafka under the Apache 2.0 license. Costs come from running it: managed services bill per task hour and per GB moved, and self-hosting pays in worker infrastructure and the engineering hours to operate it.
Is a managed connector service cheaper than self-hosting?
It depends on workload shape. A handful of standard connectors on modest volume favours managed. Many pipelines, high volume, custom transforms or data-residency requirements favour self-hosting, once network surcharges and per-GB rates are modelled at real volumes.