Confluent Pricing: What's Actually Published, What Isn't, and What IBM Changes

The Acquisition You Need to Know About First

IBM completed its acquisition of Confluent, Inc. on March 17, 2026, paying $31 per share in cash for an enterprise value of approximately $11 billion, per IBM's own newsroom announcement. Confluent still operates under its own brand: the pricing page still lives at confluent.io/pricing, and IBM's own Confluent product page routes both its "Start for free" and "Contact us" buttons straight back to confluent.io rather than hosting separate IBM pricing.

But IBM's press release is explicit that the acquisition is about embedding Confluent into IBM's existing stack, naming day-one integrations across watsonx.data, IBM MQ, IBM webMethods Hybrid Integration, and IBM Z. As of this writing, the public, self-serve pricing tiers haven't changed. What's more likely to shift is the enterprise sales motion, specifically the Dedicated tier and large committed-spend deals, since that's exactly where Confluent's revenue was already concentrated.

In its last independent quarter as a public company, Confluent reported 1,439 customers at $100,000 or more in annual recurring revenue. If you're already an IBM customer elsewhere, expect Confluent to increasingly show up as a line item in a broader IBM agreement rather than a standalone negotiation.

Confluent Cloud's Four Public Tiers, Rate by Rate

Confluent Cloud bills compute in a unit called an eCKU, short for Elastic Confluent Unit for Kafka, and per Confluent's own documentation, eCKUs scale automatically with actual usage rather than requiring manual provisioning. Basic starts at $0/month, with the first eCKU free and additional usage at $0.14/eCKU-hour, plus $0.05/GB for data in and out and $0.08/GB-month for storage. Standard starts at roughly $385/month with a $0.75/eCKU-hour rate and $0.035–$0.050/GB networking.

Enterprise starts around $895/month, with eCKUs running $1.75–$2.25/hour and networking at $0.020–$0.050/GB depending on traffic pattern. Freight, the cost-optimized, higher-latency specialty tier, starts at $2,300/month with a $2.25/hour eCKU rate, a 2-eCKU minimum, and the cheapest published storage rate at $0.03/GB-month. New organizations also get $400 in credit, valid for 30 days or until it's used, whichever comes first, on top of Basic's permanently-free first eCKU.

Every one of these numbers comes directly off Confluent's own pricing page. None of it requires a sales call.

The Fifth Tier: Why Dedicated Has No Listed Price

Dedicated clusters use CKUs instead of eCKUs: fixed, manually-provisioned capacity you pay for whether you use it or not, as opposed to the pay-for-actual-usage eCKU model on the other four tiers. Confluent's own pricing page describes Dedicated as built for "distinct networking and throughput requirements" and gives exactly one instruction for getting a rate: contact sales. This isn't a documentation gap, it's deliberate segmentation.

Confluent's own pricing FAQ confirms two negotiated mechanisms live specifically at this tier: an annual minimum-spend commitment that unlocks automatic tiered discounting as usage grows, and upfront volume-based discounts negotiated against your planned annual consumption. Practically, this means any workload with steady, high, predictable throughput, exactly the profile that pushes a team toward Dedicated instead of an elastic eCKU tier, cannot get a self-serve number from Confluent's own website at all. If your Confluent conversation has reached the point where someone is quoting you a CKU rate, you're already in a negotiation, not reading a price list.

What Else Gets Billed: Connect, Flink, and Governance

The eCKU or CKU rate covers Kafka itself, but Confluent Cloud bills four other product areas separately, and each is a real line item, not a rounding error. Connect, Confluent's managed connector layer, bills throughput at a per-GB rate plus a base fee per connector task, per task, per hour, regardless of how much data that specific task actually moves. That structure matters for CDC-style architectures where one source feeds several downstream connectors: task count, not just data volume, drives the bill.

Process, which covers Confluent Cloud for Apache Flink, bills in CFUs per minute, with Tableflow billed separately by topic-hours and gigabytes processed. Govern, Confluent's Stream Governance product, ships a free Essentials package covering 100 schemas and 10 exporters with a 99.5% Schema Registry SLA; the Advanced package, which raises the ceiling to 20,000 schemas and 100 exporters with a 99.99% SLA, starts at $1 per hour once you need it, per Confluent's own Stream Governance documentation. A team that stays inside the Essentials limits for its first year rarely notices this line item until schema growth or a real lineage requirement forces the upgrade.

The Storage Multiplier Almost Nobody Budgets For

Here's the single most common way a Confluent Cloud estimate ends up wrong: storage is billed on post-replication volume, not the raw size of your data, per Confluent's own billing FAQ. Kafka's standard replication factor is 3, meaning the data actually sitting on disk, and being billed, is roughly three times your logical data volume before any per-GB rate is even applied. A team estimating storage cost off the size of the data they're writing, rather than the size of what Kafka actually stores for availability, will underestimate that line item by a factor of three.

This isn't a Confluent markup; it's a mechanical consequence of how Kafka achieves durability. But it's rarely stated plainly in vendor marketing, and it's the first thing worth checking when a Confluent Cloud bill comes in higher than the eCKU-rate math predicted.

Self-Managed Confluent Platform: A Feature List With No Price Tag

Confluent Platform, the self-managed alternative, splits into three licensing tiers, confirmed directly from Confluent's own license documentation. Apache 2.0 covers core Kafka, Kafka Streams, and Kafka Connect: free and unrestricted. The Confluent Community License covers Schema Registry, ksqlDB, REST Proxy, and the Confluent CLI: also free, but not Apache-governed.

The Confluent Enterprise License covers the genuinely commercial layer: Confluent Server, Cluster Linking, Control Center, Confluent Replicator, MQTT Proxy, Role-Based Access Control, Tiered Storage, and Self-Balancing Clusters. A Developer license grants indefinite free use of that full commercial feature set, but only on a single-broker cluster, which makes it useless for any real production deployment. A Trial license grants 30 days of full production use, and per Confluent's own docs, adding a second broker to a Developer-licensed cluster automatically starts that 30-day trial clock, and you cannot revert back to a Developer license once it does.

After 30 days, the software stops functioning. Beyond the trial, you need a purchased Enterprise license, and Confluent's own subscription page describes what that includes, support up to what its own header calls Platinum Support, with 30-, 60-, or 90-minute response times depending on plan, but publishes no price or pricing formula anywhere. Every self-managed dollar figure you'll find in comparison articles, commonly cited in the $50,000 to $500,000-plus per year range, is a third-party estimate.

Confluent itself has never published one.

How Confluent Compares to AWS MSK and Redpanda on Published Numbers

AWS MSK is the most transparently priced of the three platforms researched here: every dimension AWS bills has an exact published rate. Provisioned Standard brokers run $0.204 to $0.21 per hour depending on instance type, plus $0.10/GB-month storage; Express brokers, a newer, higher-throughput option, run from $0.408/hour plus a $0.01/GB ingestion charge. MSK Serverless bills $0.75 per cluster-hour, $0.0015 per partition-hour, $0.10/GB data-in, $0.05/GB data-out, and $0.10/GB-month storage, all per AWS's own pricing page.

MSK Connect runs $0.11 per MCU-hour, where one MCU equals 1 vCPU and 4GB of memory. Redpanda sits at the opposite end: its own docs and pricing page publish the billing dimensions clearly, uptime, ingress, egress, partition-hours, and storage for Serverless, plus RPU-based compute for BYOC, but no exact dollar rate for any of them. The only concrete number on Redpanda's own marketing page is a comparison point aimed at Confluent specifically: a claim that Confluent Cloud Standard routes traffic through AWS NAT Gateways, incurring roughly $0.045/GB in data processing fees Redpanda's own PrivateLink-based architecture avoids.

None of that is a Redpanda price; it's a competitive argument about a Confluent cost. If you're comparing all three platforms on cost alone, AWS MSK is the only one where every number is verifiable without a sales call.

The Zero-License Option That Isn't Actually Zero-Cost

Apache Kafka itself, the open-source project all three commercial platforms build on, is licensed under Apache License 2.0, confirmed directly in the project's own LICENSE file on GitHub. That means genuinely zero license cost for running Kafka yourself. It does not mean zero cost.

Every operational responsibility a managed platform absorbs, broker capacity planning, upgrades, multi-AZ failover, monitoring, and the reliability layer a real production deployment needs, schema enforcement, consumer lag alerting, exactly-once processing boundaries where a duplicate has a real cost, becomes your own infrastructure and engineering labor instead of a vendor invoice line item. This is the actual tradeoff underneath "self-managed Kafka is free": it shifts cost from a predictable monthly bill to variable, harder-to-forecast headcount and operational risk, and most teams underestimate that shift until they're the ones getting paged for a stuck consumer group at 2 a.m.

How to Actually Estimate Your Own Number

There's no public calculator that replaces doing this walk by hand, and Confluent's own Cost Estimator does roughly the same thing. Start by identifying which tier fits your workload: Basic through Freight if throughput is variable or you want self-serve transparency, Dedicated only if throughput is steady and high enough that a negotiated provisioned rate would beat elastic per-hour billing at scale. Estimate Stream cost as eCKU-hours times the tier rate.

Estimate storage at three times your logical data volume, not the raw size, to account for replication. Estimate networking by actual traffic pattern against the tier's published range. Count Connect tasks, not just data volume, since the per-task base fee accrues regardless of how little data a given connector moves.

Check whether your schema and exporter count will cross Stream Governance's Essentials ceiling. And if you're evaluating self-managed Confluent Platform, map your required feature set against the Enterprise-license list above before assuming you need it at all: Schema Registry, ksqlDB, and REST Proxy are free under the Community License, and a lot of deployments don't actually need Cluster Linking or Tiered Storage on day one.

Related: Real-Time Data Engineering