Messaging is sold in more different ways than any other middleware, which makes it the favourite subject for the licensing question in a senior interview. The prices below are list prices observed on 26 September 2026 from the vendors’ own pricing pages unless marked otherwise; a bank pays less, but the shape of each bill does not change with a discount.

IBM licenses MQ on the two processor metrics described in web middleware licensing: the Processor Value Unit (PVU), a per-core rating by processor type, for traditional perpetual licences with annual subscription and support; and the Virtual Processor Core (VPC) — one virtual core, not a cloud Virtual Private Cloud — for subscriptions and containers. IBM’s pricing page, observed 26 September 2026, publishes starting prices for the subscriptions:

Since MQ 9.4.3, IBM also offers add-on licensing for Native HA and Cross-Region Replication in containers, and it sells lower-priced entitlements for standby replicas in some high-availability configurations — so check which nodes need a full entitlement before counting every server in a three-node group. IBM MQ on the mainframe is priced separately, under IBM’s mainframe software pricing rather than PVUs.

Confluent prices its managed Kafka on three meters: capacity in eCKUs (elastic Confluent Units for Kafka, billed hourly on what a cluster actually uses, up to a ceiling), data in and out per GB, and storage per GB-month. Dedicated clusters use fixed CKUs (Confluent Units for Kafka) paid for whether used or not. From Confluent’s pricing page:

The self-managed Confluent Platform is priced by quote only.

Apache Kafka, RabbitMQ, Apache Artemis, ActiveMQ Classic and Pulsar cost nothing to license. The licence is zero and the on-call rota is not. A production cluster needs servers, disks multiplied by the replication factor, cross-zone network charges in the cloud, and engineers able to run it around the clock — which is why the managed services exist. For RabbitMQ, remember Broadcom’s policy: patches for older release lines are for commercial customers.

Amazon SQS is the clearest per-request model. Every customer gets one million requests a month free; after that, the standard-queue rate is $0.40 per million requests and FIFO (first-in, first-out) queues cost more. AWS’s own rate table is rendered by script and did not load for us; the $0.40 figure is the one AWS-focused sources quote consistently, and the free tier and billing rules below are from AWS’s page. Two billing rules matter more than the rate: every 64 KB of payload is a separate request, so a 1 MiB message costs 16; and there is no data-transfer charge within a region. Azure Service Bus Premium is priced per messaging unit — reserved capacity billed by the hour — with its rates published through Microsoft’s calculator.

A steady 2,000 messages a second of 2 KB, around the clock: 5.18 billion messages a month. The column that matters most is the second one — a licence price and a running cost are not the same thing.

Read the table as a lesson in comparing like with like, not as a ranking. The IBM MQ figures exclude the servers and the people; the managed services include them. The SQS line shows the classic trap — ten-fold cheaper with batching, which costs nothing to implement — and how quickly per-request pricing overtakes capacity pricing at sustained volume.

Two abbreviations in the table: JMS is Jakarta Messaging, the Java messaging API; AMQP is the Advanced Message Queuing Protocol, an open wire standard.

• Commands to systems of record, mainframe or SWIFT: IBM MQ — the integration and the audit history outweigh the licence.
• Events, fan-out, replay, feeding data platforms: Kafka, managed unless the bank has a platform team that genuinely wants to run it.
• A new cloud-native application inside one cloud: the provider’s queue, with batching on from day one, until volume makes a capacity model cheaper.
• Existing JMS applications leaving a commercial broker: Apache Artemis or Red Hat AMQ, which speak the same API.
• Before any migration: fix the ILMT position and right-size the edition. Moving off IBM MQ is expensive precisely because of what it is plugged into.

• “Kafka is free. Why would we pay for Confluent?” (Operations, connectors, governance and an SLA — weighed against the team you would otherwise need.)
• “How would you cut our IBM MQ bill?” (ILMT and sub-capacity, edition right-sizing, consolidation of lightly used queue managers, then standby-replica entitlements.)
• “Price 2,000 messages a second on SQS.” (About 15.5 billion requests a month; batch to cut it ten-fold.)
• “Why does message size change an SQS bill?” (Every 64 KB is a request.)