Azure Event Hubs

Overview

Azure Event Hubs is Microsoft's managed streaming platform, described as ingesting millions of events a second with low latency. An event hub is an append-only distributed log containing one or more partitions, so the model will be familiar to anyone who knows Kafka.

The structure is namespace, then event hub, then partitions, with consumer groups giving each application its own independent view and its own position in the stream.

Kafka protocol compatibility is a first-class feature rather than an afterthought. Event Hubs exposes a Kafka endpoint, and existing applications move by changing configuration rather than code.

Know the boundaries though. The Kafka endpoint works only on standard, premium and dedicated tiers, so the cheapest tier cannot be used for a Kafka migration at all.

Three protocols run side by side, Kafka, AMQP 1.0 and HTTPS, and you can write with one and read with another, which is genuinely useful for mixed estates.

Delivery is at-least-once by design, stated as a core tenet, so consumers may receive events more than once and should be idempotent.

Kafka feature parity is partial and tier-dependent: Kafka Streams and Kafka Transactions are both public preview on premium and dedicated only, and compression is premium and dedicated only, gzip alone.

Retention is time-based per tier, one day on basic, seven on standard, ninety on premium and dedicated, so long replay windows mean paying for a higher tier.

Long-term landing is handled by Capture, which writes streaming data automatically to Blob Storage or Data Lake Storage as Avro, and bypasses throughput unit egress quotas.

Processing is external. Stream Analytics, Functions, Databricks, Data Explorer, Spark and Flink are the named options, and Microsoft states directly that Confluent's licence is why it cannot offer ksqlDB.

One cost line to check when costing a Kafka migration: on standard, the Kafka endpoint is a separate hourly charge, as is Capture.

Key features and capabilities

The same headings are used for every streaming & stream processing entry, so two tools can be read side by side.

What it does

• Namespace, then event hub as an append-only distributed log, then partitions for parallelism
• Consumer groups give each application an independent view with its own offsets
• An event hub maps directly onto a Kafka topic, and a namespace onto a Kafka cluster
• Log compaction is supported, at 1 GB a partition on standard and 250 GB on premium and dedicated

Delivery and ordering guarantees

• At-least-once by design, with the documentation recommending the idempotent consumer pattern
• Idempotent producers and consumers are supported over the Kafka endpoint
• Kafka transactions are supported but in public preview on premium and dedicated only
• Consumer progress is tracked by checkpointing per partition

Scale and retention

• Throughput units give 1 MB/s or 1,000 events a second in, and 2 MB/s or 4,096 events out, up to 40 units
• Premium scales to 16 processing units and dedicated to 10 capacity units, more on request
• Partitions, 32 per event hub on basic and standard, 100 on premium, 1,024 on dedicated
• Retention one day, seven days, or ninety days by tier; message size 256 KB, 1 MB or 20 MB

Processing and transformation

• Capture is the built-in offload, writing to Blob or Data Lake Storage as Avro, or Parquet via Stream Analytics
• Capture reads internal storage directly, bypassing throughput unit egress quotas
• Processing is external, through Stream Analytics, Functions, Databricks, Data Explorer, Spark or Flink
• Kafka Streams works but is public preview on premium and dedicated; ksqlDB cannot be offered for licence reasons

Security and governance

• Microsoft Entra ID with role-based access control, shared access signatures, or managed identities
• Kafka clients authenticate with SASL over TLS, using OAuth 2.0 or shared access signatures
• Private Link, virtual network service endpoints and IP firewall rules, none of them available on basic
• Encryption at rest with Microsoft-managed or customer-managed keys, the latter premium and dedicated only
• Azure Schema Registry for Avro and JSON Schema, with runtime audit logs on premium and dedicated only

Connectors and ecosystem

• Three protocols concurrently, Kafka 1.0 and later, AMQP 1.0 and HTTPS, mixable for read and write
• Client libraries for .NET, Java, Python, JavaScript and Go, plus any Kafka client
• Capture lands data for Synapse, Databricks and Data Explorer without custom code
• Also integrates with Kafka Connect, Spark, Flink, Samza, Storm, Akka Streams and Event Grid

Where it runs and what it costs

• Managed Azure service only, with no self-hosting or Kubernetes operator
• Four tiers, basic, standard, premium with resource isolation, and single-tenant dedicated
• Availability zones on all tiers, geo-disaster recovery from standard, geo-replication on premium and dedicated
• Availability is listed per region rather than as a total count, with a 99.99% commitment

Pricing

Usage-basedBasic at $0.015 a throughput unit hour

In East US, basic costs $0.015 a throughput unit hour and standard $0.030, both with ingress at $0.028 a million events. Premium is $1.027 a processing unit hour and dedicated $6.849 a capacity unit hour, with ingress included on both and extended retention at $0.10 a GB a month. Two standard-tier add-ons are easy to miss when costing a migration: Capture at $0.10 an hour and the Kafka endpoint at $0.09 an hour, both included on premium and dedicated. Event Hubs has no free tier.

Vendor pricing page →

Demos and videos

• Microsoft Azure channel Channel or playlist · YouTube
• Microsoft Learn shows, Event Hubs Webinar or talk · learn.microsoft.com

About Microsoft

Event Hubs is a first-party Azure platform service from Microsoft Corporation, founded in 1975 and headquartered at One Microsoft Way, Redmond, Washington, with roughly 223,000 employees as of June 2026 and listed publicly as MSFT. The service is proprietary, sold under the Microsoft online services terms, with no open-source edition and no self-hosting; the only local artefact is an emulator for developing and testing against the service in isolation. For native Kafka on Azure, Microsoft points instead at HDInsight.

Founded 1975 · Redmond, Washington · azure.microsoft.com