Kafka Alternatives in 2026: AutoMQ, WarpStream, Redpanda, NATS

IBM closed its $11 billion acquisition of Confluent on 17 March 2026, at $31 a share, which delisted Confluent and made WarpStream an IBM product by inheritance. That is the news everyone leads with, and it is the least useful thing to plan around.

The useful thing is that the argument diskless streaming was making has been won so completely that the incumbents now sell it too. Confluent has a diskless cluster type on its price list. Redpanda shipped one. Apache Kafka voted to put one in the protocol. If you are still comparing “Kafka” against “the diskless alternatives,” you are comparing a thing against itself.

Here is the arithmetic that caused all of it, and then what each engine is actually for. Prices below came off the AWS price list and the vendor pricing pages on 21 August 2026.

Where a Kafka bill goes

Take a modest workload: one tebibyte ingested per day, seven-day retention, three replicas across three availability zones. Nothing exotic.

Classic brokers hold three copies on disk. That is 21 TB of gp3, which lists at $0.08 per GB-month in us-east-1, so about $1,720 a month. Then every byte the leader accepts crosses an availability-zone boundary twice to reach its followers, and AWS charges $0.01 per GB in each direction. Two TiB a day of cross-AZ replication over thirty days is roughly $1,229 a month.

Diskless brokers hold one copy in object storage. Seven TiB on S3 Standard at $0.023 per GB-month is about $165. Replication across AZs is S3’s problem and appears on no line of your bill. You do pay for PUTs: WarpStream’s agents buffer for 250 ms or 8 MiB, whichever comes first, and at that batch size a TiB a day is under four million PUTs a month, or about $20 at $5 per million.

About $2,950 against about $185. Broker compute sits on top of both and is roughly comparable, so the ratio on a full bill is smaller than sixteen. It is still large enough that every vendor in this post reorganized around it.

The tradeoff is latency, and it is not uniform across the products, which is where most comparisons go wrong.

AutoMQ

Apache 2.0, on 1.7.3 as of August 2026, and the design choice that matters is that it keeps Kafka’s compute layer and replaces only storage. AutoMQ claims 100% Kafka API compatibility on that basis, which is a stronger claim than it used to make and is the reason Kafka Connect connectors and Streams jobs tend to work unchanged.

It also writes differently from the others. A produce lands in a write-ahead log first, on EBS or NFS or S3 depending on how you configure it, and gets acknowledged there while the upload to object storage happens in the background. WarpStream, by contrast, batches into object storage in the synchronous path. That is why AutoMQ can offer a low-latency mode at all, and why its published p99 figures for the EBS WAL are an order of magnitude below its own S3 WAL numbers. Those figures are the vendor’s own benchmarks, so treat them as a claim about the architecture rather than a measurement of your workload.

This is the one I would reach for on a new high-throughput analytics workload. Permissive license, runs anywhere, and the compatibility story means the migration is operational rather than a rewrite.

WarpStream

Now IBM’s, by way of Confluent, which bought it in September 2024. It still has its own brand, its own console, its own pricing page, and a changelog that has kept moving through 2026.

The pricing is public, which is rare enough here to be worth using: $0.01 per GiB written for the first 50 TiB a month, sliding to $0.003 at very high volume, $0.01 per GiB-month stored, and a cluster fee of $100, $500, or $1,500 a month depending on tier. New accounts get $400 in credits that do not expire.

The awkward part is thesis drift. If you picked WarpStream specifically to avoid depending on Confluent, you now depend on IBM, and that is a different vendor risk than the one you signed up for. If you did not care about that and you wanted BYOC with data staying in your account, nothing about the product got worse.

Redpanda

C++, thread-per-core, on 26.2.1. The received wisdom is that Redpanda is what you pick when you cannot tolerate diskless latency, and that stopped being the whole story in 26.1, which shipped Cloud Topics: durable object storage as the primary backing store, with Redpanda citing elimination of over 90% of cross-AZ replication cost. It is GA.

Read the license line before you plan around it. Cloud Topics requires an Enterprise license. So does Tiered Storage. The Community Edition is BSL rather than open source, with each commit converting to Apache 2.0 four years after it merges. So the free build gives you the fast broker and not the cheap storage, which is a coherent business decision and a thing you want to know before the pilot rather than after.

Pick Redpanda when the streaming layer’s latency is visible to a user: fraud scoring, order books, anything where a few hundred milliseconds changes behavior rather than a dashboard.

Bufstream, and why it is no longer on the list

Bufstream was the interesting Iceberg-native option, Protobuf-first, with every topic queryable as a table and no separate sink job. CoreWeave acquired it on 8 May 2026 and folded it into the W&B Models and Weave product lines. Buf continues independently on the Schema Registry and Protobuf tooling.

The announcement does not say what happens to Bufstream as a standalone product you can buy or self-host. Until that is clear, I would not start a migration onto it. If your end state is a lakehouse, the Iceberg integrations from the other engines are the safer route this quarter.

Apache Kafka itself

Kafka 4.3 landed on 22 May 2026. KRaft has been the only option since 4.0 and ZooKeeper is genuinely gone. Kafka 4.2 made share groups production-ready, which gives Kafka queue semantics with per-message acknowledgment, and that quietly removes one of the standard reasons to run RabbitMQ alongside it.

On diskless: KIP-1150 passed its vote on 2 March 2026 with nine binding votes. It is an umbrella proposal that establishes the direction, not an implementation, and the actual mechanics live in follow-up KIPs. Diskless topics are not in mainline Kafka today. They are, however, clearly coming, which changes how much you should weight “this vendor has diskless and Kafka does not” in a three-year decision.

Self-managed Kafka still wins when you have people who know it cold, when your throughput is low enough that the arithmetic above is a rounding error, or when a regulator has opinions about whose cloud your bytes sit in.

When the answer is not Kafka at all

NATS with JetStream, on 2.14.5, is the right shape for fan-out to many consumers: thousands of subscribers, sub-millisecond pub/sub, small enough to embed. It is not a replacement for an analytics pipeline. It is a much better answer than Kafka for the workload people keep describing as streaming when they mean broadcasting.

RabbitMQ 4.3.5 is still correct for complex routing and workloads where each message is precious. Quorum queues and streams closed a lot of the throughput gap. The reason to reach for it over Kafka’s new share groups is routing topology, not durability.

Apache Pulsar has the best geo-replication and multi-tenancy story and the operational complexity to match. The diskless wave ate most of what tiered storage used to make special about it. I would consider it for genuinely global multi-region streaming with strict tenant isolation, and not otherwise.

Managed options, briefly

MSK Serverless bills $0.75 per cluster-hour, $0.0015 per partition-hour, $0.10 per GB in and $0.05 per GB out. The partition-hour line punishes high partition counts hard, so count your partitions before you model it. MSK Provisioned charges $0.10 per GB-month for storage, which is more than gp3 costs directly.

Confluent Cloud’s own price list is the more interesting document. Basic gives you the first eCKU free then $0.14 an hour, at $0.05 per GB of ingress and egress and $0.08 per GB-month of storage. Standard is $0.75 an eCKU-hour from about $385 a month. And then there is Freight: $2.25 an eCKU-hour with a two-unit minimum, $0.014 to $0.030 per GB, and storage at $0.03 per GB-month, starting around $2,300 a month. Freight is Confluent selling you the diskless economics directly. The per-GB rate is less than a third of Basic’s.

Azure Event Hubs speaks the Kafka protocol and is fine if you are already on Azure. GCP Pub/Sub is the easiest thing here to start with and the hardest to leave, because it is not Kafka-compatible at all. Upstash Kafka is gone: deprecation began 11 September 2024 and service ended 11 March 2025.

How I would decide

If your bill is dominated by EBS and cross-AZ lines, the diagram above is your business case and you do not need a consultant to build it. Pull two numbers off last month’s invoice and you will know within an hour whether this is worth a quarter of someone’s time.

If you are on Confluent and renewing, price Freight against your current cluster type before you price anyone else’s product. The cheapest migration is the one where you change a cluster type and not a client library.

And if you are choosing from scratch, the license text is doing more work than the benchmark. AutoMQ is Apache 2.0. Redpanda’s cheap-storage feature is not in the free build. Those two sentences will shape more of the next three years than any throughput chart, and only one of them appears on a comparison slide.