WarpStream

Editor’s note: This WarpStream review is based on my hands-on test of the WarpStream Agent (v854) in playground mode and a Dev-tier BYOC cluster, in October 2026. I weighed pricing, setup, latency, Kafka compatibility, add-on products, security, and alternatives including Confluent Cloud, Amazon MSK and Redpanda.

Quick verdict: I recommend WarpStream for high-throughput streams where cross-zone networking dominates costs and sub-second consumption is acceptable. Default produce p99 is 500 ms, end-to-end p99 is 900 ms, and consumers stall during a hosted control-plane outage.

WarpStream’s Kafka-protocol-compatible, stateless Agents replace brokers and disks with object storage. Raw records stay in my VPC and buckets; signup includes $400 in non-expiring credits without a card.

Confluent acquired WarpStream on September 9, 2024; IBM completed its roughly $11 billion Confluent acquisition on March 17, 2026. WarpStream retains its own brand, console and pricing, with roughly weekly Agent releases. In this WarpStream review, I’ll take a closer look at WarpStream’s pricing, latency, Kafka compatibility and BYOC model, so you can see exactly whether it beats Confluent Cloud or MSK for your workloads.

Key Takeaways:

• Dev starts at $100/month, plus $0.01 per uncompressed GiB written and $0.01 per uncompressed GiB-month stored for the first 50 TiB.
• Agents write directly to object storage, eliminating inter-AZ broker replication traffic.
• Default produce/end-to-end p99 is 500/900 ms. Lightning Topics reach about 50 ms produce p99 but reject transactions and idempotence.
• Uncompressed-byte billing means 4:1 compression still incurs charges on four times the wire bytes.
• Ripcord protects writes, not reads: producers continue during control-plane outages; consumers wait.

Pros and Cons

WarpStream Pricing: How Much Does It Cost?

Every account starts with $400 in non-expiring credits, without a credit card. There is no permanent free plan.

WarpStream BYOC pricing combines a cluster-tier fee, metered writes and metered storage. My cloud provider separately bills Agent VMs, object storage and storage API requests. WarpStream charges no per-Agent, per-core or per-partition fees, so adding Agents does not create another software charge.

• Dev ($100/mo): for testing and non-critical workloads; no SLA, up to 4,096 partitions and $0.0345 per active 15-minute interval.
• Fundamentals ($500/mo): for production requiring a 99.9% SLA and up to 16,384 partitions. Adds topic recovery, broker-level schema validation and a lower-latency control plane; $0.1725 per interval.
• Pro ($1,500/mo): for a 99.99% SLA and up to 50,000 partitions. Adds PrivateLink and user activity logs; $0.52083 per interval.
• Enterprise (custom): for 200,000+ partitions and a dedicated control-plane cell. The SLA is 99.99%, rising to 99.999% with Multi-Region Clusters.

Plan	Monthly cluster fee	SLA	Max partitions	Notable gated features
Dev	$100	None	4,096	Agent Groups, mTLS, Lightning Topics
Fundamentals	$500	99.9%	16,384	Topic recovery, schema validation, lower latency
Pro	$1,500	99.99%	50,000	PrivateLink, user activity logs
Enterprise	Custom	99.99%; 99.999% with Multi-Region	200,000+	Dedicated control-plane cell, Multi-Region

Kafka writes cost $0.01 per uncompressed GiB for the first 50 TiB monthly, with marginal rates falling to $0.003 above 7,800 TiB. Storage starts at $0.01 per uncompressed GiB-month for the first 50 TiB, falling to $0.004 above 1,800 TiB.

Tableflow writes start at $0.04 per uncompressed GiB. Schema Registry clusters pay only cluster-minute charges.

Cluster fees accrue in 15-minute intervals receiving requests, rather than as an unconditional monthly subscription.

Is WarpStream Good Value for Money?

My arithmetic for a Dev cluster ingesting 1 TiB monthly with seven-day retention is approximately $100 for the cluster, $10.24 for writes and $2.39 for storage: roughly $113/month before cloud infrastructure. That assumes uniform ingestion and steady-state retention over a 30-day month.

Amazon MSK Express’s example costs about $1,021 for three express.m7g.large brokers ingesting and storing 1 TB for 31 days. Confluent Cloud Standard charges $0.75 per eCKU-hour, with data transfer rates of $0.035-$0.050/GB.

• Compression does not reduce metered volume: at 4:1 compression, one wire byte represents four billed bytes.
• Storage requires two budget lines: WarpStream’s $0.01/GiB-month fee excludes my bucket bill; MSK Express storage is $0.10/GB-month.
• Production starts at $500/month for me: Dev’s missing SLA outweighs its lower cluster fee for critical streams.

Reviewer’s Notes: I recommend Dev for the credit-funded trial and staging, Fundamentals as the production floor, and Pro when I need 99.99% availability, PrivateLink, user activity logs or more than 16,384 partitions.

My Experience With WarpStream

Starting With the Playground

warpstream playground starts a local Agent against the real hosted control plane, with a temporary account valid for a few hours. I installed the Agent CLI with the install script; a binary download and the Docker image are the other supported options.

warpstream playground
warpstream kcmd -type diagnose-connection -bootstrap-host localhost -bootstrap-port 9092

I ran the diagnostic command in a second terminal while the playground process remained active. Kafka listens on localhost:9092, and Schema Registry uses port 9094. I pointed a Kafka client at that local endpoint for the produce-and-consume flow.

Playground data stays in memory and disappears when the process exits. For CI, warpstream local substitutes an in-memory control plane; both modes are heavily rate-limited, so neither belongs in a throughput or latency comparison against MSK.

Creating a BYOC Cluster

For persistent storage, I moved to console.warpstream.com, where signup includes the $400 credits without a card. I created a virtual cluster and chose its tier and control-plane region.

I used Dev for the initial compatibility exercise, then created a dedicated S3 bucket and a VPC endpoint. The endpoint keeps Agent-to-S3 traffic off the NAT gateway path, avoiding an unnecessary data-processing charge.

Agents write beneath a warpstream prefix, and manually deleting those files can make the virtual cluster unusable. I kept manual cleanup away from that prefix rather than treating Agent-managed objects as disposable application files.

Deploying the Agents

WarpStream uses one stateless binary, with these deployment options:

• Docker: run the Agent in a container.
• Kubernetes: deploy through Helm charts.
• Terraform: use the provider and modules for infrastructure configuration.

I passed the bucket URL and supplied the Agent key through WARPSTREAM_AGENT_KEY. The key authenticates the Agent to WarpStream; the deployment also needs cloud permissions to access its bucket.

Agent scaling follows CPU or network demand, rather than disk utilization. Compared with operating Kafka brokers, I no longer tied partition placement to individual broker disks, but I still owned VM capacity and network configuration. Confluent Cloud is the better fit when I want to avoid managing that infrastructure.

Connecting Kafka Clients

I pointed clients at the Agent endpoint rather than a Kafka broker fleet. The officially supported choices are:

• librdkafka: the native client used by several language bindings.
• franz-go: the supported Go client.
• Java client: the standard Kafka Java implementation.

The general client-tuning path uses linger around 100 ms to batch records efficiently. I separated that throughput-oriented setting from the smaller values used for lower latency.

Agent roles and Agent Groups separate workloads inside one logical cluster. Dedicated deployments keep pipeline processing or background jobs from sharing every Agent with Kafka traffic.

What Changed Compared With Running Brokers

WarpStream removes broker disk sizing and partition rebalancing from this workflow. Retention follows record arrival time, not the producer’s CreateTime.

Compaction also differs: the first 32 L0 files compact into L1 without key deduplication. Deduplication begins at later levels, so I included compacted-topic behavior in the migration checks.

Reviewer’s Notes: Start with playground client-compatibility checks, then mirror one non-critical topic with Orbit into a persistent Dev cluster before considering a production cutover.

Latency, Lightning Topics and Kafka Compatibility

WarpStream’s default configuration delivers 250 ms produce p50 and 500 ms p99, with 500 ms end-to-end p50 and 900 ms p99. That configuration uses S3 Standard, a 250 ms Agent batch timeout, 25 ms client linger and Fundamentals.

I shortlist Redpanda or broker-based Kafka first for latency-critical request paths. WarpStream’s object-storage batching better suits streams whose cost matters more than immediate consumption.

Tuning for Lower Latency

• Reduce client linger: use 25 ms or 10 ms rather than the throughput-oriented 100 ms setting.
• Reduce Agent batch timeout: -batchTimeout defaults to 250 ms and supports values down to 25 ms. More frequent flushes increase object-store PUT costs.
• Choose Fundamentals or Pro: both offer the lower-latency control plane that Dev lacks.
• Use S3 Express for ingestion: compact into a standard bucket afterward. S3 Express storage costs roughly seven times standard storage; the mixed configuration adds about 20% to infrastructure costs on average.

With 10 ms linger, 50 ms batch timeout and S3 Express, produce latency falls below 80 ms p50 and 150 ms p99. End-to-end latency falls below 200 ms p50 and 400 ms p99.

Lightning Topics

Setting warpstream.topic.type=lightning requires Agent v743 or later. With S3 Express, Lightning Topics support under 35 ms produce p50 and approximately 50 ms produce p99 by journaling writes before asynchronously committing them to the control plane.

The faster acknowledgment does not reduce end-to-end latency. I reject Lightning Topics for producers that depend on the missing behaviors:

• Transactions and idempotence: both are rejected.
• Returned offsets: ProduceResponse offsets are always zero.
• External consistency: acknowledgment order does not guarantee visibility order across batches.
• GCP Rapid Buckets: this storage option is incompatible with Lightning Topics.

Compatibility Gaps

Standard topics support transactions and exactly-once semantics with enable.idempotence=true and a non-empty transactional.id. Standard-topic support also includes:

• Access and storage: ACLs and compacted topics.
• Consumer groups: KIP-848.
• Client metrics: KIP-714.

Check these protocol differences before migration:

• Tagged fields: ignored.
• Throttling settings: not enforced.
• Request timeouts: capped at 15 seconds, except JoinGroup and SyncGroup.
• Retention: based on arrival time rather than client timestamps.

For latency-critical workloads such as fraud scoring and transactional request/response buses, I recommend disk-based Kafka, Redpanda or MSK Express instead of WarpStream. Logs, telemetry, CDC and lake feeds are stronger WarpStream candidates.

Tableflow, Orbit, Pipelines and Schema Registry

Tableflow

Tableflow materializes Iceberg tables from any Kafka-compatible source, not just WarpStream topics. It handles schema evolution, compaction and table maintenance inside the Agent deployment, with raw data remaining in my account.

Its tables are managed, with no external writers. I recommend it for a read-oriented lake feed, not a table that another application must update independently.

Supported query and catalog integrations include:

• Snowflake and Databricks: warehouse and lakehouse access.
• BigQuery and Athena: cloud analytics.
• DuckDB and ClickHouse: alternative query engines.
• AWS Glue: catalog integration.

Writes start at $0.04 per uncompressed GiB, four times WarpStream’s first-tier Kafka write rate. Confluent Tableflow serves the corresponding Confluent Cloud use case; I prefer WarpStream’s version when the source or storage must remain in my infrastructure.

Orbit and Auto Migration

Orbit copies records with identical offsets from Kafka-compatible sources into WarpStream. It also replicates:

• Topic and cluster configurations.
• Consumer group offsets.
• Kafka ACLs.

Orbit runs inside Agents assigned the jobs role. Auto Migration, launched August 18, 2026, moves producers topic by topic without client restarts; v853 added multi-source-cluster support.

I value preserved offsets over simple record copying because a migration must preserve consumer positions as well as payloads.

Managed Data Pipelines

Managed Data Pipelines embed Bento, the Go stream processor forked from Benthos, inside Agents. Pipeline processing can increase Agent load enough to disrupt production traffic.

I recommend a dedicated Agent group with the pipelines role, rather than sharing compute with latency-sensitive Kafka traffic. Running inside my account also puts that processing capacity on my cloud bill.

Schema Registry

Schema Registry is built into the Agent binary and billed on cluster minutes only. It supports most Confluent Schema Registry APIs, but not:

• Data contracts.
• /mode and /exporters endpoints.
• Subject aliases and compatibility groups.

Broker-level schema validation starts at Fundamentals and supports Avro and JSON Schema.

WarpStream also introduced an MCP server for Claude and Cursor on April 16, 2026.

BYOC Security, the Control Plane and Uptime

WarpStream BYOC keeps raw records in my VPC and buckets, but sends operational metadata to its hosted control plane. That is a narrower boundary than putting the data plane in Confluent Cloud, not complete self-management.

The control plane receives:

• Topic names and configurations.
• File metadata: bucket names and compressed/uncompressed sizes.
• Record offsets and timestamps: never record keys or values.
• Consumer group information: names, configurations and offsets.
• Client and producer identifiers: including producer epochs and sequence numbers.

Metadata stays in the selected control-plane region, encrypted in transit and at rest. Keep sensitive values out of topic and consumer group names.

WarpStream positions BYOC for PHI, PCI and GDPR special-category data. WarpStream describes a signed BAA as generally unnecessary for HIPAA workloads in BYOC, with the requirement depending on the customer’s compliance program.

Security controls include:

• Kafka access controls: ACLs, SASL/PLAIN, SASL/SCRAM-SHA-512 and SASL/OAUTHBEARER.
• Transport protection: TLS and mTLS.
• Console access: SAML SSO; user activity logs start at Pro.
• Agent authentication: workload identity federation.
• Build artifacts: FIPS 140-3 builds and SBOMs since v853.

Ripcord keeps Produce requests flowing during a control-plane outage by journaling to object storage and committing later. Consumers remain unavailable until coordination returns.

WarpStream operates control planes in 24 regions across three clouds. Enterprise Multi-Region Clusters, available since July 2025, use two or three control-plane regions; paired with multi-region data-plane quorum writes, they provide RPO=0.

Multi-Region must be selected at cluster creation. Enterprise’s 99.999% SLA requires this configuration; without it, the tier provides 99.99%.

Contact-sales options include:

• Dedicated single-tenant control-plane cells.
• Private networking.
• A self-managed control plane.

The Agent is not published as open source. AutoMQ’s Apache 2.0 license offers a different option for open-source requirements; I do not assume a comparable fork-and-self-run exit from WarpStream.

How Does WarpStream Compare to Competitors?

WarpStream’s $500/month Fundamentals tier buys a 99.9% SLA while leaving infrastructure in my account. Confluent Cloud removes that infrastructure responsibility; MSK Express retains hourly broker charges.

• Confluent Cloud: I recommend it when fully hosted infrastructure matters more than BYOC. Standard costs $0.75/eCKU-hour with $0.08/GB-month storage. Freight costs $2.25/eCKU-hour with a two-eCKU minimum and $0.03/GB-month storage, accepting roughly 1-2 seconds of latency for relaxed workloads. Freight is a closer buying alternative for latency-tolerant streams, not a low-latency upgrade.
• Amazon MSK (Express brokers): I shortlist it for AWS-only deployments that want managed Kafka brokers. The express.m7g.large example uses $0.408 per broker-hour, $0.01/GB ingested and $0.10/GB-month storage. Broker charges continue when idle.
• Redpanda: I shortlist it for latency-sensitive applications. Serverless carries a 99.9% SLA, while BYOC offers 99.99% and a 14-day trial. BYOC requires an annual commitment and a quote, unlike WarpStream’s public cluster-tier prices.
• AutoMQ: I include this Apache 2.0-licensed alternative when an open-source option is a procurement requirement. WarpStream’s Agent is not published as open source.
• Bufstream: Its write rate is about $2 per TiB of uncompressed writes, without storage, read, per-core or per-Agent fees. I compare complete infrastructure requirements before treating that write rate as the total cost.
• StreamNative: I shortlist its Ursa engine for deployments needing both Pulsar and Kafka protocols. Serverless starts at $73/month and BYOC at $365/month, with usage charges beyond those entry figures.

Tool	Architecture	Runs in my account?	Entry price	Best for
WarpStream	Stateless Agents, object storage	Yes	Dev $100/month + usage	Latency-tolerant BYOC streams
Confluent Cloud	Fully hosted Kafka service	No	Standard $0.75/eCKU-hour + usage	Hosted infrastructure
Amazon MSK Express	Managed Kafka brokers	AWS-managed	$0.408/broker-hour + usage	AWS-native Kafka
Redpanda BYOC	Kafka-compatible brokers	Yes	Quote	Latency-sensitive shortlist
AutoMQ	Diskless Kafka, Apache 2.0	Yes	Quote	Open-source exit path
Bufstream	Kafka-compatible, object storage	Yes	About $2 per TiB of uncompressed writes + infrastructure	Lowest published write rate

Compare the same retention window and uncompressed volume across these options. MSK’s broker hours, Confluent’s capacity units and WarpStream’s cluster fee need their usage and infrastructure charges added before the totals mean anything.

WarpStream earns its place on my shortlist when removing broker replication traffic matters more than minimizing acknowledgment latency. It does not win automatically on a small or highly compressed stream.

How I Reviewed WarpStream

I evaluated Agent v854, released October 28, 2026, through the playground-to-BYOC setup path. I used the CLI connection check, selected a Dev-tier virtual cluster and configured S3-backed Agents, with Kafka clients pointed at the Agent endpoint.

My evaluation priorities were:

• Pricing and value: cluster fees, uncompressed metering, retention and separate cloud infrastructure.
• Setup: playground behavior, persistent storage, Agent deployment and supported client choices.
• Latency and compatibility: batch settings, transactions, Lightning Topics and protocol differences.
• Add-on products: Tableflow’s writer restrictions, Orbit’s preserved offsets, pipelines and Schema Registry.
• Security and reliability: metadata exposure, Ripcord, control-plane dependencies and tier-specific SLAs.

I compared WarpStream with Confluent Cloud, MSK Express and Redpanda. My cost calculation uses a Dev cluster with 1 TiB of monthly writes and seven-day retention; my production recommendation instead starts at Fundamentals because Dev has no SLA.

Prices current as of October 2026.

Should You Move Your Kafka Workloads to WarpStream?

I recommend WarpStream for multi-TiB logs, telemetry, CDC feeds and Iceberg ingestion when sub-second consumption is acceptable and broker networking dominates costs. Budget for Fundamentals at $500/month as my recommended production floor, then add usage and cloud infrastructure.

Skip WarpStream for transactional paths requiring sub-100 ms latency. Lightning Topics lower produce latency by dropping transactions and idempotence, not by making standard topics equally fast; Redpanda or MSK Express deserves the first evaluation there.

I also reject the default hosted-control-plane model if consumer stalls during an outage violate the workload’s requirements.

IBM’s ownership of Confluent makes roadmap and commercial continuity questions worth raising with sales. WarpStream retains its own console and pricing; ask how that product boundary will evolve before a major migration.

Use the $400 credits on a Dev cluster mirroring one real topic with Orbit. Compare one month’s total charges with the current Kafka invoice, including:

• WarpStream fees: cluster activity, writes and storage.
• Agent compute: the capacity running in your account.
• Object storage: retained data and API requests.

Start that trial at warpstream.com.

FAQ

Is WarpStream free?

WarpStream offers $400 in non-expiring credits, not a permanent free plan. No credit card is required to start, and Dev begins at a listed $100/month plus usage. Your cloud provider bills BYOC infrastructure separately. The free playground suits local compatibility checks, but its rate limits make it unsuitable for benchmarking.

Is WarpStream owned by Confluent?

Yes. Confluent acquired WarpStream on September 9, 2024, and IBM completed its roughly $11 billion acquisition of Confluent on March 17, 2026. WarpStream continues under its own brand, console and pricing. Raise long-term roadmap and commercial commitments with sales before making it a central streaming dependency.

Is WarpStream fully compatible with Kafka?

WarpStream supports the Kafka protocol, but not every Kafka behavior. Standard topics support transactions, ACLs and KIP-848 consumer groups. Tagged fields are ignored, throttling is not enforced, and retention follows arrival time. Request timeouts cap at 15 seconds except JoinGroup and SyncGroup. Lightning Topics additionally reject transactions and idempotent producers.

How fast is WarpStream?

Default latency is 500 ms produce p99 and 900 ms end-to-end p99. Tuned S3 Express configurations bring produce p99 below 150 ms and end-to-end p99 below 400 ms. Lightning Topics reach about 50 ms produce p99 but do not reduce end-to-end latency. Choose the configuration against your consumption deadline, not acknowledgment speed alone.

Does WarpStream work outside AWS?

Yes. WarpStream BYOC supports Amazon S3, Google Cloud Storage and Azure Blob Storage. Deploy Agents using Docker, Kubernetes Helm charts or Terraform. Billing is available directly and through AWS or Google Cloud Marketplace. Storage-specific features differ: Lightning Topics, for example, are not compatible with GCP Rapid Buckets.