Serverless Kafka without the bandwidth bill

A serverless Kafka cluster on ntail is virtual: no brokers to name, no partitions to plan, no capacity to reserve months ahead of traffic that might never arrive. Describe what you need and it's live in seconds — a Kafka endpoint your existing clients connect to straight away.

Because it's virtual, it never becomes a bill you didn't plan for. Every cluster runs against a throughput quota, so a bad deploy or a burst you didn't see coming hits a ceiling rather than your invoice. Need more room? Higher limits on request, live in minutes — not a capacity-planning meeting.

Three reasons, and none of them is a loss-leading trial price.

We don't run on an expensive public cloud. Our capacity isn't rented by the hour from a hyperscaler, so there's no cloud margin stacked underneath our price — the single biggest reason managed Kafka costs what it does everywhere else.

We built our own Kafka-compatible platform, engineered for cost. Owning every layer of the stack means nothing is paid for twice and nothing is sized for someone else's workload. We run it lean, and every efficiency we find goes straight into the price instead of into someone else's margin.

We don't resell bandwidth. Moving data is the single largest line on most streaming invoices. We don't charge for it at all, and that alone is where most customers find the bulk of their saving.

Yes — worth being precise about what that covers. Data in and data out are billed, at the per-MiB/s rate on the price list; that's your streaming throughput, metered like anywhere else. What's free is the separate toll most providers charge on top of that for moving those bytes across the public internet to and from your applications — sometimes labelled ingress and egress on other vendors' invoices. On ntail that line is $0, at any volume, wherever the traffic comes from.

There's no fair-use asterisk and no overage tier waiting further down the page. You're billed for data in, data out and storage. That's the entire invoice.

Compressed. We meter the compressed bytes as they arrive on the wire, not the raw size of your records — so the compression your producers already do works in your favour rather than ours.

For typical JSON or Avro payloads with a standard codec that is commonly 3–5× less than the uncompressed volume, which means the MiB/s you are billed for is usually a good deal smaller than the number your application thinks it is sending. Storage is metered the same way.

Pay-as-you-go, always. Everything is billed per started hour, so a cluster you run for a day costs a day. Storage bills on what you actually retain, data in and out on what you actually move, invoiced monthly in arrears.

No annual commitment, no reserved capacity to buy up front and no cluster fee. The only floor is a per-cluster minimum of 2 MiB/s of data transfer, in and out added together — $50 a month, about 7 ¢ an hour.

Yes, and it isn't a trial. The free plan gives you a managed cluster with 200 KiB/s of data in, 400 KiB/s of data out, 1-day retention and up to 10 topics, for $0 a month, with no credit card and no time limit. Because the limits are measured after compression, that's about 1,000 typical JSON events a second.

When you outgrow it, upgrade to the paid plan from the console and your clusters stay where they are. More about the free Kafka cluster.

As long as you like — retention is unlimited. Keep a topic's data for a week or for years, at the same per-TB storage rate the whole time.

There's no separate "tiered storage" product to adopt and no migration when you outgrow a retention window — just change the setting.

Production clusters carry a contractual 99.9% monthly uptime SLA — under 44 minutes of downtime a month — with service credits if we miss it.

Every cluster is spread across independent failure domains, and because compute holds no permanent state, a lost node is replaced in seconds rather than rebuilt over hours. Live status is always on our status page.

A single cluster sustains gigabytes per second, and capacity changes land in under a minute instead of during a maintenance window — so a launch, a backfill or a seasonal peak doesn't need a capacity plan filed a month ahead.

There are no brokers to size and no capacity to reserve up front — higher limits on request — and every cluster ships with throughput and consumer-lag dashboards so you can see exactly what it is doing.

Yes. ntail works with the streaming clients, connectors and analytics tools your team already uses — Spark, Flink, Snowflake, Databricks, ClickHouse, Iceberg and the rest.

Switching is a connection-string change and a credential, not a rewrite. Our team runs the migration with you at no cost, and you can move one consumer at a time with a rollback at every step.

No. There's no proprietary client to adopt, no annual contract to escape, and no public internet fee on the way out — which is usually the thing that makes leaving a streaming provider expensive.

Export everything you've stored whenever you want, and pay nothing for the transfer.

In the region you choose. ntail runs in both the EU and the US: pick the region when you create a cluster and your data is stored and processed there, encrypted in transit and at rest, on infrastructure we operate ourselves rather than rent from a public cloud.

Data residency is contractual, and ntail is operated by Factual Tech AB, a Swedish company, so GDPR is built in rather than bolted on. For regulated workloads we'll run a dedicated cluster, or run fully off the public cloud so the data never leaves the boundary you choose. See security for the full detail.