Home › Learn › Real-time analytics › What it costs
Real-time analytics · page 10 of 11
What real-time analytics costs
Rough monthly costs of real-time analytics stacks and unit prices for Event Hubs, Kinesis, MSK, Confluent, Flink, ClickHouse, Fabric and SignalR, plus what drives the bill.
In plain words
Real time is always on
A nightly job runs for an hour and switches off. A real-time pipeline runs every minute of every day: the event log, the processor and the serving store never sleep. That is why real time usually costs three to ten times as much as the batch version of the same report, and why the question "who acts, and how fast" matters.
These are rough list prices remembered from 2025 and 2026, not quotes. Prices change often; check each vendor's pricing page before you plan a budget.
Whole stacks
Small stacks, per month
| Stack | Rough monthly cost | Verdict |
|---|---|---|
| Near real time: change capture into Snowflake or Databricks every 5 to 15 minutes, Power BI refresh | $500–2k | Pursue when people are the ones acting |
| Azure-native real time: Event Hubs, Stream Analytics, Eventhouse or Azure Data Explorer, SignalR | $1k–2.5k | Pursue on Azure |
| Managed best of breed: Confluent, Flink, ClickHouse Cloud, SSE | $2k–6k | Pursue when the product must not depend on one cloud |
| Open source on your own machines | $1k–3k for machines, plus one or two engineers to run it | Drop unless you already have a platform team |
| Enterprise scale, 50 to 100+ MB a second | $20k–100k+ | Only with a clear revenue case |
Unit prices
Rough unit prices by layer
| Layer | Option | Rough price | Note |
|---|---|---|---|
| Capture | Debezium | Free; machines for Kafka Connect about $100–300 a month | You run it |
| Capture | Fivetran, Striim | Priced by rows changed; often $1k–10k+ a month | Managed; grows with change volume |
| Event log | Azure Event Hubs Standard | About $22 a month per throughput unit, plus about $0.03 per million events | Cheap at small scale |
| Event log | Event Hubs Premium or Dedicated | About $700+ a month per processing unit; Dedicated about $5k+ a month | Isolation and longer retention |
| Event log | AWS Kinesis | About $11 a month per shard, plus data | |
| Event log | AWS MSK, three small brokers | About $500–800 a month plus storage | You still tune it |
| Event log | Confluent Cloud | Small production about $1k–3k a month; dedicated clusters $5k–20k+ | Best tooling, highest price |
| Event log | Kafka on object storage (WarpStream, AutoMQ, diskless Kafka) | Often 50–80% cheaper than classic Kafka | Hundreds of milliseconds of delay; fine for near real time |
| Process | AWS Managed Flink | About $80 a month per processing unit; a small job about $160–400 | |
| Process | Azure Stream Analytics | About $80 a month per streaming unit | SQL only, Azure only |
| Process | Databricks streaming | A small always-on cluster about $500–2k a month | Licence plus machines |
| Serve | ClickHouse Cloud | Development under $100 a month; production about $500–3k | Strong price for performance |
| Serve | Fabric capacity | F2 about $260 a month; F64 about $8.4k a month | One capacity covers all Fabric work |
| Serve | Azure Data Explorer | Development about $150–250 a month; production $1k+ | |
| Serve | StarTree (Pinot), Imply (Druid) | Enterprise contracts, usually $5k+ a month | Thousands of users at once |
| Push | Azure SignalR, Web PubSub | Free for 20 connections; about $50 a month per 1,000 connections |
The bill
Five things that drive it
- Always-on compute. Nothing pauses.
- Network traffic between zones. Kafka copies every event to three zones, which can be 30 to 50 percent of the bill. Let consumers read from a copy in their own zone, or use Kafka on object storage.
- Retention. Days of events on the log's disks. Move older events to the lakehouse.
- Connectors and partitions. Managed services charge per connector task or per partition.
- People. Running Kafka and Flink yourself is the biggest hidden cost.
FAQ
Questions people ask
How much does real-time analytics cost?
Roughly, a small near-real-time stack runs $500–2k a month, a small Azure-native real-time stack $1k–2.5k, a managed best-of-breed stack $2k–6k, and enterprise scale $20k and up. These are rough list prices; check current pricing.
Why is real-time analytics expensive?
Everything runs all day and night, Kafka copies every event across zones, events are kept on disk for days, managed services charge per connector and partition, and running it yourself needs engineers.
How can I cut the cost of Kafka?
Read from a replica in the same zone, keep short retention and move history to the lakehouse, use fewer partitions, or use Kafka on object storage where a few hundred milliseconds of delay is fine.
Next
Where to go next
Next page: Which real-time technologies are established
↑ All of real-time analytics · Data platform architecture · Key concepts