Home › Learn › Data systems › Data platform › Cost › Calculator

Free tool · Databricks on AWS · list prices as of 2026-11-09

Databricks data platform cost calculator

What a Databricks data platform costs end to end, hop by hop from sources to dashboards and agents, with the streaming lane, real-time serving and the hidden costs most estimates leave out. Answer eight questions; open a hop only if you want to change more. The services around a message stream, from devices to live screens, have their own estimate below.

Eight answers

Whole platform, end to end

·

Where the money goes

Every line, and how it was worked out

Open a line to see the numbers behind it. Lines marked "hidden cost" are the ones that rarely make it into a plan.

LineKind / hopPer month

What matters

What drives your cost

Each row changes one setting both ways and keeps everything else the same. The longest bars are the settings worth getting right; the short ones you can leave at their defaults.

Spend less

Ways to spend less, for your settings

ChangeSaves per month

More detail

Adjust each hop

Every hop has its own settings. "Show expert settings" adds the rest, including the work rates the model uses, so you can replace them with numbers from your own system tables.

Streaming services

Streaming services: from the device to the screen

The whole-platform estimate above treats streaming as tables that refresh often. A stream of messages from machines, vehicles or apps also needs services of its own: a front door that accepts the messages, an event log that keeps them in order for a few days so several programs can read them, a live lane that raises alerts and keeps live screens up to date within seconds, and a platform lane that lands the same messages in lake tables every few minutes. This estimate prices those services for two stacks: everything on AWS, and AWS with Databricks for the platform lane. The live lane is the same in both. It is a separate estimate and is not added to the whole-platform total.

The design behind it, with the reasons for each choice, is in the streaming guide; its cost section has the same production example worked by hand.

Rough US-region list prices from 2025 and 2026, not quotes. Prices change often. Check the AWS Pricing Calculator and Databricks' pricing pages before you plan a budget. People and support costs are not included.

Your stream

"Show expert settings" above adds the finer settings here too.

Streaming services, per month

AWS-only
AWS + Databricks

Every line, side by side

Open a line to see the numbers behind it.

LineAWS-onlyAWS + Databricks

What moves this bill

Each row changes one setting and keeps everything else the same.

If you changeAWS-onlyAWS + Databricks
  1. Prices per message (IoT Core, S3 write requests, DynamoDB writes) dominate as the stream grows. Bundle readings at the edge, and send only changes to the live store and to screens.
  2. Compute that is always on (Flink, or a Databricks real-time cluster) is most of the live lane. Keep the live lane to what someone acts on within minutes.
  3. How often the platform lane runs sets the Databricks part, because every run pays a fixed start and checkpoint cost. Every 30 minutes costs about half of every 15 minutes; every 5 minutes about three times as much.
  4. How long the event log keeps messages. Seven days is cheap; history belongs in lake tables.
Unit prices used in this section
PriceValue

Databricks DBU rates and the S3 prices come from the price table below and follow the plan and discount chosen at the top. The rest are rough list prices for US regions, kept in the calculator's script.

How it works

How the estimate is made

Each hop turns its data volumes into work, measured in classic DBU-hours: reading new data, merging changes (plus rewriting the files those changes land in), refreshing gold, pre-computing aggregates. The compute you choose turns work into dollars. Serverless multiplies by one DBU price that includes the machines. Classic multiplies by a lower DBU price and adds the EC2 machines, start-up time and network. SQL warehouses are priced by the hours they are on, which depends on users, schedules and auto-stop, not on the number of queries.

The work rates are calibrated defaults, not measurements of your system. Treat the result as a planning estimate (the range shown is roughly minus a third to plus a half) and confirm it against system.billing.usage once you run. The Cost & FinOps page explains the drivers and hidden costs in words.

Prices used

Prices used, and how sure we are

checked: checked against a published source on the date shown. docs: a long-standing published price, not re-fetched automatically yet. estimate: our estimate. All prices are list prices for AWS US East (N. Virginia), in US dollars.

Databricks productPremium $/DBUEnterprise $/DBUStatus
Jobs compute (classic)$0.15$0.20checked 2026-11-09source
Serverless jobs and pipelines
Includes the cloud machines. Also the rate for materialized view and streaming table refreshes, predictive optimization and data quality monitoring (automated serverless).
$0.35$0.45checked 2026-11-09source
Declarative pipelines Core (classic)
Enterprise rate not confirmed; Premium value used.
$0.20$0.20checked 2026-11-09source
Declarative pipelines Pro (classic)
Needed for change data capture (APPLY CHANGES / AUTO CDC). Enterprise rate not confirmed.
$0.25$0.25docs source
Declarative pipelines Advanced (classic)
Needed for expectations (data quality rules). Enterprise rate not confirmed.
$0.36$0.36docs source
All-purpose compute (classic)$0.55$0.65checked 2026-11-09source
Serverless notebooks$0.75$0.95checked 2026-11-09source
SQL Classic$0.22$0.22checked 2026-11-09source
SQL Pro$0.55$0.55checked 2026-11-09source
SQL Serverless
Includes the cloud machines.
$0.70$0.70checked 2026-11-09source
Model serving (CPU/GPU endpoints)$0.07$0.07docs source
Other priceValueStatus
Photon DBU multiplier
Databricks' Jobs pricing page lists a Photon DBU emission rate of 2.5x or 2.9x depending on the cloud. Photon usually also finishes SQL-heavy work 2-3x faster, so the net effect is often close to even.
2.5x (range 2.5x to 2.9x)checkedsource
Serverless standard mode
Databricks says Standard mode is 'up to 70% cheaper for some workloads' than Performance Optimized. We use 25% cheaper as a cautious default.
25% cheaper than performance modeestimatesource
SQL warehouse DBU per hour
DBUs per hour per cluster; the same for Classic, Pro and Serverless. Serverless adds clusters when many queries run at once.
2X-Small 4, X-Small 6, Small 12, Medium 24, Large 40, X-Large 80, 2X-Large 144, 3X-Large 272, 4X-Large 528docssource
EC2 on-demand prices
On-demand Linux, US East (N. Virginia). DBU per hour for instances other than m5d.2xlarge are scaled estimates; the exact figure is on Databricks' pricing calculator.
m5d.xlarge $0.226/h (0.69 DBU/h), m5d.2xlarge $0.452/h (1.37 DBU/h), m5d.4xlarge $0.904/h (2.74 DBU/h), r5d.2xlarge $0.576/h (1.8 DBU/h), i3.2xlarge $0.624/h (2.0 DBU/h), c5d.2xlarge $0.384/h (1.5 DBU/h), r6id.2xlarge $0.605/h (2.0 DBU/h), m7gd.2xlarge $0.427/h (1.37 DBU/h)checked 2026-11-09source
AWS s3 gb month$0.023docssource
AWS s3 put per 1000$0.005docssource
AWS s3 get per 1000$0.0004docssource
AWS nat gateway hour$0.045docssource
AWS nat gateway gb$0.045docssource
AWS internet egress gb$0.09docssource
AWS cross region gb$0.02docssource
Vector search unit hour
One standard vector search endpoint unit kept running, per hour.
$0.28estimatesource
Llm cost per agent question
Tokens for one agent question with retrieval, on a pay-per-token foundation model. Varies 10x by model and prompt size.
$0.01estimate

Prices are refreshed by a monthly job that opens a pull request: EC2 prices come from AWS's public price list, and Databricks prices from the system.billing.list_prices table when a workspace is connected. Check Databricks' pricing pages before you commit to a budget.

Next

Where to go next

Understand the bill: Cost & FinOps: end to end and per hop.

The whole platform: The data platform, whole.