Databricks Pricing Calculator: Azure Databricks Pricing and DBU Cost
No seats, no tiers you can budget from a brochure. Databricks bills a unit of compute called a DBU, and half the published rates quietly exclude the cloud bill underneath them. Every number below came off Databricks and Microsoft this week.
Verified 20 August 2026 from databricks.com and Microsoft's retail price API · nothing estimated · Last updated August 2026
Ask your data a question:
Writing SQL… Running (read‑only)… SQL Agentsql wrote
▋
Click a question. Agentsql writes the SQL, runs it read-only, and answers.
Direct answer
Databricks charges per DBU, a unit of processing billed by the second, and sells no seats at all. On AWS Premium in US East the published rates are $0.22 per DBU for SQL Classic, $0.55 for SQL Pro, $0.70 for SQL Serverless, $0.15 for Lakeflow Jobs Classic, $0.35 for Jobs Serverless, $0.55 for all-purpose classic compute and $0.75 for all-purpose serverless. The detail that breaks most budgets is that serverless rates include the cloud instance cost and classic rates do not, so the real gap between $0.22 and $0.70 is an implied instance allowance of $0.48 per DBU hour rather than a 3.2x price difference. Azure Databricks is priced by Microsoft, not by Databricks, and its full rate card is published in a public API. All figures read on 20 August 2026.
›_ databricks pricing, the published rates
Databricks sells compute, not licences. Here is every rate that matters.
AWS, Premium plan, US East (N. Virginia), read on 20 August 2026. Databricks injects these prices with JavaScript, so they are invisible to most scrapers and to the guides that copy each other.
| SKU | Rate per DBU | Cloud instances | What you are buying |
|---|---|---|---|
| SQL Classic | $0.22 | Billed separately by your cloud provider | Interactive SQL exploration on a self-managed warehouse. The cheapest published DBU rate for BI queries, and the one every third-party guide quotes. |
| SQL Pro | $0.55 | Billed separately by your cloud provider | Adds Predictive I/O, materialized views, streaming tables, query federation, Python UDFs and Genie. 2.5x the Classic rate. |
| SQL Serverless | $0.70 | Included in the rate | Fully managed, elastic, high concurrency. Databricks hosts the compute, so this is the only SQL number you can budget without a separate EC2 line. |
| Lakeflow Jobs Classic | $0.15 | Billed separately, including idle time in instance pools | Scheduled ETL on clusters you configure. The idle-pool caveat is Databricks' own wording and it catches teams out. |
| Lakeflow Jobs Serverless | $0.35 | Included in the rate | Managed ETL in Performance Optimized or Standard mode. Databricks says Standard is up to 70% cheaper for some workloads. |
| All-Purpose Classic | $0.55 | Billed separately by your cloud provider | Interactive notebooks for data science and ad hoc analysis. The most expensive way to run an accidental overnight cluster. |
| All-Purpose Serverless | $0.75 | Included in the rate | The same interactive workload without managing instances. Photon is on automatically. |
| Genie One / Agents / Code | $0.070 | Genie compute billed separately, e.g. SQL Serverless at $0.70 | Natural-language questions, no-code agents and the AI coding assistant. One tenth the SQL Serverless rate, and it does not run the query. |
A DBU is a unit of processing capability consumed per second, so nothing here maps cleanly onto a headcount. That is the fundamental difference between Databricks and a seat-priced BI tool: compare it with Power BI pricing at $14.00 per assigned licence, or with the usage model in Snowflake pricing, which meters credits the same way Databricks meters DBUs.
›_ databricks cost, read correctly
Serverless rates include the cloud bill. Classic rates do not. That one footnote changes every comparison.
Databricks prints it in small grey text under each price card, and almost every third-party guide drops it.
Under SQL Serverless, Databricks writes "Includes underlying compute costs". Under SQL Classic and SQL Pro it writes "Plus underlying compute costs billed by cloud provider". Same page, same table, two different kinds of number. So the popular line that serverless SQL costs 3.2 times classic SQL is comparing an all-in price against a partial one, and it is wrong in whichever direction your instance bill happens to fall.
Subtracting the two published rates gives you something more useful: the instance allowance Databricks has effectively baked into each serverless SKU. That number is the break-even. Spend less than it on EC2 and classic is cheaper. Spend more and serverless is.
| Comparison | Implied instance allowance | How to read it |
|---|---|---|
| SQL Classic $0.22 vs SQL Serverless $0.70 | $0.48 | Serverless wins as soon as the EC2 or VM bill for that warehouse exceeds 48 cents per DBU hour. On a busy warehouse with modern instance types that is a low bar. |
| Jobs Classic $0.15 vs Jobs Serverless $0.35 | $0.20 | A 20 cent allowance is thin for ETL. Classic tends to win on long, predictable, well-packed jobs and lose on short bursty ones where you also pay for spin-up and idle pool time. |
| All-Purpose Classic $0.55 vs Serverless $0.75 | $0.20 | The same 20 cent allowance, but interactive clusters idle far more than jobs do, and classic bills that idle time to your cloud account. Serverless usually wins here on real usage. |
There is a second cost hiding in the classic column. Databricks' own comparison table says classic clusters make you "pay separately for underlying instances to the cloud provider, including idle time when leveraging instance pools". A warm pool that never fully drains is billed to your AWS or Azure account, arrives on a different invoice from Databricks, and is the usual reason a reconciled bill lands well above the DBU forecast.
›_ databricks pricing calculator
Monthly cost by DBU hours, at every published rate.
DBU hours per month multiplied by the AWS Premium US East rate. Find your warehouse or cluster's DBU per hour in its settings, multiply by the hours it actually runs, then read across.
| Monthly usage | SQL Classic $0.22 | SQL Pro $0.55 | SQL Serverless $0.70 | Jobs Classic $0.15 | Jobs Serverless $0.35 | All-purpose $0.55 | All-purpose serverless $0.75 |
|---|---|---|---|---|---|---|---|
| 500 DBU hours | $110 | $275 | $350 | $75 | $175 | $275 | $375 |
| 1,000 DBU hours | $220 | $550 | $700 | $150 | $350 | $550 | $750 |
| 2,500 DBU hours | $550 | $1,375 | $1,750 | $375 | $875 | $1,375 | $1,875 |
| 5,000 DBU hours | $1,100 | $2,750 | $3,500 | $750 | $1,750 | $2,750 | $3,750 |
| 10,000 DBU hours | $2,200 | $5,500 | $7,000 | $1,500 | $3,500 | $5,500 | $7,500 |
| 25,000 DBU hours | $5,500 | $13,750 | $17,500 | $3,750 | $8,750 | $13,750 | $18,750 |
Two rules before you trust a row. The three classic columns are incomplete by design: add your cloud provider's instance bill on top, because Databricks does not. And DBU hours are not wall-clock hours. A cluster emits DBUs at a rate set by its instance type, its tier and whether Photon is on, which is the subject of the next section.
›_ databricks dbu cost
Photon does not raise the DBU price. It raises the DBU count, by 2.9x.
Every Photon SKU on the Azure rate card is priced identically to its non-Photon twin. The multiplier is published in a feature table instead of a price table, which is why it gets missed.
Check the rate card and you will find Premium Jobs Compute at $0.30 and Premium Jobs Compute Photon at $0.30. Standard All-purpose at $0.40 and Standard All-Purpose Photon at $0.40. Identical, every time. Then look at Databricks' feature comparison and the real number appears: "DBU emission rate 2.9X vs. non-Photon" on Lakeflow Jobs, and "DBU emission rate 2X vs. non" on compute for data science. The price per unit did not move; the number of units did.
| Workload | Price effect | Consumption effect | Break-even rule |
|---|---|---|---|
| Lakeflow Jobs Classic | Identical $/DBU, Photon or not | 2.9x DBUs vs non-Photon | Photon has to finish the job in under 34.5% of the non-Photon runtime to break even |
| All-Purpose Classic | Identical $/DBU, Photon or not | 2x DBUs vs non-Photon | Photon has to finish in under 50% of the non-Photon runtime to break even |
| Serverless (any) | Photon is on automatically | Folded into the serverless rate | No decision to make. You are already paying for it |
The arithmetic is worth doing before you switch Photon on across a workspace. At 2.9x consumption, a job has to complete in less than 1 divided by 2.9, or 34.5%, of its original runtime just to cost the same. Photon genuinely does hit that on scan-heavy, vectorizable SQL. It does not on jobs dominated by Python UDFs, tiny files or shuffle. Measure one representative job both ways before you decide, because the billing effect is a near-3x multiplier either way.
›_ azure databricks pricing
Azure Databricks is priced by Microsoft, and the whole rate card is in a public API.
Databricks says so on its own pricing page: "Pricing for Azure Databricks is set by Microsoft and is provided for convenient reference." Every row below came from Microsoft's retail price API for East US on 20 August 2026.
This matters more than it sounds. Because Azure Databricks is a first-party Azure service, its prices are published as machine-readable meters rather than rendered in a JavaScript widget, so you can pull the authoritative rate card yourself in one request against prices.azure.com. Two structural differences show up immediately when you do.
| Azure meter | Standard | Premium | Worth knowing |
|---|---|---|---|
| Jobs Light Compute | $0.07 | $0.22 | The widest Standard to Premium gap on the card at 3.14x |
| Jobs Compute | $0.15 | $0.30 | Exactly 2x. Photon variant is priced identically |
| All-purpose Compute | $0.40 | $0.55 | 1.375x. Photon variant is priced identically |
| SQL Analytics (classic warehouse) | $0.22 | $0.22 | Identical on both tiers. The only SKU with no uplift at all |
| SQL Compute Pro | not offered | $0.55 | Matches the AWS Pro rate to the cent |
| Serverless SQL | not offered | $0.70 | Matches the AWS Serverless SQL rate to the cent |
| Interactive Serverless Compute | not offered | $0.95 | The single most expensive meter on the Azure card |
| Automated Serverless Compute | not offered | $0.45 | Serverless ETL, cheaper than the interactive equivalent |
| Delta Live Tables, Core | not offered | $0.30 | Photon and non-Photon priced the same |
| Delta Live Tables, Pro | not offered | $0.38 | Adds change data capture |
| Delta Live Tables, Advanced | not offered | $0.54 | 1.8x the Core rate for data quality expectations |
| Model Training | not offered | $0.65 | Fine-tuning foundation models |
| Serverless Realtime Inferencing | not offered | $0.07 | Plus a separate launch charge at the same rate |
| Database Serverless Compute | not offered | $0.26 | Lakebase, the managed Postgres product |
| Enhanced Security and Compliance | not offered | $0.10 | A flat per DBU hour adder on Azure, not the 15% of spend AWS charges |
| Mission Critical add-on | not offered | $0.15 | Also flat per DBU hour on Azure |
| Databricks Storage Unit (DSU) | not offered | $0.026 | Per DSU, the managed storage meter |
| Clean Rooms collaborator | not offered | $50.00 | Per collaborator PER DAY, which is $1,500 a month per partner |
| Disaster Recovery replication | not offered | $0.02 | Per GB of data processed, not per DBU hour |
First difference: the Standard to Premium uplift is not one multiplier. Jobs Light goes from $0.07 to $0.22, which is 3.14x. Jobs Compute doubles exactly, $0.15 to $0.30. All-purpose moves only 1.375x, $0.40 to $0.55. SQL Analytics does not move at all, $0.22 on both. Anyone budgeting "Premium is roughly double" is wrong on three of those four lines, and wrong by a factor of three on the first one.
Second difference, and it is the one that wrecks cross-cloud spreadsheets: the tier names do not line up. Databricks' footnote reads "The Premium tier on Azure Databricks corresponds to the Enterprise tier on AWS and GCP." A table with an "AWS Premium" column beside an "Azure Premium" column is comparing two different products, which is one reason the Interactive Serverless rate looks like $0.75 on AWS and $0.95 on Azure.
›_ databricks genie pricing
Genie is free for the thinking and billed for the doing.
Genie is $0.070 per DBU. The SQL Serverless warehouse that runs the query Genie writes is $0.70 per DBU. Exactly ten times more, on a separate meter.
Genie is Databricks' natural-language layer: ask a question in English, get an answer from your governed data. Genie One, Genie Agents and Genie Code all list at $0.070 per DBU, and each carries the note "Plus underlying compute costs, billed separately." Genie Ontology says "Price coming soon". So far so ordinary. The part that changes a budget is in the footnote attached to the free allowance.
Free until 31 January 2027
Genie One and Genie Agents carry a promotion marked FREE until 31 January 2027. Genie Code is 25% off until the same date. These are promotional prices with a published end date, not a permanent free tier, so a 2027 budget built on today's Genie line will be wrong.
150 DBUs per user from 1 February 2027
From 1 February 2027 every user gets 150 free DBUs a month across Genie One, Agents and Code. Databricks values that at $10.50 in US East, and the arithmetic checks: 150 multiplied by $0.070 is exactly $10.50. Its own FAQ says many users never exceed it.
The free allowance covers the LLM only
Databricks' footnote is explicit: "Free usage applies to Genie LLM usage only; Genie compute (e.g., SQL Serverless) is billed separately." The cheap meter is the one being given away. The expensive one, at ten times the rate, is the one that actually executes your question.
None of that makes Genie bad value. It makes the headline misleading. If you are pricing an AI analyst, the number to model is not $0.070 per Genie DBU, it is $0.070 plus however much SQL Serverless the questions consume at $0.70. And because Genie needs SQL Pro or Serverless rather than Classic, choosing it also moves your warehouse off the $0.22 rate. That is three separate line items behind one feature.
›_ databricks pricing tiers and add-ons
The compliance add-on is a percentage of list spend, so your negotiated discount does not touch it.
Enterprise tier only on AWS and GCP. Metered completely differently on Azure. Both facts are published, neither is widely repeated.
| Add-on | AWS and GCP | Azure | Availability |
|---|---|---|---|
| Enhanced Security and Compliance | 15% of Product Spend before discounts | $0.10 per DBU hour | Enterprise tier only on AWS and GCP |
| Mission Critical | 30% list, currently 15% under a promotion ending 30 June 2027 | $0.15 per DBU hour | Enterprise tier only on AWS and GCP |
Read Databricks' definition of Product Spend carefully, because it does a lot of work: spend "at list incurred in the specific workspaces where the add-on is enabled (turned on), before the application of any discounts, usage credits, add-on uplifts, or support fees". The add-on is charged on the list price, not on what you pay. So a team that negotiated 30% off and spends $350,000 against a $500,000 list still pays 15% of $500,000, which is $75,000. As a share of the money actually leaving the bank that is 21.4%, not 15%. The better your commercial terms, the worse that ratio gets.
Azure charges for the same capability with a flat per DBU hour adder instead, and a flat adder behaves very differently across SKUs. Ten cents on top of Premium all-purpose at $0.55 is an 18.2% surcharge. The same ten cents on top of Premium SQL Analytics at $0.22 is 45.5%. If your compliance-scoped workspace runs mostly cheap classic SQL, Azure's flat model costs you proportionally more than twice what it costs a workspace running expensive interactive compute.
One date to diary: the Mission Critical add-on lists at 30% of product spend and is currently 15% under a promotion Databricks says ends 30 June 2027. If your contract runs past that, model 30%.
›_ honest read
Where consumption pricing helps you, and where it bites.
We build an AI data analyst, so weigh this accordingly. Every Databricks figure above is checkable against Databricks' and Microsoft's own pages.
Nobody pays for a seat they never use
This is the genuinely good part. Power BI bills every assigned licence whether or not it is opened, and Tableau bills a Creator seat for an analyst on parental leave. Databricks bills compute. Give a hundred people read access and the ninety who never query cost you nothing.
The bill is only as disciplined as your cluster policies
The flip side of no seats is no ceiling. An all-purpose cluster left running over a long weekend bills at $0.55 per DBU plus instances, and nothing about the pricing model stops it. Auto-termination and cluster policies are not optional hygiene here, they are the budget.
Forecasting DBUs is genuinely hard
You cannot forecast DBU hours from headcount, query count or data volume with any confidence, because emission rate depends on instance type, tier and Photon. Expect the first two months to be a measurement exercise rather than a budget, and expect to revise.
It is a lakehouse first, a BI tool second
Databricks is superb at engineering and ML on very large data. If your actual need is "let the ops team ask questions of a Postgres database", you are buying a platform to solve a query problem, and the pricing reflects the platform.
Worth saying plainly, since this page discusses Genie: Agentsql does not connect to Databricks. We connect read-only to Postgres, MySQL, Snowflake and BigQuery, turn a plain-English question into SQL, run it, and show the SQL every time so an analyst can check the logic. If your data already lives in one of those four, that is a flat monthly price with no DBU meter attached, and you can read ours on our pricing page. If your data lives in a lakehouse, Genie is the right thing to price, and the numbers above are the ones to price it with.
Still comparing platforms? The seat-based side of the market is covered in Tableau pricing, Looker pricing and QuickSight pricing, the usage-metered side in BigQuery pricing and Grafana pricing, and all of them side by side in the BI tools comparison.
›_ frequently asked
Databricks pricing questions, answered.
How much does Databricks cost?
Databricks charges per DBU, a unit of processing consumed per second, not per seat. On AWS Premium in US East the published rates are $0.22 per DBU for SQL Classic, $0.55 for SQL Pro, $0.70 for SQL Serverless, $0.15 for Lakeflow Jobs and $0.55 for all-purpose compute. Classic rates exclude your cloud instance bill; serverless rates include it.
What is a DBU?
A DBU, or Databricks Unit, is Databricks' unit of processing capability billed per second of use. It is not an hour, a core or a query. Different workloads emit DBUs at different rates, so the same cluster running the same wall-clock hour can consume very different DBU counts depending on instance type, tier and whether Photon is enabled.
Is Databricks pricing based on users?
No. Unlike Tableau, Power BI or Looker, Databricks does not sell seats. You pay for compute consumed, so a hundred people who each run one query a week can cost less than one person running a badly written hourly job. The only per-user element is Genie, which grants 150 free DBUs per user per month from 1 February 2027.
Is there a Databricks pricing calculator?
Databricks publishes a cost calculator on its pricing site, and the table above does the same arithmetic in one screen. The formula is simple: DBU hours per month multiplied by the published rate for your SKU. For classic SKUs you then add your cloud provider's instance bill, which Databricks does not include.
Does the DBU price include the cloud VM cost?
Only on serverless SKUs. Databricks prints "Includes underlying compute costs" under SQL Serverless, Jobs Serverless and All-Purpose Serverless, and "Plus underlying compute costs billed by cloud provider" under every classic SKU. Missing that footnote is the single most common way a Databricks budget comes in low.
How much does Databricks SQL Serverless cost?
SQL Serverless is $0.70 per DBU on AWS Premium in US East and $0.70 on Azure Databricks Premium in East US, the same to the cent. That rate includes the cloud instance cost, so it is directly comparable to SQL Classic at $0.22 only after you add classic's separate EC2 or VM bill.
Is Databricks Genie free?
Genie One and Genie Agents are promotionally free until 31 January 2027, and Genie Code is 25% off until the same date. From 1 February 2027 every user gets 150 free DBUs a month, worth $10.50 in US East. The catch is in Databricks' own footnote: free usage covers Genie LLM usage only, and the SQL Serverless compute that runs the query is billed separately at $0.70 per DBU.
Does Photon cost more in Databricks?
The DBU price is identical with or without Photon. What changes is the DBU emission rate: Databricks states 2.9x on Lakeflow Jobs and 2x on all-purpose compute. So Photon is cheaper only when it cuts runtime by more than it multiplies consumption, which means beating 34.5% of the original runtime on Jobs and 50% on all-purpose.
What is the difference between Databricks Premium and Enterprise?
Enterprise adds the Enhanced Security and Compliance and Mission Critical add-ons, which Premium cannot buy. Naming differs by cloud, and Databricks says so: the Premium tier on Azure Databricks corresponds to the Enterprise tier on AWS and GCP, so an Azure-to-AWS comparison at the same tier name is comparing different products.
How much is the Databricks compliance add-on?
On AWS and GCP, Enhanced Security and Compliance is 15% of Product Spend before discounts, and Databricks defines Product Spend as spend at list, before any discounts, credits or support fees. A customer with a 30% negotiated discount therefore pays 15% of list, which works out at 21.4% of what they actually spend. On Azure it is a flat $0.10 per DBU hour instead.
Is Databricks free?
There is a Databricks Free Edition, which replaced Community Edition, and a 14 day free trial of the paid platform. Neither is a production tier. Free Edition is for learning; the trial runs the real product on Databricks-supplied compute for two weeks. Beyond that every DBU is billed.
Is Databricks expensive?
It is expensive when compute is left running and cheap when it is not, which is the opposite of seat-based BI tools. A team of five that queries constantly can outspend a team of fifty that queries occasionally. The two costs that surprise people most are idle classic clusters, billed to your cloud account rather than to Databricks, and the compliance add-on charged on list spend.
›_ pricing and comparisons
›_ connect your database