Databricks Pricing Guide (2026): DBU Costs, Tiers & How to Cut Your Bill

In this guide

There’s no way to sugarcoat it: Databricks pricing is confusing. If you’ve ever opened the Databricks pricing page and thought, “So how much is this actually going to cost me?”, you’re not alone. Between Databricks Units (DBUs), a long list of compute types, two current pricing tiers on AWS, and infrastructure costs that Databricks doesn’t even bill you for, figuring out your real bill can feel like a machine learning problem in itself.

The part that catches most people is the second bill. On classic compute, Databricks bills you for DBUs, and AWS bills you separately for the EC2 instances doing the work. On serverless compute, those servers are already inside the DBU price. Mix the two up and your estimate is wrong before you’ve multiplied anything.

Maybe you’re a data engineer wondering why your cluster costs so much, a FinOps practitioner building a cost model, or an engineering manager deciding whether Databricks is worth it. In this guide, we’ll cover how DBU billing works, today’s tiers, how compute types and clouds change the price, a worked example with dated list prices, the official calculator, and the questions we hear most.

By the end, you’ll know which usage, rates and AWS charges belong in your estimate, and how to check it against what you actually paid.

How Databricks Pricing Works: The DBU Model Explained

Before any numbers, there’s one concept underneath all of Databricks pricing: the Databricks Unit, or DBU.

What Is a DBU (Databricks Unit)?

Databricks defines a DBU as “a normalized unit of processing power”, used for measurement and pricing. How many DBUs a workload consumes depends on processing metrics, such as the compute resources used and the amount of data processed.

The formula is simple:

DBU formula: Cost = DBUs consumed × price per DBU

What’s less simple is that three different units hide in that sentence (earlier versions of this guide wrote “DBU rate per hour” here, which blends two of them):

• DBUs are the quantity you consumed: the usage_quantity column in Databricks’ system.billing.usage table.
• DBUs per hour (DBU/h) is how fast a configuration consumes them. On classic compute, each instance type has its own figure. The official calculator lists an m5.xlarge at 0.690 DBU/h for both jobs and all-purpose compute (Premium, AWS, October 21, 2026).
• Price per DBU ($/DBU) depends on the workload (Databricks calls it a SKU), the tier, the cloud and the region.

On classic compute, DBUs consumed = DBU/h × hours, summed over every node, driver included. We’ll put real numbers on all three in the cost example.

How Databricks Bills You: Per-Second Pricing & On-Demand vs. Committed Use

Here’s some good news: pay-as-you-go has no up-front costs, and you “only pay for the products you use at per second granularity.” Serverless compute is billed with a one-minute minimum. On classic compute, a second meter runs alongside: AWS bills the EC2 instances, and Linux On-Demand instances are billed per second, with a 60-second minimum.

Databricks also sells committed use contracts. In its own words, “The larger your usage commitments, the greater your benefits,” and commitments can be used across clouds. It doesn’t publish the discount bands, so we won’t guess at a percentage: your contract sets your rate. Promotional discounts, when they run, are marked on Databricks’ price list. When committing makes sense is covered in tip 8 of our optimization guide.

So what does a DBU actually pay for? This is the table we wish someone had shown us on day one:

	Classic compute	Serverless compute
Databricks DBUs	Yes: DBUs × $/DBU	Yes: DBUs × $/DBU
Servers (EC2)	Billed separately by AWS, in your account	Included in the DBU price
Storage (S3)	Your buckets, on your AWS bill	Your buckets, on your AWS bill; some Databricks products also bill storage in DSUs
Networking	You pay AWS directly (a few products, such as Lakebase, also carry Databricks egress charges)	Some data transfer and connectivity billed by Databricks; in-region S3 access through gateway endpoints isn’t

Sources: Databricks’ cost guidance (for serverless, “the DBU cost already includes the virtual machine costs”) and its serverless networking cost docs.

Databricks Pricing Tiers: Current Plans and Legacy Standard

Databricks pricing starts with three separate inputs: a tier (your feature set), a workload (the compute type) and a cloud (plus region). On AWS, the tier choice is short. Databricks “has two platform tiers: Premium and Enterprise,” and new accounts start on Premium. If you’re weighing Databricks Premium vs. Standard, Standard no longer exists on AWS. You’re not really buying a traditional Databricks license, either: you’re choosing a tier that changes what you can do and, for some products, what you pay per DBU.

Legacy Databricks Standard: What Changed?

Standard used to be the entry-level tier. Databricks ended Standard tier workspaces on AWS, and remaining ones were automatically upgraded to Premium on October 1, 2025. Google Cloud used the same date. Azure ran a year behind: new workspaces have had to start on Premium since April 1, 2026, and remaining Standard workspaces were scheduled for upgrade on October 1, 2026.

Databricks’ end-of-life pages for Azure and Google Cloud show Standard without Unity Catalog, Databricks SQL, serverless compute or, on Azure’s list, Lakeflow pipelines. If you spot Standard SKUs in old billing data, that’s history; you can’t model new AWS work on it. (An earlier version of this guide suggested Standard for small teams. That advice no longer applies.)

Databricks Premium: Additional Features & When to Upgrade

Premium is the default plan on AWS, so “when to upgrade” mostly answers itself: you’re already there. Databricks’ platform tiers page for AWS lists a workspace for production jobs, analytics and ML, the optimized runtime engine, and audit logs and automated policy controls. Premium workspaces also get optimized autoscaling, and the current Unity Catalog docs for AWS set no tier requirement. Unless we say Enterprise, every list price in this guide is a Premium price.

Databricks Enterprise: Pricing and Controls

Enterprise is for organizations with stricter security and compliance needs: the tiers page describes it as “advanced compliance and security for mission-critical data.” On AWS, two paid add-ons sit in the Enterprise column: Enhanced Security and Compliance (15% of product spend before discounts) and Mission Critical (managed disaster recovery plus Enhanced Security and Compliance, 30%, on a promotional 15% until June 30, 2027), as listed on October 21, 2026.

Contrary to what this guide used to say, Enterprise list prices are published. On the Lakeflow Pipelines page, serverless pipelines showed $0.35/DBU on Premium and $0.45/DBU on Enterprise (AWS, US East (N. Virginia), October 21, 2026). Committed-use discounts on top are negotiated.

Feature	Standard (legacy, retired on AWS)	Premium	Enterprise
Workspace for production jobs, analytics and ML	✓ (managed Spark workspace)	✓	✓
Unity Catalog	✗	✓	✓
Databricks SQL, Lakeflow pipelines, serverless compute	✗	✓	✓
Optimized autoscaling	✗ (standard autoscaling)	✓	✓
Audit logs and automated policy controls	Not listed	✓	✓
Security and compliance	Not listed	Cloud-native security for company-wide adoption	Advanced compliance and security for mission-critical data
Enhanced Security and Compliance; Mission Critical add-ons	✗	Not listed	Paid add-ons, % of product spend
Dedicated support and SLA	Check your contract	Check your contract	Check your contract
Pricing	No longer sold on AWS	Published list prices	Published list prices; discounts negotiated

The Standard column comes from Databricks’ Azure and Google Cloud end-of-life pages, plus the AWS compute docs for the autoscaling row. Tier names don’t map one-to-one across clouds, either: “The Premium tier on Azure Databricks corresponds to the Enterprise tier on AWS and GCP.”

Databricks DBU Rates by Compute Type (2026)

Here’s where things get really important. Two things change between compute types: the price per DBU and how many DBUs per hour a configuration emits. Let’s anchor this with one dated comparison, at list prices for Premium on AWS, US East (N. Virginia), October 2026. Jobs Compute is $0.15/DBU and All-Purpose Compute is $0.55/DBU, both plus EC2. An m5.xlarge emits 0.690 DBU/h on either, so the Databricks charge per node-hour is 0.690 × $0.15 = $0.1035 on jobs compute and 0.690 × $0.55 = $0.3795 on all-purpose. Add the m5.xlarge’s EC2 rate in us-east-1, $0.192/hour, and it’s $0.2955 vs. $0.5715 per node-hour.

Notice what happened there. The DBU price is about 3.7 times higher on all-purpose compute, but once EC2 is in, the all-in gap narrows to about 1.9 times. Always compare the all-in cost per node-hour.

All-Purpose Compute: Pricing for Interactive Workloads

All-Purpose Compute runs your interactive notebooks and development work, and of the two classic rates above, it carries the higher one. Watch the idle time. All-purpose clusters keep running until someone stops them or auto termination kicks in, and “Idle compute continue to accumulate DBU and cloud instance charges” in the meantime. You set the inactivity period yourself (10 to 10,000 minutes). So if someone runs a cell and heads to lunch, you’re paying all-purpose rates plus EC2 until the timer runs out.

Jobs Compute: Pricing for Automated Workloads

Jobs Compute is built for scheduled work: nightly ETL, data quality checks, model training runs. A new job cluster terminates when the job is complete, so there’s no idle tail waiting for someone to notice. Still running scheduled jobs on all-purpose clusters? Here’s how to move them.

Its list price per DBU is lower than all-purpose on the same plan, and Databricks says non-interactive workloads on job compute “will cost significantly less”. On your bill, classic jobs show up as “Jobs Compute” (or “Jobs Compute (Photon)”), and serverless jobs on AWS as “Jobs Serverless Compute”.

Databricks SQL Warehouses: Classic vs. Serverless DBU Costs

SQL warehouses are the compute behind Databricks SQL. There are three main types: classic and pro, whose compute runs in your AWS account (DBUs to Databricks, EC2 to AWS), and serverless, whose DBU price includes the compute. A fourth, Lakehouse Real-Time, is a serverless type for low-latency reads, in Beta. Each is priced per DBU, with DBUs per hour set by the warehouse size.

All of them charge while idle, until Auto Stop shuts them down. We’ll get into the classic vs. serverless trade-off in the serverless section.

Lakeflow Pipelines (Formerly Delta Live Tables) & Model Serving: What DBUs Do They Consume?

• Lakeflow pipelines. Databricks renames things faster than most of us update our spreadsheets: Delta Live Tables is now Lakeflow pipelines, though the classic SKUs still begin with DLT. Classic pipelines come in three editions (Core, Pro and Advanced), each with its own $/DBU plus EC2. Serverless pipelines are one SKU with two modes, Standard and Performance Optimized, compute included.
• Model Serving. CPU and GPU endpoints for models, features and agents, with the cloud instance cost included. GPU serving is billed per GPU instance per hour, at a published DBUs-per-hour figure for each configuration. For CPU serving without scale to zero, the minimum charge depends on the endpoint’s minimum provisioned concurrency.

Compute type	Best for	Unit	Servers included in DBU price?	Idle-cost behavior	Official price page
All-Purpose Compute (classic)	Interactive notebooks, development	$/DBU, plus EC2	No, AWS bills EC2	DBUs and EC2 accrue until auto termination or a manual stop	Data science & ML
Lakeflow pipelines (formerly DLT)	Streaming and batch pipelines	$/DBU per edition (classic); one $/DBU for serverless, both modes	Classic: no. Serverless: yes	Classic: AWS bills instances, including idle time in pools	Lakeflow Pipelines
Jobs Compute (classic)	Scheduled, automated jobs	$/DBU, plus EC2	No, AWS bills EC2	A new job cluster terminates when the run completes	Lakeflow Jobs
SQL warehouse (classic or pro)	SQL analytics and BI	$/DBU per size, plus EC2	No, compute runs in your AWS account	Charges until Auto Stop (default 45 minutes)	Databricks SQL
SQL warehouse (serverless)	Bursty or ad hoc SQL	$/DBU per size	Yes	Charges until Auto Stop (default 10 minutes)	Databricks SQL
Model Serving	Model, feature and agent endpoints	$/DBU; GPU configs list DBUs per hour	Yes	Scale to zero optional; CPU endpoints otherwise bill at minimum provisioned concurrency	Model Serving

Databricks Pricing by Cloud: AWS vs. Azure vs. GCP

Here’s something a lot of pricing guides skip: on classic compute, your Databricks bill isn’t your whole Databricks cost. Your cloud provider bills the infrastructure underneath, and how the two split depends on the workload (the example below shows how much).

Databricks on AWS: DBU Pricing, EC2 Costs, and Instance Choices

On AWS, your classic-compute total is DBUs plus EC2 instances, their EBS volumes, and data transfer. Databricks adds default tags such as ClusterId and JobId (jobs compute only) to the EC2 instances and EBS volumes it launches, which is how you find them in your Cost and Usage Report later. One gotcha: in Cost Explorer, those EBS volumes land under EC2-Other, not EC2-Instances.

Instance choice moves both meters. Here are three general-purpose types from the calculator (Premium, AWS, jobs compute) with their us-east-1 Linux On-Demand rates, October 21, 2026:

Instance	vCPUs / memory	DBU/h (jobs compute)	EC2 On-Demand (us-east-1)
m5.xlarge	4 / 16 GB	0.690	$0.192/hour
m6i.xlarge	4 / 16 GB	0.760	$0.192/hour
m5.2xlarge	8 / 32 GB	1.370	$0.384/hour

The m5.xlarge and m6i.xlarge cost the same on EC2, but the m6i emits more DBUs per hour, so it only wins if your job finishes sooner on it. Same with AWS Graviton: Databricks supports it and notes AWS’s claim of the best price-to-performance on EC2, but you can’t mix Graviton and non-Graviton nodes in one cluster, and some features don’t support it (see the compute docs). We dropped this guide’s old “up to 4x” Graviton figure because we couldn’t source it. Test it; don’t assume. To size from real utilization instead of guesses, see right-sizing from bottlenecks.

AWS’s usual EC2 levers apply to classic clusters, and none of them change the price per DBU:

• Spot Instances, at up to 90% off On-Demand in exchange for interruptions. With Use spot instance checked, the driver stays on-demand. More in are Spot Instances worth it in production?
• Savings Plans and Reserved Instances. Compute Savings Plans lower the price of EC2 instance usage, classic clusters included (Savings Plans vs. Reserved Instances).
• AWS Marketplace. If you pay for Databricks through Marketplace, the charges appear on your AWS Billing dashboard with your other AWS charges. Handy, but if you add “Databricks invoice + AWS bill” in a spreadsheet, you can count the DBUs twice.

Databricks on Azure: Azure Databricks Pricing & VM Costs

Azure Databricks is a Microsoft first-party service, billed directly by Microsoft, which also sets its pricing. Model DBU charges separately from VMs, storage and networking. Commitments differ, too: Azure sells prepurchased Databricks commit units (DBCUs), and the prepurchase discount applies only to DBU usage; compute, storage and networking are charged separately. So any VM discount is a separate decision. And watch the names: Azure’s Premium corresponds to Enterprise on AWS and GCP.

Databricks on GCP: DBU Rates & Committed Use Discount Interaction

Databricks on Google Cloud uses the same DBU structure, with its own prices by region, and on classic compute Google Cloud bills the underlying compute separately. Standard was retired on the same date as on AWS.

On committed use discounts (CUDs), Google ties each resource-based commitment to a region and a project. Whether it covers other projects depends on your billing account’s CUD sharing setting (on by default for accounts created on or after June 16, 2026). Before counting on CUDs for your Databricks VMs, check which project and region that compute runs in, and confirm coverage with your Google account team.

Factor	AWS	Azure	GCP
DBU pricing structure	DBUs × $/DBU; varies by region	Same structure; prices set by Microsoft	DBUs × $/DBU; varies by region
Infrastructure billing	EC2, EBS and data transfer, billed by AWS (classic)	VMs, storage and networking, charged separately from DBUs	Underlying compute, billed by Google Cloud (classic)
Best price/performance option	Test Graviton (AWS claims the best price-performance on EC2)	Check Azure pricing	Check Google Cloud pricing
Infrastructure discount path	Spot, Savings Plans, Reserved Instances	DBCUs cover DBUs only; VM discounts are separate	CUDs, scoped by region and project
Tier names	Premium, Enterprise	Premium (= Enterprise on AWS and GCP)	Premium, Enterprise

What Drives Your Databricks Costs? 7 Key Factors

So what actually makes your bill go up or down? Each factor changes either usage (DBUs, server-hours) or price (the rate per unit), and the fixes differ.

• Cluster size (usage). Every node emits DBUs and runs up EC2 hours, driver included. By default, the driver uses the same instance type as the workers, so a “two-worker cluster” is three instances on both bills.
• Cluster type and autoscaling (usage). With autoscaling, you set minimum and maximum workers and Databricks picks the number in between. That can trim node-hours on variable workloads, and the maximum keeps a busy day from becoming a spike.
• Compute type selection (price). The same m5.xlarge node-hour costs $0.1035 in DBUs on jobs compute and $0.3795 on all-purpose, as we saw above.
• Region (price). Databricks and cloud infrastructure prices may vary by region. Our example uses us-east-1; don’t assume other regions match it, or that it’s the cheapest.
• Databricks Runtime and Photon (both). On classic compute, Photon has no separate $/DBU; it emits more DBUs per hour instead: 2.9 times the non-Photon rate on jobs compute and 2 times on all-purpose. It’s also on by default when you create classic compute in the UI. Does it pay off? On an m5d.xlarge jobs node, it’s 2.001 × $0.15 + $0.226 = $0.52615 per node-hour with Photon vs. 0.690 × $0.15 + $0.226 = $0.3295 without, so the Photon run has to finish in under about 63% of the time to come out cheaper. Databricks says regular jobs “should be evaluated” for exactly that. How to test Photon on your jobs.
• Storage and network (outside the DBU price). Your S3 buckets are on your AWS bill, some Databricks products bill storage in DSUs, and serverless workloads can pick up Databricks networking charges for cross-region, internet or private connectivity. Classic workloads pay AWS for data transfer directly.
• Workload duration and scheduling (usage). Hours multiply everything else. A job that creeps from 2 to 2.5 hours a night uses 25% more DBUs and 25% more EC2 hours, with the cluster unchanged.

Illustrative Databricks Cost Example: Development and Production

So what might a mid-size data engineering team pay? This example is illustrative, built from dated list prices, not a customer’s bill. Here are our assumptions:

• Team: 5 engineers doing interactive development, plus 3 nightly production pipelines.
• Pricing: Premium, AWS, US East (N. Virginia) / us-east-1, USD, pay-as-you-go list prices as of October 21, 2026.
• Nodes: m5.xlarge for every node, drivers included. Fixed size, no autoscaling, no Spot, Photon off (the calculator doesn’t list m5.xlarge as a Photon type).
• Development: 5 separate all-purpose clusters, each 2 workers + 1 driver, billed 4 hours a day on 22 working days.
• Production: 3 separate jobs clusters, each 4 workers + 1 driver, billed 2 hours a night for 30 nights.
• Rates: $0.55/DBU (All-Purpose) and $0.15/DBU (Jobs) from Databricks’ product pages; 0.690 DBU/h per m5.xlarge from the calculator; EC2 $0.192/hour per m5.xlarge from the AWS Price List (Linux On-Demand).

First things first, the hours. Development is 5 × 4 × 22 = 440 cluster-hours, or 440 × 3 = 1,320 node-hours. Production is 3 × 2 × 30 = 180 cluster-hours, or 180 × 5 = 900 node-hours. Each dev cluster emits 3 × 0.690 = 2.07 DBU/h, and each production cluster 5 × 0.690 = 3.45 DBU/h. Now for the meat and potatoes:

Line	Cluster-hours	Node-hours	DBUs (cluster-hours × DBU/h)	Databricks $	AWS EC2 $
Dev (All-Purpose)	440	1,320	440 × 2.07 = 910.8	910.8 × $0.55 = $500.94	1,320 × $0.192 = $253.44
Prod (Jobs Compute)	180	900	180 × 3.45 = 621	621 × $0.15 = $93.15	900 × $0.192 = $172.80
Storage (S3; DSUs if used)				Your figure (DSUs)	Your figure (S3)
Network (transfer, connectivity)				Your figure (serverless paths)	Your figure
Other (EBS per node, retries, idle tails, taxes, support)				Your figure	Your figure
Compute subtotal	620	2,220	1,531.8	$594.09	$426.24

So compute comes to $594.09 + $426.24 = $1,020.33 a month, before storage, network and the other lines. Hold onto that $1,020 number.

Two things jump out. First, $426.24 of EC2 never appears on a Databricks invoice. Second, the split flips: on the dev clusters, Databricks ($500.94) is about twice EC2 ($253.44), while on the production jobs, EC2 ($172.80) beats Databricks ($93.15). So no single infrastructure percentage holds.

We also ran both rows through the official calculator on October 21, 2026 (15 m5.xlarge instances at 4 hours × 22 days on All-Purpose Compute, and 15 at 2 hours × 30 days on Lakeflow Jobs Classic). It showed 1,320 and 900 instance hours, 910.80 and 621.00 DBUs, and a monthly total of $594.09. Whew! That matches our Databricks column exactly, and it leaves out the EC2 column entirely.

Important note: these are list prices for one region and one instance type. Your bill will differ with your setup, idle time, retries and any committed-use discount, and the example can’t tell you what the same work costs on serverless (that takes a measured test run).

How to Use the Databricks Cost Calculator (Step-by-Step)

Databricks provides an official pricing calculator. Here’s how we got the numbers above (layout as of October 21, 2026):

1. Choose Select plan (Premium or Enterprise) and Select cloud (AWS, Azure or Google Cloud).
2. Pick a Compute type. Each option shows a list price, such as All-Purpose Compute ($0.55/DBU).
3. Fill in the inputs for that compute type:
   • Classic compute (jobs, pipelines, all-purpose): AWS instance type, #Instances, Hours/Day and Days/Month. Count every node, drivers included: our dev row was 15 instances (5 clusters × 3 nodes).
   • SQL warehouses: a Compute Configuration (warehouse size, 2X-Small to 4X-Large, plus 5X-Large on serverless and pro, in Public Preview) instead of an instance type.
   • Serverless jobs and pipelines: you enter DBUs and pick a region.
4. Read Instance hours, Usage (DBUs) and Price/month.
5. Click Add compute type for each additional workload. The Monthly total sums the rows.
6. Add the AWS side yourself for every classic row.

Below the calculator, the Supported Instance Types table lists a Rate ($/hour) for each instance type. Careful: that’s the Databricks charge per instance-hour (DBU/h × $/DBU), not the EC2 price. An m5.xlarge on Lakeflow Jobs Classic shows $0.1035, and EC2 adds $0.192/hour on top.

Databricks Calculator Limitations: What It Doesn’t Tell You

The calculator is useful, but it has some real blind spots:

• Cloud infrastructure. Non-serverless estimates “do not include cost for any required AWS services”, such as EC2. In our example, that’s the whole $426.24.
• Storage, network, idle time and retries. It prices exactly the hours you type in, and nothing else.
• Region labels. Classic rows have no region selector; serverless rows do. On October 21, 2026, the Compute type list showed Lakeflow Jobs Serverless at $0.40/DBU, but with US East (N. Virginia) selected, the row priced at $0.35/DBU, matching the product page. Write down which rate you used.
• Your contract. List prices only, before any committed-use discount.

Why Databricks and AWS Costs Are Hard to See Together

Even after a workload has run, getting one honest number for it takes work, because each side only sees its own half:

• Databricks sees DBUs. system.billing.usage (joined with list_prices) and the v2.0 usage dashboard show your DBUs at list price. The EC2 instances, EBS volumes and data transfer underneath aren’t in them.
• AWS sees servers. Cost Explorer and the CUR show the EC2 and EBS spend (the EBS volumes land under EC2-Other), but not which cluster or job it belongs to until you activate the ClusterId and JobId tags as cost allocation tags. AWS can backfill activated tags up to 12 months, but a tag that wasn’t on the instance when it ran can’t be added later. Pool instances get only the workspace and pool tags.
• Serverless has no servers in your account. The compute is included in the Databricks charge, so there’s no EC2 to tag (your own S3 stays on the AWS bill). Its usage lands in system.billing.usage, typically within 12 hours, where an AWS tag report can’t see it.
• It’s easy to count twice. If you pay through AWS Marketplace, Databricks bills you through AWS, so the DBUs are already a Marketplace line in your CUR. Add the list-price cost from system.billing.usage on top and you’ve counted them twice.

The native fix is to compare the estimate with actual usage once the workload has run: system.billing.usage for the DBU side, and Cost Explorer or the CUR for AWS (steps in the conclusion).

Databricks Serverless Pricing: SQL Warehouses & Serverless Jobs

On serverless, Databricks runs the machines in its own account and folds them into the DBU price. The bill gets simpler; the comparison gets trickier.

Serverless SQL Warehouses: Billing and Cost Trade-Offs

Pro and classic warehouses run on compute in your AWS account and typically take about 4 minutes to start. Serverless warehouses include the compute in the DBU price and typically start in 2 to 6 seconds.

Does serverless mean you never pay for idle? Not quite. Idle warehouses of every type “continue to accumulate DBU and cloud instance charges until they are stopped”. Auto Stop defaults to 10 minutes on serverless (5 minimum in the UI) and 45 minutes on pro and classic (10 minimum). What changes is how practical a short timeout is: Databricks notes that minutes-long startups lead many teams to leave non-serverless warehouses running. It recommends serverless where available.

Which costs less depends on your query patterns:

• Busy all day (a BI dashboard with dozens of users): there’s little idle time either way, so it comes down to DBUs used and the all-in price, including EC2 for pro or classic.
• Bursty and ad hoc (analysts who query hard for 2 hours, then stop): short Auto Stop timeouts pay off here, and serverless makes them practical.

Not sure? Run the same workload on both for a week and compare.

Databricks Serverless Jobs: Billing and Cost Trade-Offs

Serverless jobs give you job compute without a cluster to configure. Databricks picks and scales the compute, with autoscaling and Photon on automatically. You don’t pay for VM startup time; billing starts when the job starts executing, with a one-minute minimum. There are two modes on the same SKU: Standard uses fewer DBUs but can take 4 to 6 minutes to start, while Performance optimized starts in under a minute for time-sensitive work.

At Premium, AWS, US East list prices, serverless jobs cost $0.35/DBU with compute included, vs. $0.15/DBU plus EC2 for classic. Here’s a rough yardstick: our classic m5.xlarge jobs node-hour costs $0.2955 all-in, which buys about 0.84 serverless DBUs at $0.35. Serverless comes out ahead if it does the same work in fewer DBUs than that, once you count the startup and idle time you no longer pay for. Classic DBU/h figures can’t predict serverless DBUs, so measure a real run in system.billing.usage. How to tell if a workload should go serverless.

Two gotchas. Data quality monitoring and predictive optimization bill under the serverless jobs SKU, even if you’ve never run a serverless job. And serverless usage isn’t on your AWS bill (unless you pay through Marketplace), so AWS tag reports won’t see it; Databricks attributes it with serverless usage policies instead.

How to Reduce Your Databricks Costs

Understanding the pricing is great, but cutting the bill is better. The biggest levers, one line each:

• Jobs or serverless compute for scheduled work, instead of all-purpose clusters.
• Auto termination on interactive clusters and short Auto Stop timeouts on SQL warehouses.
• Right-sizing clusters from real utilization, not the original guess.
• Spot Instances for workers on fault-tolerant classic jobs.
• Photon, tested job by job against its higher DBU rate.
• Committed use contracts (pre-purchased DBUs), once usage is steady.
• Tags, so every cost has an owner.

The full playbook, with how to test each change, is in Databricks Cost Optimization on AWS: 8 Ways to Cut Costs.

Databricks vs. Snowflake vs. BigQuery: Quick Cost Comparison

Comparing Databricks with alternatives? These platforms meter compute in different units, so here’s where each bill comes from.

	Databricks	Snowflake	BigQuery
Pricing model	DBUs × $/DBU, plus the cloud bill on classic compute	Credits × price per credit for compute, plus storage and data transfer	On-demand per TiB processed, or capacity per slot-hour, plus storage
How compute is metered	Per second; classic clusters in your cloud account, or serverless	Virtual warehouses per second (60-second minimum per start), plus serverless features and cloud services	Bytes processed per query (on-demand) or slots over time (capacity)
Separation of compute / storage	Compute billed by Databricks (plus your cloud on classic); data in your cloud storage	Compute, storage and data transfer billed separately	Compute and storage priced separately
What to benchmark	The same job on jobs compute and on serverless	The same queries at your chosen warehouse size	Bytes processed (on-demand) vs. slot-hours (capacity)
Main cost levers	Compute type, auto termination, Spot (classic), commitments	Warehouse size, auto-suspend (on by default)	Partitioning and clustering to cut bytes processed; slot commitments

Which is cheapest? We haven’t run the same workload on all three, so we won’t name a winner. A fair test runs the same workload, with the same output and deadline, on each and compares the full bill.

Frequently Asked Questions About Databricks Pricing

How Much Does Databricks Cost per Month?

There’s no honest single number: it depends on your workloads, tier, region, instances and billed hours. Our illustrative example (5 engineers on dev clusters plus 3 nightly pipelines, Premium on AWS us-east-1) comes to about $1,020 a month for compute at October 2026 list prices, before storage, network and other charges. To find yours, build one line per workload in the calculator and add the AWS side for classic compute.

What Is a DBU in Databricks?

Databricks’ normalized unit of processing power. Your cost is DBUs consumed × price per DBU, which varies by compute type, tier, cloud and region; on classic compute, AWS bills the servers separately. More in the DBU model.

Is Databricks Free to Use?

Partly. Databricks Free Edition is a no-cost, serverless-only, quota-limited version for students, educators, hobbyists and anyone learning. Per its limitations page, it’s for non-commercial use with no SLA, and if you exceed your quota, compute shuts down for the rest of the day. Free Edition replaced the legacy Community Edition, which Databricks’ docs say was retired in 2025.

A free trial is different. As of October 2026, the trial docs describe usage credits valid for 14 days, while the pricing page says a trial ends when its credits are exhausted, so check the terms when you sign up. If you connect the trial to your own AWS account, AWS still charges for resources such as compute instances.

Is Databricks Cheaper Than Traditional Data Warehouses?

Sometimes; it depends on the workload. Databricks bills compute separately from storage, keeps your data in object storage like S3, and charges for compute only while it runs, all of which can work in your favor. Idle all-purpose clusters, the wrong compute type or long Auto Stop timeouts can erase that quickly. The reliable answer comes from running the same workload on both and comparing the full bill, cloud infrastructure included.

What Is the Difference Between Databricks Standard and Premium?

On AWS and Google Cloud, Standard is gone: remaining Standard workspaces were upgraded to Premium on October 1, 2025, and Azure’s deadline was October 1, 2026. Standard lacked features such as Unity Catalog, Databricks SQL and serverless compute. See Databricks Pricing Tiers for today’s Premium vs. Enterprise comparison.

Does Databricks Charge Separately for Storage?

It depends where the data lives. Delta Lake data in your own S3 buckets is on your AWS bill at S3 rates. Some Databricks products bill storage in DSUs, and serverless workloads can add Databricks networking charges. Track storage as its own line, especially if you keep a lot of history.

Is Databricks Expensive?

It can be, usually for the same few reasons: compute that keeps running after the work is done (idle all-purpose clusters, long Auto Stop timeouts), clusters sized for a peak they rarely hit, scheduled work on all-purpose compute, and an EC2 bill nobody budgeted for. The 7 cost factors show where each one comes from.

To tell whether your Databricks is expensive, build the estimate (like our example) and compare it with what you paid on both bills. A big gap usually means idle time, oversizing or retries, and that’s where to start cutting.

Start Optimizing Your Databricks Costs Today

Databricks pricing has a lot of moving parts, but it comes down to three steps: estimate each workload (DBUs × price per DBU, plus EC2 on classic), check the estimate against what you paid, then cut where the gap is. The native route for checking works well:

1. Pull the Databricks side from system.billing.usage, filtered on sku_name, usage_metadata.job_id or cluster_id, and custom_tags. Join it to system.billing.list_prices for a list-price figure, or have an account admin import Databricks’ pre-built cost monitoring dashboard.
2. Pull the AWS side from Cost Explorer or the CUR. Activate the Databricks tags you need (ClusterId, JobId) as cost allocation tags; AWS says new keys can take up to 24 hours to appear, plus up to 24 more to activate. One catch: clusters created from a pool pass only workspace and pool tags to EC2, not cluster tags.
3. Compare line by line: billed hours, DBUs, rates and node costs. Whichever moved tells you where to look, and the cost-cutting playbook tells you what to do about it.

Now jump into your workspace, pick your noisiest workload and run the numbers!

If you’d rather not rebuild that join every month, CloudForecast’s Databricks add-on for AWS is in early access. It connects read-only, puts DBUs and the AWS servers underneath in one dashboard by workspace, cluster, owner and tag, and sends a daily report and spike alerts to your inbox, Slack or Teams. See Databricks and AWS cost monitoring.