Hi everyone,
I wanted to share some insights and clarifications around Databricks Clean Rooms, especially regarding how they are designed and how costs are calculated — something that can be a bit confusing when onboarding collaborators.
What Are Clean Rooms?
Databricks Clean Rooms are secure, governed environments that allow multiple organizations to collaborate on sensitive data without exposing raw datasets. They are ideal for privacy-preserving analytics in industries like advertising, healthcare, and financial services.
Clean Rooms are powered by Delta Sharing, which enables live, read-only access to shared data across workspaces and clouds — without replication.
Design Overview
When a clean room is created:
- A central clean room is hosted by Databricks.
- A clean room object is instantiated in each collaborator’s Unity Catalog metastore.
- This object allows collaborators to share data assets (tables, views, notebooks) and approve notebooks for execution.
Importantly, a collaborator is defined by a #CleanRoom Sharing Identifier, which includes:
- Global Metastore ID
- Workspace ID
- User Email Address
This means that the clean room object is not tied to a single user, but rather to the organization’s metastore. Within that metastore, multiple users can be granted access to the clean room object using Unity Catalog privileges like MANAGE, EXECUTE CLEAN ROOM TASK, and BROWSE.
Cost Model Clarification
Here’s the key point that often causes confusion:
Databricks charges \$50 per collaborator per day — and “collaborator” means per user email.
Even if multiple users share the same meta store and workspace, each user who accesses the clean room is billed individually. There is no prorated billing — the charge applies even if the user accesses the clean room briefly or performs no operations.
Additional charges include:
- Serverless compute (Jobs Serverless on AWS or Automated Serverless Compute on Azure)
- Storage (Databricks Storage SKU)
- Data transfer, if applicable
Questions I’ve Been Exploring
As I’ve tried to understand the model better, here are the questions I’ve asked — and I’d love input from the community:
What does it mean when Databricks says “a clean room object is created in each collaborator’s Unity Catalog”?
- Is a collaborator defined by meta store ID, or is it per user email?
- Does each user need a separate clean room object?
Can a clean room object be shared among multiple user emails in the same meta store and workspace?
- If yes, why is the cost still calculated per user email?
Is the billing model misaligned with the design?
- The object is shared, but the cost is per user — is this intentional or a limitation?
Is there a way to centrally manage or limit which users can access the clean room to avoid unexpected charges?
Can we create a shared service account or group access model to reduce per-user costs?
Is Databricks planning to offer a more flexible billing model — e.g., per meta store or per workspace instead of per user email?
How are teams tracking and auditing clean room usage to ensure cost transparency?
Summary
- Clean room objects are created per Meta store, but access can be granted to multiple users.
- Billing is per user email — each user incurs a separate daily charge.
- This model ensures granular tracking and governance of clean room usage.
- I hope this helps clarify how clean rooms work and how costs are applied. If anyone has experience optimizing collaborator access or managing costs across large teams, I’d love to hear your strategies!
