10/08/2026

Databricks Budgets – Budget First. Prompt Later.

The missing piece to your Databricks AI Governance.

As AI workloads grow from small experiments to production-scale applications, keeping cloud costs under control becomes just as important as achieving model accuracy and performance. Databricks Budgets help organisations proactively manage spending by allowing teams to set budget thresholds, monitor consumption, and receive alerts before costs exceed expectations. Whether you’re running data pipelines, training machine learning models, or serving generative AI applications, budgets provide a simple but effective guardrail that improves financial visibility, encourages accountability, and helps prevent unexpected cloud bills.

In this blog we outline the pricing model of Databricks’ Genie Products, instruct you how to set up the budgets to allocate spending limits for groups and users, alert when thresholds have been met and most importantly prevent you from running out of budget.

Motivation

As of recently – 8^th of July 2026, the Databricks pricing for AI changed. Before the 8^th of July, only the compute resources that were needed to run the Databricks Genie products (listed here) were charged. However, this is not the case anymore, and all Databricks Genie products now follow the Pay-As-You-Go pricing model. For every DBU spent over the free monthly allowance of 150 DBU per user, one pays for each DBU spent on the AI product, on top of the compute that is used.
This could have a significant impact on your Databricks spending and precaution is needed now more than ever.

Solution

The threat of misusing AI by using it too much is occurring in every organisation nowadays. To mitigate this risk, Databricks implemented Databricks Budgets. The latest enhancements to the Budgets are specifically designed for governing AI usage. The Databricks Budgets are in General Availability.

Setup

To get started, the only requirement is to be a Databricks account admin. This can be set in the following URL https://accounts.azuredatabricks.net/user-management/users for users, groups or service principals.

Keep in mind that a Databricks account admin can change the budgets through the UI, whereas a Billing admin user only allows them to change budgets through API calls and therefore is more limited.

*Figure 1 — Prerequisites to setting Databricks budgets*

Open the Databricks admin console and go to Usage > Budgets (Preview) or follow the url https://accounts.azuredatabricks.net/usage/budget

*Figure 2 — Overview of the Databricks Budgets menu*

To create a new budget, press ‘Create budget’ as highlighted in the image above.

Once you are in the create budget window, the fields that need to be completed are:

Name of budget (mandatory) – for display purposes
List of workspaces – which workspaces is the budget applied to
Resource types (mandatory) – category of resources that checks are carried against, with options ‘All’ or ‘Unity AI Gateway’
Resource tags (optional)
Shared thresholds (mandatory) – threshold alerts or blocking of usage that are applied to the shared accumulated consumption of every user, group or service principal for the given resource type, tags and workspaces *
Per-user thresholds (optional) – threshold alerts or blocking of usage that are applied to all users individually or for specific users for the given resource type, tags and workspaces. This overrides the shared per-user thresholds for given users.

This is how the ‘Create budget’ menu looks like.

*Figure 3 — Overview of the Databricks Budgets ‘Create budget’ window*

When the Resource types option is set to ‘All’, the only option is to have alerting for all users and the per-user thresholds are disabled.

*Figure 4 — Databricks ‘Create budget’ window for all resource types*

Only when the Unity AI Gateway is selected, the option to override per-user thresholds is enabled.

*Figure 5 — Databricks ‘Create budget’ window for Unity AI Gateway resource types*

If the resource tags are not selected, only alerting thresholds are enabled for all users or the specific per-user thresholds.

*Figure 6 — Databricks ‘add shared threshold’ window for Unity AI Gateway without tag databricks-product: genie*

However, once the tag ‘databricks-product: genie’ is selected, the option to block usage is enabled apart from sending alerts only.

*Figure 6 — Databricks ‘add shared threshold’ window for Unity AI Gateway with tag databricks-product: genie*

This setting is crucial because it is the guardrail that will stop your organisation from going over the allowed thresholds.

It is possible to set more than one alert in the ‘Shared Thresholds’ category, but only one block statement. Interestingly, the block value is possible to be lower than the value of the biggest alert.

*Figure 7 — Databricks ‘Shared thresholds’ for Unity AI Gateway with tag databricks-product: genie*

This is because the budget might be exceeded slightly beyond the blocking threshold, which depends on how quickly the Databricks syncing happens and how costly the prompts that exceeded the blocking threshold were.

The ‘Per-user thresholds’ allows each individual user to receive an allocation specifically for them.

*Figure 8 — Databricks ‘Per-user thresholds’ for Unity AI Gateway with tag databricks-product: genie*

In the example above, all users are allowed to spend 15$ on Genie products, except for one which is an exception and has a slightly bigger allowance of 20. The actions when the threshold is exceeded are inherited from the default setting and this cannot be customized, for example to alert each user individually when they exceed their threshold, rather than the personnel that is responsible for governing the costs. In addition, if both block and alert rules are defined for all users, they apply to each users’ individual overrides.

Lastly, it is important to note that the per-user thresholds can be set on either of the following levels:

Users
Service principals
Databricks groups

This does not include Microsoft Entra ID groups by default, unless they are present as a Databricks group. It is important to note that service principals do not receive free monthly usage and are billed for all of their usage.

For now, the budgets that are set are scoped to a time window of a calendar month. This is subject to change in the future.

Good practices

The first thing we advise all organisations to do is to create initial versions of your Databricks budgets and to begin setting blocking thresholds to a value that is reasonable for your organisation. This will ensure that you have your first guardrail in place, and you can start improving and adapting it to your needs.

We recommend that you do this at least on the scope of all your workspaces and then set up the fine-grained per-workspace budgeting soon afterwards.

Furthermore, we recommend setting the blocking threshold slightly below the allocated budget. For example, if your organisation plans to allow 100$ per month on Genie products and 10$ on each user except for one admin who has 20$ allowance, we advise setting up your first budget as follows:

Alert when 80% of the budget has been used (in this example 80$)
Alert when 90% of the budget has been used (in this example 90$)
Block Genie products until the end of the month when 95% of the shared budget has been used (in this example 95$)
Alert when 100% of the budget has been used, as you might slightly exceed the blocked amount. It is important to know how much it exceeded.
Alert and block each user when they reach 95% of their allowance (in this example 9.5$) except for one admin who is allowed 19$ monthly.

*Figure 9 — Databricks create budget recommended starting point*

The budget alerts and thresholds are great features to improve your awareness of being close to the allowed budget, but do not allow for a good overview of where the organisation as a whole or certain teams/departments/users are in their budget. Databricks provides native dashboards to each budget to track the consumption

*Figure 10 — Databricks monitoring of existing budgets*

The last advice is setting up the Databricks Budgets using Infrastructure as Code. The industry-standard tool Terraform already supports this and is documented at https://registry.terraform.io/providers/databricks/databricks/latest/docs/resources/budget. Together with source control, this will allow your organisation to see how the budgets evolved over time, restore them to a stable and desirable configuration in case they get deleted or change to a version that is not desired.

What can Databricks Budgets do for you?

✅ CAN DO	❌ CAN’T DO
Limit shared spending of Genie products	Blocking of shared spending of non-Genie products
Limit users to spending a monthly budget assigned specifically for them for Genie products	Blocking of individual usage of non-Genie products after individual or shared budgets are exceeded
Warn a list of people that individual users are close to going out of budget for Genie products	Help users optimize their prompts to lower budget spent
Override the default per-user spending limits for certain users on Genie products	Overrides for non-Genie products
Alert within a 24-hour window of budget exceeded	Alert immediately after budget is spent
Setting limits and alerts based on consumption in USD	Set thresholds for spending on compute resources on Genie products
	Consider billing discounts when calculating budget usage
	Be in immediate sync with system table system.billing.usage as it updates less frequently
	Budget details could be delayed after creation and are not applicable immediately
	Setting limits and alerts based on consumption in EUR

Is your organisation having difficulties with AI governance?

We help organisations not only implement AI solutions, but also empower governed and responsible AI usage. Get in touch with us!

Written by

Kristijan Shirgoski

Consultant @ Lytix