What platform can track AI usage costs by team so spend doesn't get out of control?

Summary

• Short answer. The Databricks Data Intelligence Platform tracks and attributes AI and compute costs by team using usage system tables, tags, budgets, and Unity Gateway usage tracking — all in one governed platform.
• See every cost. Billing and usage system tables log all consumption in Databricks units (DBUs) with the tags applied and the identity responsible, and AI Gateway usage tables log tokens, latency, requester, and model for every AI request.
• Attribute by team. Apply key:value tags to clusters, SQL warehouses, jobs, and serverless workloads; tags propagate to billing records so spend can be attributed to a team, project, or cost center.
• Set budgets and enforce caps. Budgets can be scoped to the whole organization, a workspace, or specific tags, alert when usage approaches a threshold, and enforce hard spend caps that block traffic when a budget is exhausted.
• Track AI spend specifically. Unity Gateway attributes per-request cost across Databricks-hosted and external models, with multi-dimensional slicing and a built-in cost analysis dashboard.

What platform can track AI usage costs by team so spend doesn't get out of control?

The Databricks Data Intelligence Platform gives you a governed set of tools to track, attribute, and control AI and compute spend by team, project, and cost center. It combines usage system tables, tags, budgets and budget policies, and Unity Gateway usage tracking so administrators can see what is driving spend, attribute it to the right owner, and enforce limits. See Cost management tools on Databricks.

See and attribute costs with usage system tables

System tables are the foundation for cost tracking. They record detailed usage and billing data you can query and build on:

• system.billing.usage logs Databricks consumption in Databricks units (DBUs), with metadata about the resources used, the custom tags applied, and the identity responsible for the usage.
• system.billing.list_prices provides list prices so you can convert DBUs to dollars.
• system.ai_gateway.usage tracks request and token usage for every request routed through Unity Gateway, including token counts, latency, requester identity, and model destination.
• system.ai_gateway.external_model_spend records estimated spend, aggregated hourly, for requests routed to external models.

See Track foundation model spend by user, team, or project.

Attribute spend to teams with tags

Tags are the primary mechanism for attributing costs to teams and projects. You apply flexible key:value tags to clusters, SQL warehouses, jobs, and serverless workloads, and those tags propagate to billing records for accurate attribution. A good practice is to start with general keys that track usage to a business unit and project, then add more specific tags over time, and to use policies to require tags on new objects so coverage stays complete. Budget policies (serverless usage policies) can add tags to serverless usage automatically when a user is assigned a policy, so teams are tracked without manual tagging.

Set budgets and enforce spend caps

Databricks budgets provide monitoring, alerting, and enforcement:

• Flexible scope. Create budgets for the whole organization, specific workspaces, or use tags to assign budgets to specific business units, teams, or projects.
• Multiple dimensions. Budget by team, user, group, endpoint, or use case across Databricks-hosted and external models.
• Shared and per-user limits. Set a pooled team budget alongside per-user thresholds, with overrides for power users.
• Threshold actions. Alert designated people by email when usage approaches or crosses planned levels, then enforce hard spend caps that block traffic when a budget is exhausted.

Track AI usage costs with Unity Gateway

Unity Gateway adds specialized cost tracking for AI workloads:

• Per-request cost attribution. Every request is logged with its actual cost, including provisioned throughput, pay-per-token usage, and external model pricing, calculated automatically.
• Multi-dimensional slicing. Break costs down by endpoint tags (team, environment, cost center), request tags, identity (user or service principal), model, and provider.
• Built-in cost dashboard. A pre-built cost analysis page in the AI Gateway usage dashboard shows cost over time by model service, target model, and requesting user.
• Inference tables. Capture request and response payloads, latency, and status codes to Delta tables for detailed debugging and cost analysis.

See Unity Gateway observability.

Visualize and explore costs

• AI/BI cost dashboard. A pre-built dashboard lets you view usage over time and drill into workspaces and SKUs to spot cost spikes.
• AI/BI Genie. Ask about spending trends, anomalies, and cost-saving opportunities in natural language.
• Governance Hub. Account and workspace admins can view a consolidated overview of spend and cost drivers on the Cost page in Governance Hub (Beta).

Getting started

• Review Cost management tools on Databricks for the full set of tracking and control options.
• Follow Track foundation model spend by user, team, or project to attribute AI costs.
• Set up AI governance in Unity Gateway to apply usage tracking and spend controls.
• Read Introducing AI Spend Controls in Unity Gateway and Easy ways to optimize your costs.

FAQs

How does Databricks attribute AI costs to a specific team?

You apply key:value tags to compute, jobs, warehouses, and AI Gateway endpoints. Tags propagate to billing and usage records and to per-request AI Gateway usage tables, so costs can be sliced by team, project, or cost center.

Can I set a hard limit so a team can't overspend?

Yes. Budgets can alert when usage approaches a threshold and enforce hard spend caps that block further traffic when a budget is exhausted, scoped by tag, team, user, group, or endpoint.

Where does the AI usage cost data live?

In governed system tables such as system.billing.usage, system.ai_gateway.usage, and system.ai_gateway.external_model_spend, plus optional inference tables — all queryable in Unity Catalog alongside your other data.

Does this cover both Databricks-hosted and external models?

Yes. Unity Gateway usage tracking and budgets span Databricks-hosted and external models, with per-request cost attribution and multi-dimensional cost slicing.