Predictable AI spend in BigQuery: Introducing cost management for generative AI functions

Google Cloud has made token-based cost management for BigQuery generative AI functions generally available, allowing administrators to set daily quotas at the project and per-user levels. These quotas are managed through Google Cloud Identity and Access Management and apply globally across all regions, preventing unnoticed cost accumulation. The system includes four daily quotas that cover both input and output tokens, including reasoning tokens that count toward Gemini platform billing.

Administrators can adjust quotas directly in the Google Cloud console under IAM & Admin → Quotas & System Limits without requiring support cases. Quotas can be tightened immediately or increased, with many requests auto-approved. Required permissions include Owner, Editor, Quota Administrator, or Service Usage Admin roles to manage these settings.

For programmatic control, quotas can also be managed via the Cloud Quotas API, gcloud CLI, or Terraform, enabling consistent guardrails across multiple projects. The Quotas & System Limits page provides historical token consumption data with customizable time ranges, helping teams detect usage spikes and optimize budgets based on trends in generative AI workloads.

Token quotas serve as a preventive, AI-specific cost control by stopping work near the limit during query execution, with separate limits for projects and users. They complement broader financial guardrails like Cloud Billing spend caps, which operate at the dollar level and may incur minor overages due to enforcement delays.