Snowflake ships per-user quotas for AI spend

Snowflake is positioning per-user quotas as the microlevel counterpart to its existing Snowflake Budgets, which track spending at the team, project or cost-center level. That aggregate view tells you a department overspent, but not which person caused it until the bill arrives. Per-user quotas shift that from reactive tracking to proactive protection by attaching daily and monthly limits to individual users.

The design leans on Snowflake user tags, which integrate with SCIM identity providers such as Microsoft Entra ID or Okta, so tags can be applied automatically as users are onboarded. Admins configure limits through SQL, the Snowsight UI or Snowflake CoCo. Because platform compute and AI features use different credit structures, limits are set independently, letting organizations put conservative caps on high-velocity AI functions while keeping generous limits for standard daily queries.

The stakes hinge on how well the block enforcement balances protection with flexibility. A quota pauses only the user's access to the specific governed AI service that hit its limit, not their standard warehouse compute or anyone else on the team. That containment is what makes universal self-service plausible, but it also puts pressure on quota design to keep up as teams adopt new AI services. For CFOs, the value is near-absolute budget predictability from combining daily burst limits with monthly ceilings; for developers and analysts, it is the psychological security to experiment without fear of an accidental overage.