Unexpected High Cost on Simple Use Case

Question

Has anyone had challenges controlling cost on IQ, specifically from Ontology? 

 

We enabled IQ based on these instructions at tenant level: Ontology (preview) required tenant settings - Microsoft Fabric | Microsoft Learn

 

The sample data we are working with has about 11 tables and ~100k records total across all tables. We spent a few hours during one day experimenting to set up an ontology on top of this schema and it cost us ~$200. Checking the Metrics app we can see this came from underlying Graph usage. 

 

Does anyone know how to better control this cost? 

Answer 1

Post/ask in the Fabric community:

Welcome to the Fabric Community

Answer 2

IQ Ontology and its underlying Graph item consume Fabric capacity continuously while models are running and while graph operations (indexing, refresh, queries, AI reasoning) are executed. Even with relatively small data volumes, a few hours of experimentation can generate noticeable cost if the ontology and graph are left running at full capacity or refreshing frequently.

To better control cost for this scenario:

1. Pause capacity when not actively using IQ
   • Fabric capacities can be paused and resumed to avoid charges when experimentation is not in progress.
   • After finishing ontology/graph work for the day, pause the Fabric capacity and resume it only when work needs to continue. This directly limits model uptime charges.
2. Reduce model uptime and complexity
   • Ontology charges are affected by how long the ontology model is running and by model complexity (number of definitions, relationships, etc.).
   • During early experimentation, keep the ontology as small and simple as possible and delete or consolidate unused entity types and relationships.
   • Close or stop test ontologies that are no longer needed so they do not keep consuming capacity.
3. Tune Graph refresh behavior
   • The Graph child item of the ontology can be configured to refresh automatically on a schedule, and each refresh consumes capacity.
   • Review the Graph item’s refresh schedule in the workspace and either:
     • Reduce refresh frequency to the minimum needed for testing, or
     • Temporarily disable scheduled refresh while experimenting and trigger refresh manually only when required.
4. Manage ontology operations and AI usage
   • Ontology logic and operations (queries, indexing, refresh rates, idle time) all contribute to capacity usage.
   • Limit heavy or repeated graph traversals and complex queries during testing.
   • AI reasoning and natural language queries are billed based on tokens used; keep AI-based querying to what is strictly needed while validating the model.
5. Monitor usage patterns
   • Use Fabric monitoring tools and the Metrics app to identify which ontology or graph operations are driving the most consumption, then adjust refresh schedules, query patterns, and model size accordingly.

These steps focus on reducing model uptime, controlling automatic graph refresh, and limiting expensive query/AI operations, which are the main levers for lowering IQ Ontology and Graph costs.

---

References:

• Capacity consumption for ontology (preview)
• What is Fabric IQ (preview)?
• What is ontology (preview)?