Plan Microsoft Fabric capacity with clearer, usage-based cost signals. Effective October 1, 2026, Microsoft changes how consumption is measured for selected Microsoft Fabric artificial intelligence (AI) experiences.
Summary
- A dynamic consumption model replaces the fixed model for selected Fabric AI experiences
- Consumption remains billed in Fabric Capacity Units (CUs)
- Consumed capacity varies by the AI model used and its reasoning effort, the complexity of the request, the tools and services involved, and the processing required
- Fabric operations that use large language models (LLMs) are affected, and further capabilities may follow
- Existing workloads keep running, but Capacity Unit consumption can change after October 1, 2026
- Microsoft will publish more details on October 1, 2026 on the Microsoft Fabric blog
What is changing
Until now, the selected Fabric AI experiences drew capacity at a fixed rate per operation. From October 1, 2026, Microsoft measures the resources that each task actually needs and charges that consumption in Fabric Capacity Units. The billing unit stays the same, while the amount consumed per task becomes variable.
Four factors drive the consumed capacity: the AI model used and its reasoning effort, the complexity of the request, the tools and services involved, and the processing required. Two requests that look comparable can therefore consume different amounts of capacity. Microsoft will publish further details on October 1, 2026.
Which Fabric AI capabilities are affected
The change applies to Fabric operations that use large language models (LLMs). Microsoft names the following capabilities and states that further ones may follow:
- Copilot in Fabric
- Fabric data agents
- AI Functions
- Fabric IQ ontologies
What is Microsoft Fabric?
Microsoft Fabric is Microsoft’s analytics platform that brings data integration, data engineering, data science, real-time intelligence, and business intelligence together in one environment. It is licensed through capacity: an organization buys a Fabric capacity, and every workload draws from the same pool of Capacity Units. Because analytics workloads and AI operations share that pool, a change in AI consumption also affects the headroom left for reporting and data processing.
Why this matters now
Capacity is a budget, not only a technical setting. When AI operations consume a variable number of Capacity Units, cost forecasting turns into continuous monitoring rather than a fixed calculation. Organizations that record their consumption before October 1, 2026 keep a baseline they can compare against, which lowers the risk of throttling and of unplanned capacity upgrades. Teams that use Copilot in Fabric or Fabric data agents at scale benefit most from an early review.
What organizations should do now
- Review current AI usage in the Fabric Capacity Metrics App and record a consumption baseline before October 1, 2026
- Identify the workspaces that rely on Copilot in Fabric, Fabric data agents, AI Functions, and Fabric IQ ontologies
- Monitor Capacity Unit consumption weekly during the first month after the change and compare it with the baseline
- Validate that the current capacity reservation still covers peak analytics and AI demand
- Update cost forecasts and internal chargeback models for variable AI consumption
- Follow the details Microsoft publishes from October 1, 2026 before committing to a capacity change
Ready to review your Microsoft Fabric licensing?
Check with our licensing experts whether your Fabric capacity commitment, your Microsoft agreement, and your buying program still fit your AI consumption. Contact us at https://www.schneider.systems/contact/ to optimize your Microsoft licensing position.
Sources
For the announcement of Microsoft Fabric: Dynamic Consumption Measurement for AI Capabilities, please visit: https://blog.fabric.microsoft.com/.
For useful software licensing information on Microsoft products, please visit: https://www.schneider.systems/software/microsoft/
