New Fabric AI Billing: Higher Variable Cost

• Fabric AI operations changed from a fixed rate to a variable rate based on the tool, model, and complexity.
• Rates can differ by five times between AI Functions models and four times between Copilot and Fabric IQ, so similar workloads consume capacity resources at very different rates.
• Admins and developers should reevaluate AI workloads in Fabric and confirm that the tools and models selected are still the most cost-effective choice.

Fabric’s change to AI metering moves cost risk from Microsoft to its customers. Beginning Oct. 1, 2026, most Fabric AI operations switched from a fixed rate to a variable rate that’s determined by the tool used, the model selected, and the complexity of the request. The change allows Microsoft to charge more for larger, reasoning-heavy models, which agents and Fabric IQ ontology queries use. There’s no immediate concern about existing workloads failing, but the cost for many processes will go up. Customers should reevaluate the tools and models used and ensure that the lowest cost configuration for the task at hand is selected. Surprisingly, for some workloads, the best cost choice may be to move the process outside of Fabric entirely.