AI costs are becoming harder for businesses to predict, Splunk report says

Businesses are finding AI costs harder to predict as vendors shift toward consumption-based pricing and AI agents take on more work, according to Splunk’s State of Observability 2026 report, as described by TechRadar’s Craig Hale.

Agents can operate independently and use widely varying amounts of tokens depending on the task, its complexity and the model selected. Among surveyed organizations, 55% say they cannot effectively control AI costs. Meanwhile, 57% are concerned that telemetry costs exceed the cost of the infrastructure they monitor.

The report also points to closer collaboration between observability and security teams. Four in five organizations (81%) say monitoring AI workloads has increased that collaboration. Some 73% say they can accurately trace problems to underlying security causes, up from 65% last year. Cisco Chief Product Officer Jeetu Patel said observability and security data are converging and teams need to collaborate more.

Among observability leaders, 91% say they can accurately attribute AI costs to particular agents, models and workloads, compared with 69% of non-leaders. The report also says three in five respondents find unexpected AI bill spikes more financially stressful than a consistently high baseline cost, amid the shift toward consumption-based pricing.