Dive Brief:
- AI costs are getting harder to control as the rush to deploy agents outpaces enterprises’ ability to monitor the technology, according to Splunk’s State of Observability 2026 report. The Cisco-owned data and observability platform surveyed 2,050 ITOps and engineering professionals for the report.
- More than half of respondents said they cannot control AI costs as each prompt, re-run and token comes with its own price tag, the report found. Another 60% said unexpected AI-related billing spikes create operational strain over high baseline costs.
- The cost of AI has transitioned from a fixed line item to a variable cost that compounds at scale with little visibility, Mitch Ashley, VP and practice lead for CIO and technology buyers at The Futurum Group, told CIO Dive in an email. “A monthly report can’t govern a cost that spikes in one afternoon,” he said.
Dive Insight:
Unpredictable AI costs are leading CIOs to implement tools to track spend and strategies for cost management.
“The most useful tools track spend back to its source and manage what’s consumed at that point of origin,” Ashley said. “They tie tokens and calls to a specific agent, workflow, team, and trace, and enforce AI FinOps budget thresholds and rate limits at the agent operational layer before a request is sent.”
Beyond using tools to track spend and control costs, CIOs should assign cost ownership to users who start an AI workload, include a named budget for the agent or workflow at the outset and conduct a review if the workload drifts, Ashley said.
Routing routine requests to smaller, cheaper models and writing consumption caps into contracts can also help manage AI costs, Ashley said.
Vendors are responding to the AI cost management challenge. In August, Google launched flexible billing options and cost controls for AI agents, such as a monthly cap on AI spend. Oracle earlier this year rolled out limited token bundles to better tie agentic pricing to business value, and AWS previewed a FinOps agent to investigate cost anomalies.
AI is also prompting enterprises to look to observability tools for greater visibility into increasingly complex technology environments, an Enterprise Management Associates report found. However, enterprises are also struggling with observability tool sprawl. Three in four enterprise IT organizations are using up to 12 observability tools across cloud, infrastructure and network environments, according to the consultancy. That, too, adds to mounting AI costs.
Agents are also affecting what questions observability tools in particular need to answer, Ashley said. Observability tools should be able to explain agent resource consumption, decision loops and system interactions, and CIOs will want to press observability vendors on “how they price, sample and retain AI telemetry,” he said.
“AI workloads emit a trace for every prompt, tool call, and model response, so the observability bill climbs with the model bill,” he said.
