It’s about surprise charges…
For the past two years, the biggest challenge for organisations was figuring out how to use AI. Now many are facing a different problem: Understanding where AI is being used, what value it’s creating and accurately tracking what it’s costing them.
Splunk’s State of Observability 2026, based on a survey of more than 2,000 ITOps and engineering professionals worldwide, including in New Zealand and Australia, found 55 percent of organisations can’t accurately control AI costs, while 60 percent say unexpected billing spikes are creating more operational strain than the underlying cost of AI itself. For New Zealand, it’s an even bigger issue, with 68 percent saying they can’t accurately control AI costs.
“For three years, AI ran like a developer preview with pilot projects, flat-rate plans, and subsidised inference. That era is ending.”
The conversation is shifting away from experimentation and towards management, as businesses discover that deploying AI at scale is creating financial and governance challenges that many were not prepared for.
For many organisations, AI spending is no longer confined to a handful of pilot projects. Instead, generative AI tools, coding assistants and AI agents are increasingly being deployed across software development, customer service, security operations and business workflows. And as adoption expands, usage can rise rapidly, often in ways organisations don’t anticipate.
The rise of the surprise bill
Unlike traditional software licenses with relatively fixed costs, AI consumption can fluctuate significantly depending on how heavily employees, customers or automated systems use the technology. That makes forecasting difficult and creates the potential for unexpected jumps in spending.
“For three years, AI ran like a developer preview with pilot projects, flat-rate plans, and subsidised inference. That era is ending. Spend now tracks directly to tokens, and usage is increasingly hard to predict,” the report notes.
It found organisations are less concerned about AI’s baseline cost than they are about unpredictability. Surprise billing spikes were the primary operational strain for 60 percent of respondents.
Most businesses can budget for a known expense. It becomes much harder when usage changes unexpectedly, new AI services are introduced, or successful projects scale more quickly than anticipated.
“The instinct is to assume falling token prices mean falling costs. The opposite is true,” the report notes. “Consumption is growing faster than prices are dropping, so total spend keeps climbing.”
The issue isn’t simply that AI is expensive, but that many organisations no longer know exactly what drives the bill. “When organisations run dozens of agents across multiple models, the costs stack up in ways that are incredibly difficult to track in real time.”
The report argues this represents the largest operational spending shift since the emergence of cloud FinOps.
Connecting AI to business value
And it’s not just about the cost – it’s whether that cost is actually producing anything worthwhile. “Without visibility into agent behaviour, the performance of the entire AI stack and token consumption, teams can’t connect spend to outcomes. And if you can’t make that connection, every dollar looks the same — whether it’s a good investment or pure waste.”
At the same time, organisations are continuing to invest heavily in AI despite widespread concerns about governance and oversight.
More than four in five respondents – 83 percent – said they are concerned about unintended consequences from AI agents. (Australian concerns are lower, at 79 percent). Yet enterprises continue to accelerate adoption as they seek productivity gains, automation opportunities and competitive advantages.
The result is a growing tension between enthusiasm and control, with the report suggesting businesses are still working through questions around accountability, measurement and value, even as AI becomes increasingly embedded in day-to-day operations.
The findings suggest a growing divide between organisations that have established clear controls around AI and those that have not. Businesses with stronger governance practices are significantly more likely to report confidence in their ability to manage AI costs and understand how AI workloads contribute to business outcomes.
