Chris Andersen, CFO at Flexera,
As AI adoption accelerates, many organisations are discovering they have far less visibility into AI consumption than they thought. As a result, businesses are finding it increasingly harder to price and budget for AI services. Goldman Sachs expects the volume of tokens businesses consume to rise 24-fold by 2030, even as the price of each token keeps falling. While the cost of token consumption is a real issue, it is an indication of a much larger challenge for finance teams.
As AI moves from pilot projects into everyday operations, finance is inheriting a new category of spend often without a reliable way to trace where the money is going. The issue many finance teams face is that AI costs rarely arrive as a clean line item. Every AI decision and action has a price tag. The challenge is that those costs are often spread across cloud platforms, SaaS contracts, data services and infrastructure, making them difficult to identify and attribute.
The result is a widening gap between how much organisations are spending on AI and how much they actually understand about that spend. Flexera’s 2026 State of ITAM report found that only 31% of organisations have visibility into their AI software costs, while just 36% have complete visibility across their broader IT estate. Reuters reported separately that 71% of companies surveyed had experienced AI cost overruns. For most organisations, AI adoption is running ahead of the accounting practices needed to understand what it costs.
Why AI costs are difficult to trace
The problem starts with how AI gets bought. Traditional software procurement runs through a small number of controlled channels, which gives finance a reasonably clear picture of what has been purchased and by whom. AI adoption has not followed that pattern. A team can sign up for a new tool without it ever touching procurement, while the AI features embedded in an existing cloud contract or SaaS subscription often go live with a price increase nobody flagged as an AI cost. AI spend is showing up everywhere.
None of this is done to obscure spend deliberately. It happens because AI has been adopted faster than the systems built to track it.
Finance teams can see the total bill, but that figure alone can’t tell them which department generated the cost, which models are being used, which agents are consuming resources or which workloads are driving consumption. It says nothing about what work was actually completed, or whether any of it delivered a measurable return.
Why this matters more as AI scales
This blind spot is often manageable while AI spend is low. However, it can quickly become a serious governance issue once that spend starts to represent a meaningful share of the technology budget, which is exactly the direction most organisations are heading in. Costs that cannot be attributed cannot be managed. They can only be tolerated until the point where a finance director asks a straightforward question about return on investment and nobody in the business can answer it with certainty.
This is where the overrun figures from Reuters start to make sense. Overruns rarely happen because AI is inherently expensive. They usually occur because organisations lacked visibility into consumption until after the bill arrived. Usage-based pricing and unpredictable token consumption, combined with the sheer number of entry points for AI spend, mean that waiting for the invoice is no longer a viable form of oversight.
Visibility before optimisation
Freezing AI adoption isn’t the answer either. What’s needed is a shift in how organisations account for what they are already spending, starting with an accurate, unified view of AI cost that spans cloud, SaaS and infrastructure instead of treating each as its own expense stream. Spend also needs to be attributed to the team or project driving it, so cost and value sit side by side rather than cost being judged on its own months later. Visibility only holds, though, if governance is built around it.
Enterprise leaders need clear accountability for how AI tools are introduced, governed and measured. Otherwise, spend can begin long before anyone understands the long-term financial implications. A default ‘yes’ on a small invoice today can quickly show up as an unquestioned six-figure token bill a year from now.
Usage needs to be checked against outcomes on a regular cycle and not just when the budget review forces the question. That’s the difference between teams reacting to AI costs and teams shaping the decisions before they happen.
Organisations that get this right won’t necessarily spend less on AI. They’ll be the ones who can say exactly what their investment is costing and what it’s returning, and that confidence is what lets a programme scale rather than stall. The businesses building that visibility now are setting themselves up to invest boldly over the next year, not because they’re being more cautious, but because they finally know what the number actually is. Ultimately, the question isn’t simply what AI costs. The more important question is what it returns. What business value does it create? What outcomes does it improve? What’s it worth? Organisations that can answer those questions with confidence will be in a far stronger position to scale AI adoption while maintaining financial discipline.