Every company is spending money on AI right now. Very few of them can tell you exactly where that money is going.
KPMG surveyed more than 2,100 senior business leaders across 20 countries for its latest AI Pulse report and found that only 26% have full, real-time visibility into what their AI systems cost to operate. The gap matters because it is tied directly to results: companies with clear cost visibility are five times more likely to report an established return on their AI investment than companies without it.
This is not a story about AI being too expensive. It is a story about AI being unpredictable. Two forces are driving that unpredictability, and neither one shows up on a typical budget spreadsheet.
The first is usage. AI tools are typically billed by consumption, meaning you pay for how much you actually use, similar to an electricity bill rather than a fixed subscription. The per-unit price of running AI has been falling, but usage has grown even faster.
Every time a company rolls out a new AI feature or a new use case, that adds more usage, and more usage means a bigger bill regardless of how cheap each unit gets.
The second force is your existing software vendors. Companies like Salesforce and Microsoft are building AI into the tools you already pay for, and they are raising prices to fund it.
Research firm Zylo found that business software spending rose nearly 8% last year even though companies are not buying more separate tools than before. Gartner projects global software spending will hit 1.43 trillion dollars in 2026, a jump of more than 15% from the prior year, with AI pricing changes cited as a leading cause. In plain terms: your software bill is going up whether or not you have decided to invest more in AI yourself.
There is a deeper issue underneath the cost problem: most AI projects simply are not working yet. A widely cited MIT study found that 95% of generative AI pilot projects inside companies fail to deliver a measurable return, usually not because the AI itself is bad, but because it was never properly built into how the business actually operates.
That connects directly to a separate finding from PwC, which surveyed more than 4,000 CEOs worldwide. Companies with a solid technical foundation and clear rules for using AI responsibly were three times more likely to report meaningful financial returns than companies without that foundation.
Put these numbers together and a pattern appears. Buying AI tools is the easy part. Cleaning up your data, training your people, and redesigning how work actually gets done is the hard and expensive part, and it is the part that actually determines whether AI pays off.
For any business leader watching an AI budget line grow this year, the lesson is not to spend less. It is to demand visibility before adding more tools. If you cannot see what you are already spending and why, adding another AI feature will only make the picture blurrier, not clearer.