Skip to content
Theme
OnDevice Tools

GadgetsLaptops & PCs

Business AI costs push over half of firms past budget

New Capgemini research finds 59% of executives overspent on AI, with token, training and governance costs all rising even as model prices fall.

Illustration of business leaders reviewing rising AI cost charts on a laptop screen

Key takeaways

  • Capgemini's September 2026 survey of 500 senior executives found 53% are worried about AI costs.
  • 59% said their expected AI spending was exceeded, and 20% went more than 30% over budget.
  • Falling token prices have not lowered bills, as training, governance and oversight add costs.
  • Three in five executives plan to adopt a multi-model strategy within the next 12 months.
In this article
  1. What the survey found
  2. Why AI bills keep climbing
  3. The shift to multiple models
  4. Measuring value, not just cost

What the survey found

Capgemini questioned 500 senior executives at large organisations during September 2026. Of those, 53% said they are concerned about the cost of AI, and 59% reported that their expected AI spending had been surpassed.

The overruns vary in size. Around 20% of organisations went more than 30% past their AI budget target, the report said. This pressure has drawn chief financial officers into AI spending decisions more often than before.

The concern covers several types of cost, not just one. Spending on tokens, training and governance all feature, along with extra costs for human review and redoing work.

Why AI bills keep climbing

Illustration: AI, price, business, crypto and the labels AI Yield, Capgemini

The report notes that the cost of AI tokens has fallen for the most widely used models. Tokens are the units of text or data that a model reads and produces, so cheaper tokens might be expected to lower bills.

In practice, other expenses are pushing totals up. Training, governance, human review and work that has to be repeated all add to the bill, according to the research. Together, these make AI spending harder to forecast and control.

The shift to multiple models

Many organisations are responding by spreading their work across several AI models instead of relying on one. This approach is known as a multi-model strategy. Three in five of the executives surveyed said they expect to adopt one within the next 12 months.

Capgemini says the choice of model will rest mainly on cost and capability. Other factors include latency, meaning the delay before a model replies, and risk.

How companies measure spending is changing too. At present, 60% track their total AI spending against budget. Capgemini expects that figure to fall to 44% in the next six to 12 months, as firms focus more on results.

Measuring value, not just cost

Capgemini argues that businesses should judge AI by the value it delivers, not only by its price. It highlights a measure it calls AI Yield, which is designed to record and confirm the value created by an AI result. Better measurement of this kind can help justify requests to spend more on tokens.

Volker Darius, CEO of Capgemini Invent, said cost control alone is not the answer, arguing that "cost discipline alone will not create a profitable AI estate". He said organisations should weigh cost, quality and business value together, and choose the most suitable model for each task.

Darius added that as AI moves deeper into daily operations, companies should stop treating its cost as a purely technical or purchasing matter. Instead, he said, they should manage AI as "a portfolio of economic outcomes".

Sources

  1. Over half of businesses admit they're facing significantly higher AI costs than they expectedTechRadar

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.