According to The Register, a recent survey by one of the "Big Four" accounting firms, KPMG, reveals that many corporate executives are being "scared" by AI's new pay-as-you-go billing model. In the past, companies could use fixed-price contracts to have AI companies subsidize the costs of large language models. However, as computing power prices continue to rise, the entire tech industry has been forced into a defensive position, making the era of low-cost AI usage unsustainable.

KPMG's survey covered 2,145 senior executives across 20 countries or regions, uncovering an unexpected reality. As many as 29% of respondents did not know where the rising AI costs came from, and nearly one-third of executives admitted they didn't understand the economics of AI. This issue is directly affecting the actual deployment of AI in work scenarios. The report authors pointed out that as pay-per-use models become more common, many organizations are still building the capability to effectively predict, monitor, and manage AI spending.

When the Billing Bill Turns, All the Problems Are Exposed

In short, one-third of executives haven't figured out how to truly leverage AI efficiently. Once AI is no longer as easy to use as a monthly subscription, the problems are exposed once the billing bill starts turning. Companies used to be accustomed to the "monthly subscription model," but now they are moving to real-time billing based on token consumption, which makes costs go from predictable to uncontrollable, greatly increasing budget management difficulty. Many companies set their AI budgets at the beginning of the year based on old pricing models, only to find that their bills far exceed expectations, forcing management to reassess the cost-benefit ratio.

This finding also confirms the judgment of many employees who are forced to use AI tools in their work: many enterprise leaders see AI as an immediately usable cost-cutting tool but don't really understand how to use it. From procurement decisions to implementation, the role of AI within enterprises remains in a fuzzy area — seen both as a necessary strategic opportunity and as an immediate efficiency tool. The gap between these two expectations is the root cause of the current challenges in enterprise AI deployment.

From "Buying Tools" to "Calculating Detailed Accounts," Enterprise AI Enters a Rational Phase

Current enterprise AI applications are going through a key turning point. When computing power costs shift from being subsidized by AI companies' "acquisition incentives" to becoming user-facing "operational expenses," companies can no longer blindly chase trends. Instead, they need to establish a full-chain cost management system, from procurement to monitoring and optimization, just like managing any core infrastructure cost. While KPMG's survey reveals some confusion, it also means that companies that truly understand the economics of AI will gain a first-mover advantage in the next round of competition.