A survey of 515 business and IT leaders in the U.S. spanning enterprises with more than $500 million in annual revenue finds just under a third (30%) are already seeing their organizations consume more than 10 billion tokens per month when accessing artificial intelligence (AI) tools and applications.
Conducted by Deloitte, a total of 61% expect their organizations to consume more than 10 billion tokens per month by 2028.
That level of consumption makes it clear most organizations are going to need to adopt best practices for optimizing prompts to reduce costs, says Chris Thomas, U.S hybrid cloud infrastructure leader for Deloitte. “There is going to be a lot more focus on tokenomics,” he adds.
Similar to the best FinOps practices that many organizations have adopted to control cloud costs, the need to ensure prompts are efficiently consuming infrastructure resources is becoming critical, notes Thomas. In fact, a full 86% of respondents expect AI infrastructure budgets to increase in the next three years, with allocations expected to, on average, more than triple, the survey finds.
The degree to which those costs are sustainable will vary from one organization to another, but they do suggest that as AI applications become more complex, the number of more expensive reasoning tokens consumed will increase. For example, nearly all respondents (96%) rate their AI workloads as being of a medium or high level of complexity.
Nearly two thirds of respondents (64%) have also already started limited or at-scale deployments of AI factories, with 88% expecting to achieve that goal by 2028. Nearly three quarters (73%) expect to be running AI factories at scale within the next three years, with nearly all respondents (97%) reporting they are confident in their ability to run AI at scale.
At the same time, AI workloads are moving to the network edge, with well over a third (36%) having deployed them at the network edge. Just under three quarters (72%) expect to achieve that goal in the next three years.
Not surprisingly, 81% of respondents said they believe IT teams have the technical and financial acumen to manage AI projects at scale, with just over half (51%) reporting IT leaders own AI infrastructure integration decisions.
Overall, the percentage of respondents with a high number of AI pilots (31 or more) was almost 50% in 2025 and is expected to jump to nearly 70% by 2028. Similarly, the percentage of respondents with the highest AI production-ready use cases (31 or more) is expected to go from 44% in 2025 to 67% by 2028.
Given the number of projects, it’s apparent business and IT leaders will need to closely monitor costs, says Thomas. Just about every aspect of the AI supply chain from silicon to personnel is currently constrained, so it is not likely any of those costs will be declining soon, he adds.
Hopefully, as the number of projects organizations launch increases it will become simpler to manage AI applications at scale. In the meantime, however, while IT teams might hope for the best they may want to start planning now for any one of many possible worst-case scenarios.
