The Decisions That Define AI At Scale

Nearly nine in 10 enterprises have already limited or redesigned AI use cases due to rising token consumption.

That finding from Cloudera’s 2026 State of Enterprise AI survey captures what can happen when an AI use case gains traction. An experiment may be designed around a predictable level of demand, only for its economics to change when usage grows. The infrastructure gets tested, too. Production can expose performance limitations or make an existing deployment prohibitively expensive, while the data powering the application must remain protected wherever the workload runs.

Enterprise AI is reaching a point where growth depends on what organizations can sustain. Decisions made early in an AI project can look very different under the demands of production, which is forcing enterprises to reconsider how their applications are built and where they run.

The Economics Of AI Are Changing The Equation

A promising AI use case can become a victim of its own success. More users generate more inference, and a workload that looked economical during development can take on a very different cost profile once it is used throughout the business.

For many enterprises, the economics are already finely balanced. Fifty-eight percent say the value generated by their AI initiatives is approximately equal to what they cost, while less than a quarter (23%) say AI generates more value than it costs.

Yet precision around those costs remains limited. Only 27% say they can measure end-to-end AI costs very accurately. That leaves enterprises making decisions about growth without always having a clear picture of what additional usage will mean financially.

The consequences now reach the infrastructure layer. Forty-five percent have already moved an AI workload to another cloud, a private environment or an on-premises platform due to higher or more volatile costs.