The question for business leaders is no longer whether AI works. It is whether their organization can afford to run it at scale—and continue to get meaningful value from it.
In a recent Fortune commentary, IBM Chairman, President and CEO Arvind Krishna argues that enterprise AI economics turn on three questions: what drives the cost, who sets the price, and what the organization is actually spending.
AI is becoming an operating cost
As AI moves from pilots into everyday workflows, inference becomes a recurring operating expense. The total cost depends on more than a model’s listed price. It can also reflect repeated processing, the amount of context a system handles, infrastructure utilization, and the engineering needed to keep workloads reliable and governed.
Krishna points to caching and parallel processing as ways to reduce repeated work. In an IBM test described in his article, the same hardware handled seven times more work after those optimizations. The broader lesson for teams is to examine how a system is designed and used, not just the price of an individual model call.
Three questions to make AI spending clearer
- What is driving the cost? Look at the workflow end to end: model usage, repeated context, infrastructure, and the work required to operate the system.
- Who controls the price and the underlying economics? Compare provider pricing with the flexibility and responsibility involved in running models on infrastructure your organization controls. The right choice depends on the task, quality requirements, security needs, and operating capacity.
- What are you spending—and what is that spend accomplishing? Make costs visible by workload, then compare them with useful outcomes such as time saved, decisions improved, or work completed.
Optimize for value, not just token price
The lowest-priced token is not automatically the best value. A cheaper option can cost more overall if it needs repeated attempts, misses the required quality, or adds operational work. The more useful measure is token yield: how much meaningful work or business value each dollar of AI spend produces.
That means matching models and infrastructure to the work. A task that is exploratory may have different quality, latency, and cost requirements than a decision that must be highly reliable. Clear measurement helps teams make those trade-offs deliberately instead of restricting usage across the board.
Start with the work your teams need to do
For a customer or business team, a practical review starts with a specific workflow and the outcome it should improve. Track the cost and quality of that work, identify avoidable repetition, and decide which requirements—such as accuracy, speed, privacy, or control—matter most. Then evaluate whether the chosen model and operating setup meet those needs as usage grows.
AI economics are shaped by decisions made around the model: how data and workloads are handled, how costs are monitored, and how performance is judged. The aim is to give teams room to use AI where it helps while keeping spending understandable and tied to outcomes.
Source: Arvind Krishna, Chairman, President and CEO of IBM, “IBM CEO: Rationing AI Is a Slow Way to Lose,” Fortune, November 5, 2026.
If your team is weighing how to scale an AI initiative, I’m happy to compare notes on the questions you’re working through.
Newsletter
Imran's Insights
Perspectives to help your team navigate AI, software delivery, engineering and high-performance computing.
