85% of Tech Leaders Can’t See What Their AI Costs in Real Time

AI can create value and still become economically inefficient.

That is what makes one of IBM’s latest findings so striking: 85% of technology leaders say they lack real-time visibility into AI spending. At the same time, organizations report cloud costs running 48% higher than initially projected as AI workloads consume more infrastructure than expected.

That is not just a budgeting problem. It is a decision problem.

How do you know where to put the next AI dollar if you cannot see what the current one is really costing?

Value is only half the equation

Most AI conversations begin with the upside:

• Did productivity improve?
• Did cycle time fall?
• Did revenue increase?
• Did employees save time?

Those questions matter. But IBM makes an equally important point: creating value is only half the equation. The other half is what it costs to create that value.

And AI economics are unusually dynamic. Workloads fluctuate, compute demand spikes, models require tuning, cloud usage changes, new models arrive, and prices fall. A workflow that looked economically attractive six months ago may no longer be the best way to produce the same outcome today.

Without visibility, leaders may never know the economics changed.

The business case can expire faster than the investment cycle

This is the part I think matters most. AI business cases are often built at the beginning:

• Expected usage
• Model choice
• Cost per query
• Cloud demand
• Human review burden
• Expected productivity
• Expected value

Then reality moves. Adoption grows faster than expected, usage patterns shift, review costs rise, inference prices drop, and cheaper or premium alternative models appear. The original business case may still be sitting in a deck while the economics underneath it have already changed.

IBM explicitly warns that quarterly reviews may be too slow. By the time cost overruns show up in financial reporting, the underlying drivers may have been embedded in workflows for months.

That means AI economics cannot be treated as a static approval exercise. They have to be monitored as an operating condition.

The most powerful model is not always the best economic choice

IBM makes another point that should get more attention: not every use case requires the newest or most powerful model.

Organizations using multiple models in a more integrated, fit-for-purpose way expect materially better outcomes than those relying predominantly on large pre-trained models. IBM reports expectations by 2030 of:

• 24% greater productivity gains
• 55% greater improvement in operating profit margins
• Roughly twice the reduction in process cycle and project delivery times

The management question is simple:

Are we paying for more intelligence than the task requires?

A premium model may be justified for a complex, high-consequence workflow. It may be wasteful for a high-volume, low-complexity task that a smaller model can handle just as well. But you can only make that choice if you can see both what the model produces and what it costs to produce it. Otherwise, model selection becomes a technical preference instead of an economic decision.

Cost visibility should change what gets funded

This is where cost visibility becomes strategically useful. IBM recommends treating AI as a living portfolio: continuously analyzing spending, ownership, performance, and expected returns at the use-case level so leaders can adjust capital as costs and returns change.

That changes the operating loop:

measure cost → compare outcome → evaluate alternatives → reallocate → measure again

• If a lower-cost model produces the same business outcome, move the workflow.
• If costs are rising faster than value, intervene.
• If an initiative is improving, consider scaling.
• If economics deteriorate, constrain or redesign it.

The objective is not simply to spend less. It is to make better allocation decisions.

The hidden risk is stale economics

This is where weak visibility becomes expensive. A company may keep scaling an initiative because it worked well at launch. But maybe its unit economics are deteriorating, usage volume increased faster than expected, indirect costs are rising, or the model is now overpowered because a cheaper alternative exists.

Maybe the value is still real—but the cost of producing that value is no longer competitive. Without current visibility, the organization keeps acting on yesterday’s economics. And AI moves too quickly for that to be safe.

Better visibility improves more than budgeting

IBM argues that stronger financial discipline is associated with better AI execution. Organizations with tighter visibility and coordination are more likely to exceed cloud expectations, deploy more AI agents per budget dollar, and report greater readiness for future AI adoption. IBM emphasizes early visibility, tighter coordination, and more deliberate capital allocation rather than simply higher spending.

That is the deeper point. Cost visibility helps leaders answer:

• Which model should we use?
• Which initiative deserves more capital?
• Which workflow is becoming too expensive?
• Where are we paying premium prices without premium value?
• What should be constrained?
• What should be scaled?
• Where should the next AI dollar go?

Those are not accounting questions. They are operating decisions.

AI value without economic visibility is incomplete

IBM’s 85% figure matters because it exposes a basic weakness in how many organizations are managing AI. They may know that AI is producing something useful, but they often cannot see clearly enough whether the economics are improving or deteriorating.

That leads to an important distinction:

• AI value tells you what the system produces.
• AI economics tells you whether producing it still makes sense.

Companies need both. Because the most important cost question is not: What did this AI initiative cost us?

It is:

Knowing what we know now, would we fund it the same way again?

That is where real-time visibility becomes valuable.