AI costs are falling, but enterprise AI spending keeps rising. Organizations must treat AI as a capital allocation decision and control the AI operating layer to maximize value.

By Neil Dhar, Global Managing Partner, IBM Consulting

The executive brief

• Falling AI costs are not translating into lower AI spending, making AI economics a growing leadership challenge.
• Organizations generate the most value when AI investments are treated as capital allocation decisions tied to specific business outcomes and measurable ROI.
• Managing AI consumption requires focusing on economic returns, not just usage or budget controls.
• Control of the AI operating layer gives organizations the ability to optimize cost, performance, governance and value as AI technologies evolve.

The cost of using AI keeps falling. Yet, for most organizations, AI spending continues to rise.

This conundrum is not new. As AI gets cheaper and more capable, companies are finding more ways to create value. And whether you were an early adopter or are racing to catch up, chances are your AI bill is a lot higher today than it was six months ago.

This surge in spending has become an omnipresent topic in board and C-level discussions across industries. The technology is powerful, but the bills are growing, and the question of who controls the economics of AI becomes much harder to ignore. 

Treat AI investments as a capital allocation decision

What many business leaders underestimate is just how much control they have over this. 

My background is in finance and private equity, which means I look at problems through the lens of ROI and value creation. AI is no different. Like every other investment, it’s a capital allocation decision, and it needs to be treated like one.

"AI is a capital allocation decision, and it needs to be treated like one."

Yet what I often see is companies approaching AI primarily as a technology deployment. They’re running experiments and science projects to find out what AI can do. That’s an important first step, but it also means the cost base keeps growing before anyone has established how value will be measured.  

Every dollar invested in AI should have a purpose. What benefit is that capital expected to generate? Is it driving revenue? Increasing profit? Enhancing cash flow? If you can’t answer those questions, there’s a high chance you’re wasting money.

Start with the business process and expect results

The best companies are clear about what the investment is going to deliver. They align AI investments to specific business processes, whether that’s in sales, marketing, HR, finance, procurement or R&D. Then they take those processes apart, looking for bottlenecks and points of friction where the technology can reduce costs, speed up work or deliver a better experience in ways that show up in revenue or profit.

Where I often see companies go wrong is treating AI as a long-term bet that won’t be measured for a year or more. The more disciplined companies run on a much shorter clock, tracking whether AI is delivering the improvements they targeted and expecting tangible results within three to six months.

Don't put a cap on innovation

A clear business case brings discipline to AI spending, but you also need control over how those costs grow. 

Many frontier-model providers charge based on usage, which effectively puts customers on a meter the provider controls. The more a company uses, the more it pays, at a price set by someone else.

"Capping usage may contain the bill, but it can also limit innovation."

I recently spoke with a client that had adopted virtually every major model and AI tool available.

Their philosophy was that if you give employees access to everything, innovation will follow. Six months later, they found themselves facing a different reality. Costs had ballooned, operational control was limited, and the connection between spending and business value remained unclear.

Capping usage may contain the bill, but it can also limit innovation. A more effective approach is to shift focus from budgets towards economics, looking past the bill to what each dollar delivers. For example, rather than tracking token spend alone, consider token yield: the value of successful outcomes relative to the total cost of tokens used. 

Why organizations need control of the AI operating layer

What companies ultimately need is control over the AI operating layer, the environment that governs how AI is used across the business. Without it, it’s hard to turn investment discipline into day-to-day decisions about cost, consumption and value.

For a long time, the assumption was that the best results came from using the most powerful models available. That’s rarely true. You don’t need a brain surgeon for a paper cut, and most AI tasks don’t require the equivalent of one either. 

Owning that layer means you set the rules. You decide which models your people, applications, and AI agents can use, what they can do with them, and under what conditions. You keep a clear record of what’s happening and what it costs, rather than paying a meter someone else controls.

"You keep a clear record of what’s happening and what it costs, rather than paying a meter someone else controls."

This also gives you more choice in where and how AI runs. Highly capable open-weight models can now be run on your own infrastructure, giving organizations greater flexibility over both cost and performance. Using this approach, IBM recently demonstrated the ability to run 3,000 AI agents simultaneously.

Taking back control

The pressure on AI spending isn’t going away. Costs, models and capabilities are changing too quickly for any company to set a strategy once and walk away. Whether cheaper, more capable AI improves the economics of the business depends on where leaders invest, how work gets done, and what earns more funding.

Owning the operating layer gives leaders the control to act on those decisions and adapt as better options emerge. Companies that take control of their AI economics will be better positioned to see AI show up in revenue and profit. Those that simply pay the bill will find someone else is making those decisions for them.