Companies may be enthusiastic about integrating AI into their operations, but the payment stage is proving far from simple. A large-scale survey of 2,145 senior executives across 20 countries found that 29% of business leaders are struggling to understand and control operating costs as they begin scaling AI adoption.

This “budget shock” comes from a simple reality: the way major technology companies charge for AI has changed.

As Anthropic and OpenAI move toward usage-based billing, the true cost of AI is becoming more visible.

When AI Is No Longer Sold as a Flat-Rate Package

Recently, major names such as Anthropic, OpenAI and GitHub have begun shifting some of their services from flat-rate subscriptions to usage-based billing.

This change is similar to moving from unlimited internet access to paying for every unit of data used. It makes budget forecasting far more complicated. Many organizations admit they still lack the ability to forecast, monitor and manage AI spending effectively as this new pricing model becomes more common. Even with the rise of AI Agents, around one-third of senior executives view limited understanding of AI costs and economics as a major barrier.

Because costs are rising and pricing structures keep changing, many companies have been forced to recalculate their plans. Nearly half of the surveyed organizations said they had delayed or slowed AI deployment after realizing that costs were exceeding the actual value generated. Instead, they are shifting toward cheaper but high-accuracy AI models, a segment that is increasingly influencing corporate AI strategies.

However, this does not mean businesses have lost faith in AI. In reality, they are becoming more practical: carefully assessing where AI creates real value and directing capital toward areas with the clearest returns.

The Spending Race Among Tech Giants

While businesses are struggling to balance their AI budgets, major cloud infrastructure providers continue pouring massive amounts of money into the sector. Amazon is expected to spend around $200 billion in capital expenditures this year, up 50% from last year, mainly to upgrade AI-processing capacity for AWS data centers. Microsoft is also moving aggressively, with expected spending of up to $190 billion, a 61% increase.

To stimulate demand and help customers spend more effectively on their infrastructure, both giants are investing heavily in direct engineering-support teams. Amazon has announced a $1 billion investment in AWS Forward Deployed Engineering to help customers adopt AI Agents more easily and shorten deployment time. Microsoft has committed $2.5 billion to a new operating entity called Microsoft Frontier Company, designed to help customers optimize AI capabilities while shaping their own competitive value.

Governance Challenges and AI Backfiring

Beyond money, the next major barrier for AI remains governance: who is responsible for wrong decisions or hallucinated information generated by AI?

Experts say management accountability is crucial, but the success or failure of AI governance ultimately depends on real day-to-day operational processes.

Notably, even major research and consulting organizations sometimes stumble because of AI. Recently, one analysis group reviewed an AI report from a major consulting firm and discovered a surprising fact: only 5 out of 45 cited sources in the report were accurate. The remaining sources were either incorrect, fabricated, or too vague to verify. The flawed report was later quietly removed from several websites, along with a promise to review human-oversight procedures when using AI.

Clearly, AI is smart and convenient, but if companies allow the technology to run freely without strict controls over both spending and information accuracy, the price may come in the form of huge bills and costly brand-damaging lessons.