WASHINGTON DC. WA, October 5, 2026 — Artificial intelligence — and especially, especially generative AI — is cheaper than originally thought during their budgets with businesses across industries. A recent study has shown that above half of firms overshoot on their AI cost forecasts by double digits. And almost one in four miss those forecasts by over fifty percent.
Budgets Built on Shaky Assumptions
As projects transition from pilot programs to full production scale, corporate AI spending has taken off like a rocket. According to the 2026 Global Forecast, spending on AI technology around the world will be at $2.59 trillion this year as per Gartner. That figure is up forty-seven per cent year on year.
Most of that growth comes from inference costs, which are the ongoing operating costs to run AI (after it has been trained). Inference costs scale with usage, unlike training which is typically a one off cost. When firms use AI for more complex applications, those are dramatically less able to predict accurately.
How the cost estimates go wrong
Finance teams frequently fail to account for the resources required to acquire, store, and protect the data upon which AI systems rely. These indirect costs are seldom included upfront in project bids—a practice that results in cost overruns once systems approach production scale. Smaller companies (less than 10-50M in revenue) seem to have an easier time with their forecasts than larger enterprises.
That shell has been hardened, or at least thus far, and new AI use cases suffer a design/complexity tradeoff that translates into rising corporate AI budgets per industry analysts—Ernst & Young have identified this complexity as an emerging price sensitivity due to the addition of agentic systems. Executives said the shift from testing to production has left many finance departments flat-footed. Such overruns, some have said, were a considerable surprise blow to profits.
The Global Large Company Trend
The problem of forecasting itself is not confined to any industry or geography. A few months later, the chief executive of Commonwealth Bank of Australia forecast that token-based AI costs will increase in ways that are not so simple as tasks become more complex. As adoption of the technology accelerates around the world, he said, companies will likely intensify their scrutiny over AI spending.
According to research from IDC, AI infrastructure spending surges this year alone by double digits. That growth has spurred many chief financial officers to seek more transparency about return-on-investment numbers before signing off on additional AI expansion. While your tenure may be uncertain, almost all companies have kept increasing their AI budgets — never pulled back.
How Companies Are Responding
Even with pressure mounting over costs, most technology leaders say they have no plans to slow down AI adoption. About four in ten leaders said they are advocating for widespread use of AI as long as it creates impact. Others fly over the entire range of use cases, but only invest in those that remain cost-effective as costs fluctuate.
Still, some executives contend the upswing in costs might serve a useful purpose by deterring weak AI projects. The tighter budgets may lead teams to focus on applications with a more apparent impact on the business rather than use cases for experimentation. The finance and technology leaders will all continue to watch closely throughout 2026 to see if AI spending can finally become predictable.
Vendors have launched more granular billing tools to enable real-time tracking of usage by companies. In some cases, finance departments have brought new staff to better monitor AI usage across teams and projects. The change echoes how companies developed cloud cost management capabilities due to the difficulty in forecasting early cloud expenditure.
As in previous years, smaller, more AI-native companies are better at forecasting than larger enterprises. Part of this, analysts say, is due to tighter internal processes designed from the get-go around usage-based pricing models. In the case of larger organizations, AI tools are usually acquired through departmental silos without functional data and cost-tracking.
With more pilot projects advancing into large-scale production, the next year is likely to herald a period of closer examination. Executives that can’t show a clear return on AI spending may face pressure to throttle back their expansion ambitions. For now, the gap between AI ambition and AI budgeting discipline is one of the defining challenges (along with how to conduct an ML development process that leads to high-performance results).
