Companies shift from bigger AI spending to cutting AI costs

In short: Companies are realizing that using advanced AI is pricier than expected, so they are shifting from open-ended spending to cost cutting and tighter tracking.

What's going on

Many businesses are finding that training and running AI systems costs more than their plans assumed. One analysis says 80% to 85% of large companies miss their AI cost forecasts by more than 25%. The extra spending often comes from graphics chips called GPUs (special chips used for heavy computing), cloud bills, and the cost of each AI answer.

AI is also becoming a bigger part of overall budgets. Companies expect to spend about 1.7% of revenue on AI in 2026, more than double 2025 levels. Some firms think AI could take 25% to 50% of the entire IT budget within two years.

Another issue is that time savings are not always as large as promised. A global survey cited in 2026 reporting found that for every 10 hours AI saves, workers spend almost 4 hours checking and correcting its output. That is like hiring a fast assistant who still needs a lot of proofreading.

Even so, many executives say AI is paying off. In a 2026 “state of AI” survey, 88% said AI increased annual revenue and 87% said it reduced annual costs.

What to watch

Expect more “AI frugality” tactics. These include reusing repeated text so companies do not pay twice (like saving common phrases), sending simple tasks to cheaper AI tools, and buying AI work in bulk for discounts. CFOs are also pushing for clearer rules and better metrics, such as cost per successful task, instead of paying just to access the most powerful model.

Source: NYTimes