By 2028, about one in five organizations worldwide will scale back or abandon AI in parts of their operations and return to conventional software development, unable to rein in runaway costs, according to research from Gartner.
The retreat is expected to concentrate in areas where the return fails to justify the spend.
The reversal marks a sharp turn from the early land-grab era, when leading developers dangled free or heavily discounted access to win market share.
As the large language model market consolidates around a handful of dominant suppliers, flat monthly subscriptions are giving way to pay-as-you-go pricing, handing pricing power to vendors and leaving customers unable to forecast their final tab.
The strain is already visible at the world's largest firms.
Microsoft in May halted Claude Code licenses for thousands of internal staff and shifted them to its own GitHub Copilot, while Uber Technologies burned through its entire 2026 AI budget in about four months after urging some 5,000 engineers to lean on AI coding tools, later capping monthly spending at about $1,500 per person.
A McKinsey survey of 75 companies in May found that 93 percent had blown past their AI budgets.
Agentic AI, which chains together multiple reasoning steps, drives the steepest overruns.
Such systems can consume up to 30 times more tokens than a standard chatbot on the same task, and a failed output that forces the model to reprocess everything from scratch can push consumption as high as 50 times, with most of the cost accruing not in the first attempt but in the grind of revisions.
Gartner has urged firms to adopt "AI FinOps," a discipline for tracking spending and performance in real time, and warns that the divide between companies that master cost control and those that cannot could widen across entire industries.
AJP Takeaways
• Gartner forecasts that by 2028 about one in five organizations worldwide will scale back or abandon AI in parts of their operations and revert to conventional software development because of uncontrolled costs.
• Uber Technologies exhausted its entire 2026 AI budget in about four months in 2026 and capped engineer spending at about $1,500 a month, while Microsoft dropped Claude Code for its own GitHub Copilot in May 2026 to contain expenses.
• A McKinsey & Company survey of 75 companies conducted in May 2026 found that 93 percent of respondents had exceeded their AI budgets, underscoring how token-based pricing is reshaping enterprise AI economics.
Copyright ⓒ Aju Press All rights reserved.