Why Nearly Half of Companies Are Pausing AI Projects Over Cost

More companies than ever are putting AI into daily work, and almost half have paused or scaled back AI projects because the cost was not worth it. Both findings come from the same KPMG survey, published this week, and together they describe the stage enterprise AI reached in mid-2026: adoption is rising, budgets are under scrutiny, and the companies that can see what AI actually costs are the ones reporting returns.

What did KPMG's Global AI Pulse find?

KPMG's Global AI Pulse for the second quarter of 2026 found that 22% of organizations have reached the "driving adoption" stage, embedding AI in everyday workflows, up nine percentage points from the previous quarter, IT Brief UK reported on Friday. At the same time, almost half of businesses had paused or trimmed AI initiatives after concluding the cost outweighed the value.

The most useful number for executives is the third one. Organizations with full visibility into what their AI costs to run reported established returns at five times the rate of those without it.

IT Brief's report did not include the survey's sample size or the countries covered, so the figures are best read as direction rather than precise measurement. The direction is consistent with what enterprise practitioners have been saying for a year.

Why are companies pausing AI projects?

The pauses are not a retreat from AI. They are a correction to a period when usage grew faster than anyone's ability to price it.

Uber offered a public example in May. The company used up its entire 2026 budget for AI coding tools in four months after encouraging employees to adopt them, including through an internal leaderboard that ranked teams by usage, Fortune reported. Andrew Macdonald, Uber's president and chief operating officer, said it was hard to tie the rising use to features customers would notice. "That link is not there yet," he said on a podcast.

That is the pattern KPMG's numbers describe. Usage-based pricing makes AI cheap to start and hard to forecast. Agentic tools make it harder. Fortune's report cited Gartner's view that even as inference prices fall, agentic models use far more tokens per task, so cheaper tokens do not necessarily mean cheaper enterprise AI. A rollout that looks inexpensive per user can become a large line item once thousands of employees and agents are running.

Why does cost visibility predict AI ROI?

KPMG's five-to-one finding does not prove that tracking costs causes returns. A company that can see its AI spending is probably also better at defining use cases, measuring outcomes and shutting down projects that are not working. Cost visibility is a sign of operational maturity as much as a cause of success.

But that is the point. The organizations reporting returns treat AI like any other business investment, with a budget owner, a unit cost and a measure of value. Speakers at AI Connect events have pressed this test for years. "If the use cases are not clear and it cannot be tied to a measurable business outcome, it's probably hype," Girish Bhat of Revefi said during a AI Connect webinar on choosing AI use cases that pay off. In the same session, Dave Mariani, CTO and co-founder of AtScale, offered a simpler filter: "If AI wasn't in the name of the project, would I do it?"

At COLLIDE 2025, data leaders from The Home Depot, Equifax and Mars Wrigley described how they tie every AI use case to a business OKR, trading off speed, accuracy and cost before a project starts rather than after the bill arrives.

Are falling model prices the answer?

This week brought plenty of price competition. OpenAI released GPT-5.6 in three tiers, with its cheapest, Luna, listed at $1 per million input tokens and $6 per million output tokens, against $5 and $30 for the flagship Sol, Nerova reported. Meta opened a paid API for Muse Spark 1.1 at $1.25 and $4.25, with chief executive Mark Zuckerberg saying the company aims to deliver capable models at very low cost, Reuters reported.

Cheaper tokens help, but they do not fix the problem KPMG identified. A company that cannot see what it spends will not know whether a cheaper tier is saving money or simply encouraging more use. Lower prices can also make it easier to launch projects that never had a business case.

And not every cost is falling. Infrastructure is getting more expensive in places. Contract prices for DRAM memory rose about 44% from the prior quarter and NAND flash about 53%, according to Citi Research data cited by The Next Web after Samsung forecast record profits on Tuesday. Enterprises running their own AI hardware or buying GPU capacity will feel that.

What does good AI cost management look like?

The companies doing this well tend to share a few habits, based on what practitioners have described at AI Connect events and what this week's data suggests:

• A unit cost for every use case. Cost per resolved ticket, per document processed or per question answered, not just a monthly invoice.
• A budget owner outside IT. The business unit that benefits should see and defend the spending.
• Spending limits and alerts. Caps per user, per agent or per project, with alerts before budgets run out rather than after.
• Model routing. Sending routine work to cheaper models and reserving premium models for tasks that justify them.
• A shutdown rule. A clear point at which a project that is not meeting its target gets paused, which is what nearly half of KPMG's respondents ended up doing anyway, only later.

AI Connect panelists have also warned that pilots stall for organizational rather than technical reasons, a theme in our coverage of how enterprises move AI pilots to ROI. Cost visibility is part of that organizational work.

What should leaders take from this?

The KPMG survey is a reminder that enterprise AI's next phase will be judged by finance teams as much as by technologists. Adoption is no longer the hard part. Knowing what each use of AI costs, and what it returns, is what separates the companies scaling AI from the ones pausing it.

Frequently asked questions

What did KPMG's Q2 2026 AI Pulse find?

22% of organizations are at the 'driving adoption' stage, almost half paused or scaled back AI projects over cost, and firms with full cost visibility reported ROI five times as often.

Why are companies pausing AI projects?

Usage-based pricing and agentic tools make costs hard to forecast, and many firms cannot link rising spending to business results.

How can enterprises manage AI costs?

Track a unit cost per use case, give budgets to business owners, set spending limits and alerts, route routine work to cheaper models and define shutdown rules.

Filed under: AI Strategy & ROI, People, Culture & Leadership