According to a report EY released this week, in a survey of 500 U.S. decision-makers at the senior vice president (SVP) level or above, more than four-fifths of companies investing in AI reported internal concerns about token usage and related costs. Yet despite these concerns, 37% of companies were actually expanding the scope of their AI adoption, while only 15% narrowed it.
Dan Diasio, EY Global AI Consulting Leader, noted in the report that "companies are shifting from pushing adoption forward indiscriminately to setting priorities," adding that "those priorities should focus on doing different things, not doing the same things differently."
Why Cost Concerns and Expanded Investment Are Happening Simultaneously
The same report found that three-quarters of senior leaders with ongoing AI investments said off-the-shelf software solutions do not meet their IT needs. Nearly nine out of ten companies said they have either fully built internal AI or are running pilots. This suggests that cost concerns are not a signal to halt adoption altogether, but rather a signal to shift from off-the-shelf solutions toward in-house development.
Vendors have also picked up on this cost sensitivity and responded accordingly. OpenAI announced earlier this week that it had lowered costs for its Luna and Terra models, while Oracle and AWS unveiled product changes last month designed to help leaders manage costs more effectively. Diasio said, "Since the target keeps shifting every six weeks with this technology, it's understandable that executives haven't yet established the right roadmap or narrative for change." However, he added, "organizations need to deliberately choose which initiatives to pursue and drive them forward based on financial value. Otherwise, they'll just be running in place."
Another Challenge: Lack of Visibility
Alongside cost, visibility issues are also coming to the fore. According to Flexera data released in June, more than two-thirds of companies lack accurate visibility into their AI software usage, and about three-fifths reported that AI overspending had increased year-over-year. FinOps practices, originally aimed at cloud costs, are now expanding their scope to include AI cost management. The Tokenomics Foundation, launched last month as a spin-off of the Linux Foundation, aims to bring enterprise-wide AI spending under control.
What's Different for Korean Companies
Because this survey targeted U.S. decision-makers at the SVP level and above, it's difficult to apply the findings directly to the organizational structures or decision-making systems of Korean companies. Still, the pattern of "cost concerns and expanded investment happening in parallel" suggests that rather than framing adoption as a binary choice, the right order of operations is to first understand exactly which tasks are consuming how many tokens. If a company is weighing off-the-shelf solutions against in-house development, that decision only has a solid basis once visibility into token usage and costs has been established.
For organizations considering AI adoption—or already in the process—the priority this week should be checking just how accurately they can track AI-related software and compute usage by department.