The latest findings from the fifth EY US AI Pulse Survey discover that token costs are changing where companies deploy artificial intelligence, who receives access and how usage enters the budget. The survey results, however, do not point to a broad retreat from AI investment.
Among senior leaders at organizations investing in AI, over 80% said their organizations were concerned about token use and related costs. Only around 65% said their organizations actively monitored consumption and had clear spending guardrails.
Token costs prompt strategic shifts
Among respondents whose organizations used token-based tools, 98% said token use and related costs had prompted their organizations to reconsider some part of their approach. 37% said their organizations were reconsidering where to prioritize AI investment across business functions, 35% were doing so across use cases and 32% were reconsidering which employees should receive access. Another 30% were reconsidering where to host AI workloads, including whether to use on-premises systems or the cloud.
37% said token costs had led their organizations to consider expanding the scope of their rollout, compared with 15% considering a reduction. 29% were accelerating deployment, while 15% were slowing it.
“‘AI saves time’ is no longer a sufficient business case when the costs are mounting and difficult to ascertain over the long run,” EY Global AI Consulting Leader Dan Diasio said Taken together, the EY findings indicate reassessment across business functions, use cases and employee access, but not a broad move toward narrower or slower rollouts. The survey does not establish whether those changes subsequently reduced costs or improved returns.
The budgeting challenge of token economics
Direct model-provider API usage differs from conventional seat licensing because spending varies with the tokens consumed by each request. A FinOps Foundation working group said this economic model “defies easy budgeting with traditional tools,” citing developer-led purchasing, opaque billing, limited native allocation and pricing that varies by model tier and use case.
The group recommends inventorying provider accounts and API keys, adding usage tags and measuring costs per query, user or workflow. Those measures are intended to connect consumption to a business activity rather than leave spending as a single provider-level total. The guidance describes a practitioner approach, not evidence that the controls have been widely implemented or have delivered savings.
Separate survey data indicates that AI spending is moving into the remit of established cost-management teams. The State of FinOps 2026 report said 98% of respondents now manage AI expenditure, up from 63% in 2025. Its AI section covered 693 practitioners, while the wider global survey received 1,192 responses. Participants most frequently reported difficulty seeing AI costs, allocating them to business units and determining value.
A mixed approach to building and buying software
The EY survey also records a mixed build-and-buy response, although the published results do not establish that token costs caused it. The share of senior leaders saying their organizations were focusing AI investment on off-the-shelf solutions rose from 56% in 2025 to 66%. At the same time, 76% said off-the-shelf software no longer met their organizations’ specific needs, while 87% had deployed or were piloting programs to develop internal software with AI. 82% expected per-seat software pricing to become less relevant in their industries over five years.
Workflow and productivity tools for specific teams were the most commonly reported category. Among organizations running or piloting internal-development programs, 60% were building those tools, while 39% were augmenting and 33% were replacing existing enterprise software.
Internal development introduced additional management questions. 72% reported challenges slowing their progress. Unmanaged employee-created applications were cited by 34%, followed by regulatory and governance concerns at 33% and cybersecurity exposure at 32%.
Spending falls short of early expectations
Reported spending also remained below earlier expectations. In the April 2025 survey, 35% expected their organizations to be spending at least $10 million on AI a year later. 23% reported that level in the April–May 2026 wave. EY kept the tracking questions and audience definition unchanged, but its methodology does not say it followed the same respondents or organizations over time.
EY also found that 98% reported some positive return from AI. Its published results did not disclose average returns, token expenditure, costs by workload or the share of AI budgets represented by token charges. The available evidence therefore supports a shift toward tighter usage control and more selective allocation, but not a conclusion that companies have established the full cost or financial return of their deployments.
The report was based on an online survey of 534 US-employed decision-makers at senior vice president level or above, conducted by a third-party vendor from April 24 to May 17, 2026. It covered 10 industry groups. EY reported a margin of error of plus or minus four percentage points for the total sample at the 95% confidence level but did not publish the subgroup sizes for the token-related questions.