Dive Brief:
- CIOs are struggling to predict token consumption and clearly link spend to business outcomes as rising AI costs become a primary concern for executives, a report from IT services company Accenture found. The report is based on a July 2026 survey of 750 senior executives from enterprises and 15 interviews with Fortune 500 technology and finance leaders.
- Enterprises spent roughly $2.5 billion on tokens in 2025 and token consumption is expected to grow 78% over the next 24 months, the report found. Token consumption is the third-largest driver of AI costs behind infrastructure and software development and maintenance.
- While AI is generating productivity gains and improving customer outcomes, drawing the line between returns and token spend is tricky. “Only one dollar in five of enterprise token spend shows up as a quantified financial outcome,” the report said. “The remaining four dollars sit in a space companies believe is productive but cannot prove.”
Dive Insight:
Despite concerns about rising costs and the challenges of connecting AI spend to business outcomes, enterprises are pressing ahead with adoption efforts.
More than 4 in 5 businesses reported concerns about AI token usage and costs of AI implementation, yet 37% of businesses still plan to expand the scope of their AI deployments, according to an EY report released in July.
Additionally, while tokens are starting to become more cost efficient, falling prices will not always translate to reduced expenses. When asked what they would do if token prices fell 25% or more, 42% of surveyed executives said they would expand existing AI workloads, according to the Accenture report.
Earlier this year, the Linux Foundation launched the Tokenomics Foundation to address the issue of token consumption leading to higher AI costs. As managing AI spend becomes a critical focus area, CIOs can take steps to help tie token consumption to outcomes, the report said.
CIOs will need to improve visibility into token consumption at the workload level by deploying an AI gateway or observability layer for each AI interaction, including the model used, token cost and output. This strategy allows 53% of token usage to be tracked to a team or individual user, according to the report.
After improving visibility, making teams such as engineering responsible for their token costs will encourage managers to justify spend for each use, choosing the right AI models for the job while keeping expenses in mind, the report said. CIOs should also make model costs transparent to individuals or teams, as most developers and employees tend to choose the most capable models because they can’t see cost differences, the report said.
Sending workloads to cheaper models is critical in managing token costs as only about 10% of workloads need advanced — and more costly — frontier-level reasoning, the report said.
Vendors have started adding cost-control features that allow queries to be routed to appropriate AI models in response to enterprise worries. Snowflake launched a model routing feature within its Cortex AI Gateway and other AI products that automatically chooses the best AI model for the task based on quality and cost. Meanwhile, Google began offering flexible billing options that allow enterprises to cap monthly AI spend.
Lastly, before committing to new AI workloads, CIOs should define what the process costs currently in time or money, what financial outcome will be improved by AI and “how that outcome will be measured in dollars,” Accenture’s report said.
