Only 11 percent of roughly 400 surveyed businesses report being able to accurately forecast their artificial intelligence spending, according to survey and research data reported by the Financial Times on October 5, 2026. The findings highlight persistent operational challenges for enterprise technology departments attempting to predict and control generative AI usage costs.
Token Consumption Drives Budget Variance
Artificial intelligence usage is primarily billed on a per-token basis, measuring the structural units of text or data processed during queries and output generation. However, because a model's token consumption for any given task can be highly unpredictable, organizations struggle to establish dependable baseline projections for long-term expenditure.
Cheaper Models Can Increase Total Expense
The unpredictability of token usage directly challenges the assumption that lower-priced AI models always save money. A study analyzing more than 6,800 tasks found that lower-priced models actually ended up costing more than higher-priced alternatives across 32 percent of evaluated workloads.
When a less expensive model requires significantly more tokens or longer iterations to complete a task, the total computational volume inflates. Consequently, lower nominal rates per token do not guarantee lower final operational costs for nearly one-third of benchmarked enterprise tasks.
Widespread Enterprise Forecasting Challenges
The survey data indicates that the vast majority of organizations remain unable to reconcile token-based AI consumption with traditional fixed software procurement models. With just over one in ten enterprises reporting confidence in their financial forecasts for AI deployments, variable token efficiency across workloads remains a primary source of budget variance.
AI-generated editorial illustration; not a documentary photograph.
SIGNAL TITANS
Leave a Reply