AI tokens are catching some companies by surprise — and spurring big bills

Uber has restricted employee use of AI tools like Anthropic’s Claude after depleting its 2026 AI coding budget in just four months, highlighting a growing issue for businesses as AI token costs escalate faster than anticipated. AI tokens, used in models like ChatGPT and Claude, function as a metered billing system for businesses, charging per usage beyond allocated limits. Patrick Farrar, CEO of AI for Canadians and head of AI at Toronto Metropolitan University’s DMZ, notes companies often overlook workflow adjustments and employee training, leading to unplanned expenses. A simple prompt may use fractions of a token, while complex requests or autonomous AI agents can consume hundreds of tokens, multiplying costs significantly. Canadian businesses are adopting AI to drive innovation but struggle with budgeting for token usage. Annie Veillet, a partner at PwC Canada, emphasizes that organizations underestimated costs tied to employee access and AI model selection. While Ottawa’s national AI strategy aims to boost adoption through business integration and workforce education, companies now face financial surprises as AI tools become embedded in daily operations. Experts stress the need for better cost management strategies to align with rising AI dependency. Token pricing varies, with basic prompts costing fractions of a cent per word, while advanced tasks like visualizations or automated workflows can escalate expenses dramatically. Farrar warns that autonomous AI agents—tools operating independently—can consume 50 to hundreds of times more tokens than manual interactions, straining budgets. This trend underscores the gap between pilot-phase AI adoption and large-scale implementation, where unchecked usage leads to unforeseen financial burdens. As businesses scale AI integration, cost control remains a critical challenge. Companies must reassess token allocation, employee training, and workflow design to mitigate unexpected expenses. The shift from experimental AI use to operational reliance exposes financial risks, particularly for firms unprepared for token consumption patterns. Industry observers urge proactive budgeting and monitoring to sustain AI-driven growth without cost overruns.