Businesses are facing escalating costs associated with artificial intelligence, threatening to undermine anticipated productivity gains linked to the technology’s adoption, according to a recent analysis by consulting firm Accenture. This surge in expenses is largely attributed to the transition from basic chatbots to more advanced autonomous AI agents, which consume significantly more computational resources and, consequently, generate substantially higher bills.

The Albanese government has projected that AI could contribute up to A$600 billion annually to the national economy by 2030, highlighting the technology’s economic potential. However, Accenture’s findings suggest that without proper budget oversight and usage controls, organisations risk quickly exhausting their AI budgets, often within months of deployment.

One key factor driving the rising costs is the token-based pricing model used by AI providers. Under this system, users pay for every "token"—roughly equivalent to three-quarters of a word—that the AI processes. Autonomous agents, which operate through multi-step, iterative workflows and complex information retrieval loops, can consume tokens at rates far exceeding those of human users, sometimes using six times as many tokens for a given task.

Examples cited include Uber, which reportedly spent its entire AI allocation within four months on services offered by Anthropic, the creator of the Claude AI model. Microsoft also scaled back on internal Claude licenses, redirecting staff to use the more cost-effective GitHub Copilot. This highlights a broader industry challenge where employees often default to premium AI models for tasks that require less computational power, driving up expenses unnecessarily.

Stacy Pence, Accenture’s AI and Data lead, warned that the rapid adoption of agentic workflows—autonomous AI processes operating without human intervention—is the fastest-growing category of token consumption. These workflows currently account for 11 percent of enterprise token spending but are expected to increase sharply, outpacing company budget growth and rendering static annual IT allocations inadequate.

Moreover, many companies lack sufficient governance mechanisms to monitor and control AI usage effectively. Without clear policies or cost management strategies, autonomous AI systems risk rapidly exhausting budgets through continuous execution cycles. Pence emphasized that organisations must implement governance frameworks to identify when to deactivate such systems to prevent budget overruns.

Accenture’s research also indicates that over half of AI requests within enterprises are routed to expensive, frontier-class models, despite evidence that 90 percent of workplace tasks do not require such advanced processing capabilities. This mismatch between model selection and task complexity further accelerates cost escalation.

In response to these challenges, businesses are reportedly slowing the rollout of advanced AI capabilities to establish better controls and ensure sustainable spending on AI initiatives. The findings underscore the importance of strategic management and oversight as companies navigate the evolving landscape of artificial intelligence investment.