Agentic AI to Increase Token Use 100-Fold, Forcing Pricing Model Shift

What happened

New analysis from Futurum Research indicates that the adoption of agentic AI systems could dramatically increase the volume of token usage per task, potentially by a factor of up to 100 times. This surge in token consumption is predicted to accelerate a fundamental shift in how AI services are priced, moving away from the current per-token models. The research suggests that the current pricing structures, which often charge based on the number of tokens processed, may become unsustainable or uneconomical as AI agents perform more complex, multi-step operations. The study highlights that agentic AI, characterised by its ability to autonomously plan, execute, and refine tasks, inherently requires more iterative processing and communication with large language models (LLMs). This iterative nature directly translates into a significant uplift in token exchange compared to simpler, single-query interactions. Consequently, the industry is likely to pivot towards alternative pricing mechanisms that better accommodate the operational realities of these advanced AI systems.

Why it matters

For business leaders, AI practitioners, and policymakers, this projected shift in token usage and pricing models carries significant implications for AI governance, ethics, and strategic planning. The increased computational demands of agentic AI could lead to higher operational costs, necessitating a re-evaluation of budget allocations for AI initiatives. Furthermore, the transparency and explainability of these more complex, multi-step AI processes become even more critical when resource consumption escalates, raising new questions about cost attribution and efficiency. From an ethical and governance perspective, the potential for 100-fold increases in token use underscores the need for robust monitoring and control mechanisms. Unchecked, such a rise could lead to unforeseen environmental impacts due to increased energy consumption, as well as exacerbate issues of bias amplification if agents are allowed to autonomously iterate on flawed data or instructions without sufficient oversight. Businesses must therefore proactively develop strategies for optimising agentic AI workflows, not just for cost efficiency but also for responsible resource utilisation and ethical performance.

By the numbers

• 100 times: The potential increase in token usage per task with agentic AI.

The Institute take

The anticipated explosion in token usage driven by agentic AI is more than just a technical detail; it's a bellwether for a fundamental re-evaluation of AI's economic and ethical footprint. While the industry fixates on the immediate cost implications of per-token pricing, the deeper issue is the unconstrained computational appetite of increasingly autonomous systems.

The real challenge isn't merely adjusting pricing models, but instilling a culture of computational frugality and purpose-driven AI design. Responsible AI leaders must move beyond simply optimising for task completion and actively design agentic systems with built-in constraints and efficiency parameters, prioritising not just what an AI can do, but what it *should* do, and with what minimal resources. Otherwise, we risk inadvertently scaling inefficiency and environmental impact alongside capability.