The Unforeseen Costs of AI Innovation
The widespread adoption of free AI tools like ChatGPT, Claude, and Gemini has provided millions of users with access to sophisticated technology developed through colossal investments by tech giants such as Microsoft, Google, and Anthropic. While these free versions offer immense utility, the companies behind them are naturally seeking to monetize their substantial research and development expenditures. They achieve this by offering premium, paid-for versions equipped with advanced functionalities, including coding assistance and billing integration. Concurrently, a growing ecosystem of third-party firms is developing and marketing specialized services powered by AI agents, often built upon existing large language models (LLMs). However, establishing a consistent and predictable pricing model for these advanced AI services has proven remarkably difficult.
Simon Gooch, representing identity management firm Saviynt, which is integrating agentic AI into its offerings, highlights the inherent challenge. "Trying to tie someone into a cost model for the next 12 months, two years, three years, it doesn't make any sense, honestly, because we don't know," he states. This uncertainty stems primarily from the volatile economics surrounding 'tokens,' the fundamental units that underpin LLMs and agentic AI systems.
Understanding Token Consumption and Its Variability
When a user interacts with an LLM by submitting a query or requesting a task, the input is broken down into mathematical segments known as tokens. The AI model then processes these tokens to generate a response, which is subsequently converted back into human-readable text, code, or commands. The core difficulty lies in the unpredictable nature of this process. Minor alterations in a prompt can lead to vastly different outputs, and even identical prompts may not always yield the same response. Furthermore, different AI models will produce varying results, contributing to the overall variability.
In the realm of agentic AI systems, this complexity is magnified. Businesses frequently deploy multiple AI agents in concert to facilitate decision-making and execute actions. This collaborative approach significantly increases both token utilization and the inherent unpredictability of costs. Despite a reported decline in the cost of individual tokens and their associated credits in recent years, a Goldman Sachs analysis indicates a dramatic surge in overall token consumption. The bank projects a staggering 24-fold increase in token consumption between 2026 and 2030, reaching 120 quadrillion tokens monthly, driven by the expanding use of AI agents across industries.
The Challenge of Managing Unseen Costs
Many organizations and individual users of AI systems often struggle to accurately gauge their token expenditure until they either deplete their allocated resources or receive an unexpectedly high monthly bill. This issue has even reportedly affected large corporations; Microsoft, for instance, has reportedly curtailed its engineers' use of certain third-party coding tools, and Uber reportedly exhausted its annual AI coding token budget within a few months earlier this year.
Will Venters, an Associate Professor of Digital Innovation and Information Systems at the London School of Economics, observes that companies can be caught off guard as they experiment with or implement AI internally, with staff inadvertently consuming a large volume of tokens. "People are finding it really hard to manage that cost… it's a non-deterministic output, so it's a non-deterministic value," he explains.
Strategies for Cost Mitigation and Future Pricing Models
Despite these challenges, companies are exploring various strategies to manage AI-related expenses. Oliver King-Smith, founder of smartR AI, notes that smaller organizations sometimes utilize flat-fee personal accounts, a practice he believes major AI vendors will eventually curtail once shareholder pressure for profitability intensifies. He advises companies to be more discerning in their choice of AI models. Rob Steele, CFO of iplicit, a UK accounting software firm, emphasizes the importance of precise prompting, likening it to providing detailed instructions for a shopping trip to ensure desired outcomes and avoid unnecessary expenditure.
The cost management problem becomes particularly acute when AI is embedded into products intended for thousands of users, potentially leading to ballooning expenses. Managers may discover that tokens are required not only for core software development but also for crucial ancillary tasks like testing, security, and implementing protective guardrails. Venters points out that deploying additional AI agents can be done with a single click, contrasting sharply with the careful consideration and planning involved in expanding a human workforce.
However, Venters also offers a nuanced perspective, suggesting that despite the unpredictable token costs, the value derived from AI usage might ultimately outweigh the expense. "It's not quite the same as a calculator," he says. "The more you give it, the more expensive it is, but the better the result may be."
The ultimate challenge remains how to pass these variable and often unpredictable costs on to end customers. Bill Peterson, Senior Director of Product Marketing at Sumo Logic, acknowledges that "Nobody's really figured it out." His software firm, currently previewing new security services based on agentic AI, is actively discussing pricing strategies with corporate clients. Options being considered include across-the-board price increases, performance-based billing, or charging for bundles of incidents. Peterson highlights the inherent instability of such models, noting that they could be disrupted if major LLM providers alter their own pricing structures. "You get into variable pricing, and it's changing every couple of months," he states, adding that "Customers don't like that. That's not how anybody builds a budget." The search for a stable and equitable pricing model for AI services continues.