Tokenomics: Why making AI pay is tricky

## The Tricky Business of Pricing AI: Why "Tokenomics" is a Headache

Major artificial intelligence companies offer premium versions of their technology, but the economics of making AI pay are proving surprisingly complex. While free versions of services like ChatGPT provide immense value, the firms behind them – including Microsoft, Google, and Anthropic – have invested hundreds of billions in developing the underlying Large Language Models (LLMs). These companies naturally seek to recoup their investments by offering paid tiers with advanced features for tasks like coding or billing.

Simultaneously, third-party companies are developing and selling services powered by AI agents, often built upon LLMs and trained for specific functions. However, determining a price for these services presents a significant 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," states Simon Gooch of Saviynt, an identity management company integrating agentic AI into its offerings.

This difficulty stems from the rapidly evolving economics surrounding "tokens," the fundamental units of LLMs and agentic AI. When a user interacts with an LLM like ChatGPT or Claude, their prompt is broken down into mathematical "tokens" for processing. The LLM's response is also generated in tokens, which are then converted back into text, code, or commands.

The core issue is the inherent unpredictability of this process. Minor alterations in a prompt can yield different answers, and even identical prompts may not always produce the same result. Furthermore, different models will generate varying responses. In agentic systems, where multiple AI agents collaborate to make decisions and take actions, both token consumption and unpredictability are further amplified.

While the cost of individual tokens (or the credits used to purchase them) has significantly decreased in recent years, according to Goldman Sachs analysis, the overall volume of tokens consumed by businesses and consumers has dramatically increased. The bank projects a 24-fold surge in external token consumption between 2026 and 2030, reaching 120 quadrillion tokens per month, as companies increasingly adopt AI agents.

Yet, companies and individuals using AI systems often have a limited understanding of their token usage until they either exhaust their allocation or receive their monthly bill. Microsoft, for instance, has reportedly scaled back its engineers' use of certain third-party coding tools, and Uber reportedly depleted its annual AI coding token budget in just a few months earlier this year.

Will Venters, Associate Professor of Digital Innovation and Information Systems at the London School of Economics, notes that companies can be caught off guard as they experiment with or implement AI internally, with staff rapidly consuming 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.

Companies are exploring strategies to address this. Oliver King-Smith, founder of engineering software firm smartR AI, suggests that smaller organizations can "fly under the radar and use [flat fee] personal accounts which I am sure the big vendors don't like." However, he believes this practice is unsustainable, predicting that once major AI platforms face shareholder pressure for profitability, "They will start clamping down." King-Smith also advises companies to be more deliberate in their choice of AI models.

Rob Steele, CFO at UK accounting software firm iplicit, emphasizes the need for greater precision in prompts. "You wouldn't send someone in your family out to get the weekly shop without any kind of detailed instructions as to what you expect in that shopping basket, right?" he asks.

Venters highlights the difficulty in controlling costs when AI is integrated into a product destined for thousands of users. AI expenses can quickly escalate as managers realize tokens are needed not only for core software development but also for tasks like testing, security, or implementing guardrails. "It's particularly hard when you're looking at agentic processes," Venters adds, noting that deploying more AI agents is a simple click, unlike the careful discussions involved in expanding a human workforce.

Despite the unpredictable nature of token costs, Venters points out that companies might ultimately derive greater value from their AI token usage. "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."

However, companies still face the challenge of passing these costs on to their own customers. "Nobody's really figured it out," admits Bill Peterson, senior director of product marketing at Sumo Logic. His software firm is currently previewing new security services based on agentic AI and is engaged in discussions with corporate clients about pricing. "We're still having some fun conversations about this internally," he remarks drily.

Peterson suggests options such as across-the-board price increases, payment by results, or charging for "bundles" of incidents. Yet, any chosen price structure could be disrupted if and when large language model providers alter their own pricing strategies. "You get into variable pricing, and it's changing every couple of months," he says. "Customers don't like that. That's not how anybody builds a budget."