Corporations are tightening AI token usage limits as costs rise unexpectedly, marking a shift from previous encouragement of AI integration.
In a dramatic turnaround, many large corporations that once encouraged rampant AI usage are now tightening access and enforcing strict limits on token consumption. This shift comes as companies face unexpectedly high expenses associated with tokens, the units of measurement for AI services.
Tokens, which represent the smallest units of data processed by AI models, were once seen as a minor cost. However, following the monetization of AI services, particularly after the release of GPT-4 in March 2023, they have become a financial burden. Pricing for tokens has been standardized, with costs reaching $30 for 1 million input tokens and $60 for output tokens.
Despite a decrease in the average cost of tokens—down approximately 98% over the past three years—corporate AI spending has exploded, rising by about 320%. Annual AI budgets have ballooned from $1.2 million in 2024 to $7 million in 2026, driven by the increasing integration of AI solutions and a surge in token usage.
Notably, agent technologies have significantly amplified token consumption, with costs for complex workflows escalating dramatically. Analysts have identified this phenomenon as the Jevons Paradox, where lower prices lead to increased consumption, a trend that has been particularly pronounced in the AI sector.
Corporate Reactions to Rising Costs
Major companies are now grappling with the financial implications of their AI strategies. Uber, for instance, initially encouraged extensive use of AI tools but recently announced a $1,500 monthly limit per employee on AI services due to excessive spending. Uber's CTO admitted to exceeding the budget during a demonstration, highlighting the unforeseen costs of token use.
Meanwhile, Microsoft has curtailed its use of Anthropic's Claude Code after a costly experiment, directing developers to migrate back to its own GitHub Copilot. The decision underscores concerns regarding the sustainability of AI spending amid rising token costs.
Amazon is also restructuring its partnership with Anthropic to pay for tokens rather than computing hours, a change that could increase its overall costs, although the company has denied this assertion.
Some firms have reported staggering expenses; one company incurred a $500 million bill for AI computations due to a lack of spending limits. Others, like Priceline, have begun imposing restrictions on AI usage following similar experiences.
Looking Ahead
While the so-called "tokenpocalypse" presents challenges, large corporations are likely to absorb these costs more easily than smaller firms. In response to the growing financial pressures, the Linux Foundation has launched the Tokenomics Foundation to promote better cost management practices in AI spending. As the industry evolves, companies will need to adjust their budgets and strategies to account for these new expenses, ensuring that AI remains a valuable asset rather than a financial liability.