Why Companies Struggle to Price AI Services Amid Unpredictable Token Costs

Enterprises integrating AI agents built on large language models such as ChatGPT, Claude or Gemini encounter significant challenges when trying to price their services, because token usage is unpredictable. A token represents a fragment of a prompt or response, and slight variations can change the number consumed, while combining multiple agents multiplies the effect. Although the cost per token has dropped, analysts at Goldman Sachs expect monthly token consumption to rise twenty-fourfold between 2026 and 2030.

Companies like Microsoft and Uber have already hit token limits unexpectedly, prompting internal cost controls. Smaller vendors are using flat-fee personal accounts to stay under the radar, but industry insiders warn this will not last as big platforms pressure for profitability. Executives are debating pricing strategies ranging from across-the-board increases to usage-based bundles, yet any shift in the underlying model providers' rates could upend these plans.