Despite the remarkable free offerings from giants like Microsoft, Google and Anthropic, businesses are struggling to price the AI services that power them.
Large language models consume “tokens” – small units of text or code that the model processes. Users pay for both the input and output tokens, and the total cost varies unpredictably: wording, model version, or even the moment you run the query can change the token count.
Goldman Sachs notes that while the cost per token has fallen, the volume of tokens a company or individual uses has jumped 24‑fold between 2026 and 2030 – projected at some 120 quadrillion tokens a month. That volatility makes setting a consistent price model a moving target.
Microsoft tightened its engineers' use of third‑party coding tools, while Uber reportedly blew through a yearly token budget in just a few months.
Simon Gooch, an identity‑management entrepreneur, warns that “trying to tie someone into a 12‑year cost model is a useless exercise when token usage is that unpredictable.”
Smaller firms like smartR.ai’s founder Oliver King‑Smith say they can skirt prices by using flat‑fee personal accounts, but that strategy may soon be unsustainable as big vendors clamp down.
Enterprise leaders are developing practical tactics: precision prompts that reduce token waste, a careful selection of models, and integrated cost dashboards to monitor usage in real time. Nonetheless, the number of tokens used by each employee—whether to write code, run tests, or solve negotiations—remains opaque until the bill appears.
Bill Peterson of Sumo Logic highlights the challenge of passing AI costs on to customers. “We’re still debating whether to increase base prices, charge by results, or bundle incidents,” he notes. “But if provider prices shift again, our models could become unstable.”
As the AI economy matures, experts agree that transparent usage data, clearer pricing frameworks, and industry‑wide standards will be essential to keep the technology affordable and accountable.
