The rapid expansion of Large Language Models (LLMs) and agentic AI has created a financial headache for businesses that are discovering just how expensive these tools can be.
While tech giants like Microsoft and Google have poured billions into these models, the actual cost of operation is tied to 'tokens'—the mathematical chunks processed by AI—which are notoriously difficult to predict.
Because AI outputs are non-deterministic, the same request can result in vastly different token usage, leaving companies with little control over their monthly expenditures. The scale of this problem is massive, with Goldman Sachs forecasting that token consumption will skyrocket to 120 quadrillion tokens per month by 2030.
Major firms are already feeling the heat; Uber reportedly exhausted an entire year's worth of AI coding budget in just a few months, and Microsoft has had to rein in its own internal use of third-party tools.
As companies scramble to find a sustainable pricing model, many are finding that traditional budgeting is impossible when the underlying costs are constantly shifting.
Whether it is through flat-fee workarounds that big vendors are likely to shut down or the implementation of stricter prompt guidelines, the reality is that businesses are currently flying blind.
Until these platforms face shareholder pressure to prove profitability and stabilize their pricing, the era of unpredictable, ballooning AI costs will continue to squeeze corporate bottom lines.
Tags
