The Uber COO Andrew Macdonald warned that AI spending is hard to justify, echoing concerns that enterprises lack clear ROI. A month later, a firm dumped $500 million on Anthropic’s models after failing to cap usage. These incidents expose a broader trend: companies can’t gauge the cost of token‑driven AI.

Microsoft’s GitHub Copilot shifted users to token‑based billing, letting some burn half their monthly credits in a single prompt. The move highlighted how subscription plans mask true costs, driving overuse. When enterprises adopt token models without visibility, they risk inflating expenses while chasing performance, a pattern mirrored across Anthropic, OpenAI and other providers.

Token pricing forces developers to track usage closely, yet most AI services still hide granular costs. The lack of transparent metrics turns every misstep into a hidden fee, undermining the promised efficiency of LLMs. As firms grapple with rising token bills, the industry must shift toward clear accounting or risk continued waste and eroded trust.

Without a standard ROI framework, decision makers rely on anecdotal benchmarks that vary by prompt and model. As the sector matures, regulators and vendors must standardize token accounting and publish cost‑per‑task metrics. Only then can organizations evaluate true value and prevent the runaway expenses that have already cost companies hundreds of millions in hidden AI spend.