To rein in wanton AI spending, we need AI ‘nutrition labels’

As tech companies increasingly rely on artificial intelligence, they face a growing concern: the hidden costs of AI usage. The problem is that many employees and decision-makers aren't aware of just how much energy and computing resources are required for even a single AI prompt. Google, for example, now processes over 3.2 quadrillion tokens a month, seven times more than a year ago. Uber, on the other hand, depleted its entire AI budget by the fourth month of 2026.

Experts believe companies need "AI nutrition labels" – a transparent way to gauge the cost and energy consumption of each prompt. Just as nutrition labels on food help consumers make informed choices, these labels could guide employees to use AI more efficiently. By showing token estimates and computational intensity, these labels would help employees weigh the cost against the value their work delivers.

An AI label could reveal unnecessary complexity in a workflow or prompt, allowing users to adjust their requests before wasting tokens. This could prevent multiple attempts, irrelevant context, and failed outputs that add up to significant costs. For example, a single inefficient AI workflow might run up a $2.25 bill, while two such prompts a day could cost a company $99 a month. Over 1,000 employees over a year, that's $1.19 million in needless expenditure – not to mention the wasted energy.

These AI nutrition labels would also help companies make smarter purchasing decisions, similar to Energy Star ratings for appliances. As AI systems become more autonomous, the lack of transparency about costs and energy use will become untenable. Companies will demand this visibility, and AI makers will eventually provide it. In the end, understanding the "ingredients" of an AI answer will be as important as the answer itself.