Unit economics can help, but only if the units are real. Cost per transaction, cost per customer, cost per claim, cost per inference, or cost per order can be useful. A vague internal allocation metric that nobody uses to change behavior is accounting theater, not unit economics.

AI will expose weak spots

The AI findings are probably the most important part of the report, but even there, the story should be read with more skepticism. Flexera says generative AI public cloud services are now used by 58% of respondents, and all respondents use generative AI in some capacity. That sounds dramatic, but “using generative AI” can mean many things. It could mean a production AI product or a developer assistant or an experiment with a cloud API. It could mean a pilot that will never survive procurement review.

AI will make the cloud cost problem worse. GPU resources, inference costs, data movement, storage growth, model monitoring, vector databases, and duplicated experimentation environments will create bills that many enterprises do not yet understand. The old cloud problem was that teams could provision infrastructure with a credit card. The new problem is that teams can provision very expensive AI services before finance, architecture, security, and data governance know what happened.