Who’s tracking AI spend?
A survey posted on social said 45% of organisations reward AI spending but only 33% enforce spend limits and 22% report no tracking at all (x.com). Other practitioners recommended tagging resources and setting quotas to prevent agentic AI overruns, and one post claimed 92% of such overruns come from poor visibility ( ).
A new survey making the rounds in the AI operations community says many companies still treat artificial intelligence spending as growth, not something to cap. (x.com) In that poll, 45% of respondents said their organizations reward AI spend, 33% said they enforce spend limits, and 22% said they do not track the spend at all. The post was published on X and circulated this week among cloud and FinOps practitioners. (x.com) The basic problem is simple: every prompt, model call, agent step, and cloud workload can add cost, often in tiny increments that only become visible after the bill lands. OpenAI’s usage dashboard, for example, shows usage and cost data, but its help center says data is not consolidated across organizations by default. (help.openai.com) That accounting gap gets wider with agentic AI, where software can make repeated model calls on its own. KPMG said in January 2025 that a majority of large-company leaders were looking to agentic AI to help scale AI across the enterprise. (kpmg.com) Practitioners responding to the survey argued for old cloud-finance habits: tag resources, assign owners, and set quotas before usage scales. One X post recommended tagging and quotas specifically to stop agentic AI overruns. (x.com) The large cloud vendors already expose those controls. Amazon Web Services says cost allocation tags let customers categorize and track costs, and Google Cloud says billing budgets can track actual costs against planned costs and trigger alerts or automated responses. (docs.aws.amazon.com) (docs.cloud.google.com) Microsoft has gone a step further in recent guidance for Azure OpenAI users. In a March 2026 post, Microsoft said Azure Budgets, Action Groups, and Automation Runbooks can automatically disable Azure OpenAI access when a budget threshold is breached. (techcommunity.microsoft.com) Amazon is also adding more AI-specific cost visibility. AWS documentation published this month says Amazon Bedrock now supports cost allocation by Identity and Access Management principal identity and tags, so organizations can map model usage to the user or role that made the call. (docs.aws.amazon.com) Not every claim in the social thread is easy to verify. A separate X post said 92% of agentic AI overruns come from poor visibility, but the post did not cite a public study or methodology. (x.com) The thread’s main point still lines up with where enterprise AI is heading: more autonomous systems, more model calls, and more pressure to tie each dollar to a team, project, or product. Without that tagging and budgeting layer, “AI spend” is just another line item that grows faster than anyone can explain. (mckinsey.com) (docs.aws.amazon.com)
