There is a striking twist in the AI rush: In some high-usage enterprise deployments, the systems meant to improve efficiency are also becoming a major cost themselves. Axios reported that some companies and teams see AI compute costs rivaling or exceeding the salary costs of the employees using the tools. That is of course not a common situation but it is still a meaningful signal about how expensive heavy AI usage can become.

The most notable example in the report came from Nvidia. According to Axios, Bryan Catanzaro, Nvidia’s vice president of applied deep learning, said that for his team, “the cost of compute is far beyond the costs of the employees.” The comment stood out because it came from a senior Nvidia executive. Nvidia is one of the companies most closely tied to the AI infrastructure boom, so an internal remark like that offers a useful reality check on the economics of large-scale AI use.

The reason is not hard to understand, even if the exact bill varies by company and workflow. Large models require substantial compute, and intensive use of coding agents or developer assistants can drive up token consumption quickly. A light chatbot workflow, such as drafting emails or summarizing notes, is one thing. A coding workflow that repeatedly analyzes code, generates revisions, and runs multiple prompt cycles and tests is another. 

Axios also pointed to a report from The Information about Uber’s internal AI tooling costs. According to that report, Uber’s 2026 budget for AI coding tools had already been exhausted early in the year, with CTO Praveen Neppalli Naga saying the budget had been “blown away already.” 

AI has often been framed as a way to amplify labor, automate repetitive work, or in some cases reduce headcount pressure. In practice, at least in some high-intensity deployments, it is also becoming a new infrastructure cost center. That does not mean the productivity upside is fictional. It means the old assumption that more automation automatically leads to lower costs is proving too simplistic, even if true in many instances.

The direction is clear enough. AI infrastructure, software, and cloud services are pushing enterprise technology budgets higher, and buyers are under more pressure to justify where that spending goes.

The more real AI becomes inside large organizations, the more it gets judged like any other enterprise tool. Does it save time, improve output, reduce bottlenecks, or create enough business value to offset its cost? Those questions are less glamorous than product demos, but they are the ones that decide whether a technology becomes durable infrastructure or just an expensive experiment.

There’s one thing most people overlook in the conversation: computing efficiency is going to skyrocket in the coming months and years. Tokens will be generated at a much cheaper price, and tasks that have reached a “good enough” level will cost dramatically less, freeing budget for higher-value workloads.

Filed in . Read more about AI (Artificial Intelligence), Office and Social.