Enterprise AI is running into a cost wall that the sales pitch mostly skipped over.
The problem is not just “AI is expensive.”
The deeper problem is that enterprise software budgeting was built for predictable licensing, while frontier AI pushes firms into variable consumption.
One recent report gives an extreme example.
An AI consultant said a client spent $500 million in a single month on Claude after failing to set usage limits.
The mechanism is not mysterious.
At enterprise scale, these systems do not sit idle waiting for a person to type.
They can run coding tasks, test code, browse, call models repeatedly, trigger more calls, and spread across teams and workflows before anyone has a clear view of the burn.