A company that introduced generative artificial intelligence (AI) across its organisation was billed as much as $500 million, or about 754 billion won, in usage fees in just one month, drawing renewed attention to corporate AI cost management.

BeInCrypto, a blockchain media outlet, reported on May 29 that a company, speaking on condition of anonymity, experienced an unexpected surge in costs after opening Anthropic's AI model Claude to all employees. It was reported to have operated the service without setting per-user usage limits or a budget cap.

The problem arose as a usage-based billing structure combined with companywide adoption of generative AI. As employees began using the AI without limits, token usage surged, and costs rose exponentially.

Cost increases were especially pronounced among heavy users handling complex tasks. Engineers building agent-style workflows or performing large-scale code generation work were reported to be able to generate costs of several hundred to several thousand dollars per person per month. Running AI agents around the clock to autonomously perform repetitive tasks makes costs rise even faster.

The generative AI industry has recently focused on agent-style AI and Extended Thinking features as new productivity tools, but they are also cited as drivers of massive token consumption. Unlike typical question-and-answer use, they repeatedly perform multiple steps, requiring far more computing resources for the same task.

The case is drawing more attention as more companies face similar concerns. Microsoft was reported to have reduced internal Claude code licences after monthly AI costs per engineer rose to $500 to $2,000, the report said. Ride-hailing company Uber was also reported to have spent most of its 2026 AI budget before April. Uber Chief Operating Officer Andrew Macdonald (앤드루 맥도널드) said it is becoming increasingly difficult to justify the cost-effectiveness given how companies are using AI and their business priorities.

Amazon also faced a similar problem. The company once ran an internal AI usage leaderboard, but ended the programme after employees repeatedly entered simple prompts unrelated to productivity to compete in the rankings. Usage rose but did not lead to meaningful work outcomes, and the company judged that it only increased infrastructure costs.

In the industry, many companies are criticised for treating generative AI as a flat-fee service like existing SaaS, or software-as-a-service. In reality, costs vary greatly depending on the model type, prompt length and whether agent functions are used, but companies underestimated that structure.

Anthropic provides enterprise control tools such as an administrator dashboard, per-user usage limits and compliance management features. But companies must set those features themselves, and the report said they were effectively not applied in this case.

As a result, companies are shifting their focus from expanding AI adoption to controlling costs. Major companies are introducing policies such as per-user spending caps, role-based access control, real-time cost monitoring systems and prioritising lower-cost models for simple tasks.

Claude is growing rapidly, securing hundreds of corporate customers that spend more than $1 million annually. But the case is being assessed as showing that generative AI can improve productivity while also posing a new risk that can threaten corporate budgets without proper governance.