More companies that actively distributed AI tools to employees are restricting usage because of the cost burden.

In a recent report, the Financial Times said Amazon, Walmart, Cisco, Uber and Meta are placing usage limits, curbing wasteful use and encouraging cheaper models as they move to control AI spending.

The newspaper said the moves show a new phase in corporate AI adoption. As employees begin using AI agents that autonomously carry out complex tasks beyond chatbots, far more computing resources are needed. It said companies are in a position where they must weigh whether the tasks assigned to AI agents are worth the money.

As Anthropic and OpenAI shift some services from flat-rate plans to token-based billing that charges by the amount of data processed, the cost burden appears to be growing further. Companies could have 10, 100 or, more aggressively, 1,000 agents per employee, and if those agents keep running, computing resource consumption is bound to rise.

Software company Workato saw usage surge after about 1,300 employees began using AI agents last summer. In May, after Anthropic switched to token-based billing, the company’s AI spending jumped.

Kosti Perikoc (코스티 페리코스), Deloitte’s global leader for generative AI, said, "Compute costs are now starting to enter the minds of CFOs and boards." He said, "Consumers and companies have been trained that AI is cheap or free, but it is never that."

OpenAI CEO Sam Altman (샘 알트먼) said, "Costs have emerged as a 'huge problem' for customers this year," adding, "Last year there was no such issue at all."

Goldman Sachs analysts last month forecast token consumption would rise 24-fold by 2030 due to AI agent use, and that a chip shortage would worsen over the next 12 to 18 months because of a surge in computing demand.