AI agents are getting pricier, pushing Amazon, Meta and others to curb usage

Major companies that rushed to give employees AI tools are now trying to rein in their use as costs rise and corporate budgets come under pressure. Amazon, Walmart, Cisco, Uber and Meta, all early adopters, are setting usage caps, warning against overuse or steering staff toward cheaper AI tools as they try to control spending.

The shift reflects a new phase in corporate AI adoption. Employees have moved from chatbots to AI agents that can carry out more complex tasks autonomously, but those systems require far more computing power. The pressure intensified after Anthropic and OpenAI moved parts of their services from subscription pricing to token-based billing, which tracks the amount of data processed and exposes companies more directly to the cost of each automated task.

Executives say the economics are changing fast. Deloitte generative AI head Kosti Prikous said computing costs are now reaching finance chiefs and senior management, adding that the belief AI is cheap or free is wrong. OpenAI CEO Sam Altman said earlier this month that costs are becoming a major issue for consumers this year. Uber President and CFO Andrew Macdonald said it is becoming harder to justify AI token spending, and Uber has capped monthly token costs for each worker at $1,500 after exhausting its 2026 AI budget by April.

Walmart also limited use of one internal AI agent after imposing token caps, and Cisco says companies must balance the push to deploy agents with token availability and cost. Goldman Sachs analysts warned last month that AI agents could drive a 24-fold jump in token consumption by 2030 and worsen chip shortages over the next 12 to 18 months. The squeeze is also affecting smaller firms such as Workato, whose AI bill jumped sevenfold when Anthropic switched billing in May. Amazon and Meta have taken similar steps, while Microsoft, Amazon and Google are also offering routing tools that steer requests to the cheapest suitable model.