Companies that rushed to put artificial intelligence tools in the hands of employees are beginning to restrict their use as the cost of deploying the technology at scale strains corporate budgets.

Amazon, Walmart, Cisco, Uber and Meta have introduced spending caps, discouraged wasteful use or pushed workers toward cheaper models as they seek to control rising AI costs.

The shift comes as employees move beyond basic chatbots toward AI agents capable of carrying out complex tasks autonomously. These systems require significantly more computing power and can continue consuming resources without direct human oversight.

The pressure has increased as AI providers including Anthropic and OpenAI move some customers from flat subscriptions to token based billing, exposing companies directly to the cost of each prompt and automated workflow.

OpenAI chief executive Sam Altman said this month that cost had become a major issue for customers after receiving little attention last year.

Uber has capped employee spending at $1,500 per month on individual AI tools after using its entire 2026 AI budget by April.

Uber president and chief operating officer Andrew Macdonald said it was becoming harder to connect growing token spending with measurable improvements in consumer products.

Walmart has also capped the number of tokens employees can use through its internal AI agent after adoption of its Code Puppy coding platform surged.

Amazon warned employees last month to stop using AI simply for the sake of using it after engineers began deploying agents to climb internal adoption leaderboards. Meta introduced similar measures in April.

Cisco president and chief product officer Jeetu Patel said the infrastructure required to operate an agent is meaningfully greater than that needed for a chatbot.

A single employee could eventually oversee tens or hundreds of agents that continue running and consuming compute, he said.

Goldman Sachs analysts expect AI agent adoption to drive a 24 fold increase in token consumption by 2030 and intensify chip shortages over the next 12 to 18 months.

Smaller companies are facing similar pressure. Software group Workato said its AI spending increased sevenfold in one day after Anthropic moved the company to token based pricing in May.

Workato has responded by encouraging employees to use older and cheaper models rather than restricting access entirely.

AI platforms are also introducing tools that automatically direct tasks toward less expensive models when frontier systems are unnecessary.

Microsoft, Amazon and Google have launched routing systems that select the most suitable model from a customer approved group, while some companies are turning to open source models that can run on their own servers or employee devices.

The push to control spending could pressure the growth of leading AI labs as companies begin demanding clearer evidence that higher token usage is producing meaningful gains in productivity and revenue.