Uber Caps AI Coding Costs After Using Up Annual Budget

Uber Caps AI Coding Costs After Using Up Annual Budget

Uber is limiting its employees to $1,500 in monthly token spending per artificial intelligence coding tool, Bloomberg reported Tuesday (June 2).

The limits apply only to agentic coding software, and they apply to each tool, so employees can spend $1,500 on each different tool, according to the report, which cited an Uber spokesperson.

Uber provides employees with a dashboard on which they can track their usage of different tools, and it allows individuals to seek permission to exceed the limits, per the report.

“We think this is all a pretty straightforward way to responsibly encourage agentic AI adoption and experimentation at scale across the company,” the Uber spokesperson said in the report.

The limits were implemented in recent months after Uber used up its AI budget for the full year by April, the report said, noting that The Information reported that fact in April.

In The Information’s report, Uber Chief Technology Officer Praveen Neppalli Naga, who shared the news about the AI budget, said the company was “back to the drawing board.”

Advertisement: Scroll to Continue

Bloomberg reported Monday (June 1) that Walmart limited employees’ AI usage amid rising demand. After previously providing workers with an unlimited number of tokens to use with its in-house AI agent Code Puppy, the retailer has begun offering a set number of tokens. A Walmart spokesperson said in the report that the company wants employees to apply AI to create value and that it is guiding workers to use the right AI for the task.

Fortune reported in May that Microsoft hit a ceiling too and began canceling most internal Claude Code licenses by mid-May, while redirecting engineers across its Experiences and Devices division to GitHub Copilot CLI by June 30, the end of its fiscal year.

PYMNTS reported in February that as enterprises launch production-scale AI deployments, they are discovering that enterprise AI is tethered to billing models that scale just as elastically as earlier ones but not as transparently.

The commercial infrastructure underpinning traditional software-as-a-service (SaaS) doesn’t translate cleanly to systems that meter value by computation rather than by user. In the case of AI, billing models charge by token, per application programming interface (API) call, per generated image, per inference cycle, per autonomous workflow, or, in some cases, for all of them simultaneously.