AI promised cost savings, but Microsoft and Uber say it’s costing more than human workers

Microsoft has reportedly begun revoking most direct Claude Code licenses and redirecting its engineering workforce to the GitHub Copilot CLI instead. The rollback comes just six months after the tech giant opened up access to Claude Code to thousands of developers, project managers, designers and other staff, spurring widespread experimentation with AI-powered coding.

The adoption was quick and enthusiastic. Maybe he was too fast. According to a report by , the sheer scale at which employees have adopted the tool has now led to the firm retreating from the technology on which its own engineers have become increasingly dependent. threshold.

The decision will not affect Microsoft’s broader business relationship with Anthropic. The company’s Foundry deal, which includes an investment of up to $5 billion in Anthropic and gives Foundry customers access to Claude models, remains intact, as does Anthropic’s $30 billion commitment to purchase Azure computing capacity.

Uber Consumed Its Entire 2026 Artificial Intelligence Budget in Four Months

Microsoft is not an isolated case. Praveen Neppalli Naga, Uber’s chief technology officer, said: InformationHe said in April that the ride-hailing company had exhausted its entire 2026 budget for AI coding tools in just four months of the year.

This statement is particularly notable given that Uber actively encourages adoption and uses internal leaderboards to rank teams based on their use of AI tools.

The overall trend at both companies points to a tension that gets little attention in workplace AI discussions: The more companies push workers to use the technology, the faster the costs pile up.

AI Token Economics: Why Don’t Cheaper Prices Lead to Cheaper Bills?

At the heart of the issue is how AI computing is priced. Large language models charge per token, which is the basic unit of text that the model processes and creates. Luck. According to this model, greater efficiency and greater utilization are financially indistinguishable: both increase total spending.

Many large technology companies are actively increasing token consumption even further. Amazon has encouraged its staff to use the term “tokenmaxx,” meaning to use as many AI tokens as possible. An employee at Meta created an internal monitoring tool called “Claudeonomics” to track which employees were using AI the most.

Goldman Sachs estimates that agency AI systems that act autonomously in multiple steps rather than responding to single queries could lead to a 24x increase in token consumption by 2030, reaching 120 quadrillion tokens per month as businesses deploy AI agents at scale.

The unit price of these tokens is expected to drop significantly. research firm Gartner’s It predicts that by 2030, running inference on a large language model with one trillion parameters will cost AI providers almost 90% less than by 2025. But Gartner warned that this price deflation will not translate into lower corporate bills. Agency models require significantly more tokens per task than standard models, consumption growth may outpace falling unit costs, and AI providers are unlikely to pass on the full benefits of cost reductions to enterprise customers.

“Product Managers should not confuse the deflation of commodity tokens with the democratization of border logic,” said Will Sommer, senior director analyst. Gartner’s.

The Cost of Computing Already Exceeds the Cost of Employees

Perhaps the most striking acknowledgment of AI’s cost problem came from the technology industry itself. Bryan Catanzaro, NVIDIA’s VP of applied deep learning, addressed the issue head-on in a recent interview. axios.

“The cost of computing for my team far exceeds the costs of employees,” Catanzaro said.

This comment carries weight given that NVIDIA is a major supplier of chips powering AI infrastructure globally. It suggests that the economics of replacing or augmenting human labor with AI may be much more complex than initial estimates implied.

Agency AI Goals Face a Challenging Financial Reality

The cost pressures now emerging contrast with the broad visions for AI deployment that tech executives have publicly expressed. NVIDIA Chief Executive Officer Jensen Huang said he expects to one day have 100 AI agents working alongside every human employee at his company.

Huang is part of a broader chorus of corporate leaders advocating for an efficient future where digital workers perform tasks across the organization with limited human supervision.

If token consumption continues to rise faster than the decline in unit costs, this future could impose a much heavier financial burden than executives have openly acknowledged. Luck guesses. Early signals from Microsoft, Uber and others suggest that, under current pricing and usage models, the economics of large-scale AI deployment remain deeply unresolved.