Companies today are pressuring employees to use AI as much as possible in order to reap the productivity benefits of the technology. But that pressure is leading to cracks, and those cracks may be irreparable.
Microsoft has reportedly begun revoking most of its direct cloud code licenses The VergeThis led engineers to use the GitHub Copilot CLI instead. This comes just six months after the company first opened up access to cloud code, encouraging thousands of its developers, project managers, designers, and other employees to experiment with coding. The technique became popular rapidly. Probably very popular. The scale at which employees use it is now prompting the company to reverse course on a tool its own engineers had come to rely on. Canceling the cloud code license will not affect Microsoft’s Foundry deal, which includes investing up to $5 billion in Anthropic and providing Foundry customers access to the cloud model, as well as Anthropic’s $30 billion commitment to purchase Azure compute capacity. The Verge.
Microsoft isn’t the only company that is reducing its internal AI usage. Uber CTO Praveen Nappalli Naga told Information Back in April, the company spent its entire 2026 AI coding tool budget in just four months. This came after the company actively encouraged adoption of the AI tool through internal leaderboard ranking teams.
This report may spoil the bets placed on technology by the biggest technology companies. While some are clinging to the promise of an AI “renaissance” or “revolution,” the cost of adoption is proving to be a stubborn obstacle. These developments also suggest that the economics of replacing or augmenting human labor with AI may be more complex than some early predictions originally implied. That falls in line with what Brian Catanzaro, vice president of applied deep learning at Nvidia, recently said in an interview with Axios.
“For my team, the cost of computation far outweighs the cost of staff,” he said.
Anthropic did not immediately respond. LuckRequest for comment. Microsoft did not provide any comment.
An emerging AI paradox: cheap tokens, big bills
Uber and Microsoft aren’t the only companies pushing employees to use AI as much as possible. Like Uber, a Meta employee created a leaderboard, dubbed “Cloudonomics” after Anthropic’s AI model, to track which employees are using AI the most. Amazon is motivating its employees to use “tokenmax” or as many AI tokens (the basic building blocks of AI computation) as possible.
But with a token-based pricing system, work becomes more expensive with greater usage and better efficiency. Goldman Sachs recently predicted that agentic AI could drive a 24x increase in token consumption by 2030 as consumers and enterprises adopt AI agents, reaching 120 quadrillion tokens per month. As businesses turn to AI agents to increase productivity, the total cost can increase rapidly even if the price of each token falls.
But as consumption increases, the cost of individual AI tokens is expected to decline rapidly. A recent report from research firm Gartner found that by 2030, estimating on a trillion-parameter LLM – in simple terms, a highly sophisticated AI model – AI will cost firms about 90% less than by 2025. Nevertheless, Gartner predicted that cheap tokens would not translate into cheap enterprise AI because the agent model requires far more tokens per task than the standard model, increased consumption could outweigh the falling unit cost, and AI providers would not fully support AI. Passing on lower costs to consumers. In turn, estimated costs are likely to increase.
“Chief product officers (CPOs) should not confuse the deflation of commodity tokens with the democratization of marginal logic,” Gartner senior director analyst Will Sommer warned in a statement.
This reality may complicate grand plans by some companies to deploy AI agents. Nvidia CEO Jensen Huang recently said that he thinks 100 AI agents will one day work with every employee at his company.
Huang is part of a broader wave of CEOs who are promoting an agentic future in which digital workers work across the entire enterprise. But if token consumption grows faster than unit costs decline, the future could lead to a much heavier bill than officials expect.
<a href