Microsoft Made A BIG Mistake About AI, They Want Their Real Workers Back

A year ago, the biggest names in technology were racing to push their employees to use as much artificial intelligence as possible, building internal leaderboards, offering incentives, and framing aggressive AI adoption as the competitive edge that would separate the winners from everyone else in the years ahead. Now some of those same companies are quietly scaling back, and the reason is creating one of the more awkward conversations in the industry. Microsoft has begun canceling most of its direct Claude Code licenses for employees, directing engineers to migrate to a different tool by the end of June, just six months after enthusiastically rolling access out to thousands of its staff. Uber's CTO revealed in April that the company had burned through its entire annual AI coding tools budget for 2026 within the first four months, the direct result of an internal leaderboard the company had created to rank teams by how heavily they were using the technology. Bryan Catanzaro, a vice president at Nvidia, the company that sells the chips the whole AI industry runs on, summarized the situation bluntly: for his team, the cost of compute is far beyond the cost of the employees.

The core issue is a paradox built into how AI is priced. AI tools charge by the token, a unit that roughly corresponds to a chunk of text processed, and while the cost of each individual token has been falling, the emergence of so-called agentic AI, where AI models take sequences of autonomous actions to complete tasks rather than answering a single question, has caused total token consumption to explode. Agentic AI can use anywhere from tens to a thousand times more tokens than a standard AI query, depending on how many steps a task requires, and Goldman Sachs recently forecast that this dynamic could drive a 24-fold increase in total token consumption by 2030 even as per-token prices continue to fall. The result is a system where efficiency improvements make the total bill larger rather than smaller. For individual engineers at Uber, token spending was running between $500 and $2,000 per person per month, a figure that makes the economics of replacing or augmenting human labor with AI look quite different from the projections many companies built their AI strategies on just two years ago.

Source:https://fortune.com/2026/05/22/microsoft-ai-cost-problem-tokens-agents/