The biggest surprise in the latest wave of enterprise AI adoption is not that usage is rising, but that the bill can grow faster than expected. Microsoft and Uber are both signaling that AI does not automatically cost less than human labor, especially when it is pushed into daily work across large teams.
That shift matters because these companies have been among the most aggressive in bringing AI into the workplace. From software development to team productivity, broader use has also exposed a less convenient reality: token charges, compute demand, and infrastructure costs can stack up quickly.
Microsoft begins tightening AI spending
Microsoft is reportedly pulling back on some direct Claude Code licenses after the coding tool spread widely among employees. The move comes as the company tries to rein in AI spending that has been climbing.
Instead of continuing that approach, Microsoft is steering developers toward GitHub Copilot CLI. The decision follows a period in which the company encouraged thousands of employees to experiment with AI-assisted coding.
That push was not limited to software engineers. Microsoft also promoted AI use among designers and project managers, which widened adoption across multiple parts of the organization.
The problem is that broad usage brings broad costs. The more employees rely on AI for everyday tasks, the larger the operational burden becomes.
Why AI costs can rise so quickly
AI pricing works differently from a normal salary expense. Most systems charge by token for each task they process.
Every prompt, every response, and every automated workflow consumes tokens that must be paid for continuously. Once many teams start using the same tools at scale, token usage can rise sharply and create a large extra expense.
That is why an AI subscription is only one part of the total bill. Compute and infrastructure add another layer of cost as usage intensifies.
Uber runs into the same problem
Uber has faced a similar pressure, and on a significant scale. CTO Praveen Neppalli Naga said earlier this year that the company had already used up its entire AI coding budget for 2026 in just four months.
The case shows that cost overruns do not come only from having AI available. They also come from actively encouraging employees to use it more often.
Uber has expanded AI use through internal rankings and productivity dashboards. Those tools can drive adoption, but they also increase token consumption, compute needs, and infrastructure spending.
Human labor still wins in many routine tasks
On paper, a single AI token may look inexpensive. In practice, repeated costs across very high volumes can become difficult to manage.
That is why AI is not automatically cheaper than people. For many routine tasks, continuous AI use can still cost more than hiring human workers.
Research from the Massachusetts Institute of Technology supports that view. The study says AI is economically efficient in only a limited set of roles, not across the wide range of jobs often assumed to be replaceable.
A similar assessment has also come from the AI chip industry itself. Nvidia executive Bryan Catanzaro has acknowledged that in some situations, running AI systems can cost more than employing people.
AI still has value, but not as a universal replacement
These developments do not reduce AI’s usefulness. The technology still helps teams move faster and work more efficiently.
But at least for now, it is not always a sensible full replacement for human labor. Companies still have to ask whether the operational benefit is worth the cost.
The financial pressure may also continue to grow. Industry players expect agentic AI, which can handle multi-step tasks more independently, to increase enterprise AI spending in the years ahead.
That capability offers deeper automation, but it also demands more processing, more tokens, and larger infrastructure. Microsoft and Uber now show that the central AI question is no longer only about what the technology can do. It is also about whether the cost of running it makes sense at enterprise scale.
Source: sundayguardianlive.com