The latest move by Microsoft has sparked a debate across the tech industry. The company is cutting back on direct access to Anthropic’s Claude Code tool for most employees. The decision comes only months after Microsoft encouraged workers to use it.
The shift does not end Microsoft’s larger partnership with Anthropic. Microsoft still supports customer access to Claude models through its Foundry platform. But the change sends a clear signal: large-scale internal AI use may cost more than companies expected.
According to reports from Fortune and The Verge, Microsoft asked employees to move away from Claude Code and use GitHub Copilot CLI instead. The move affects engineers, designers, project managers, and other staff who had started using the tool in large numbers.
The story points to a growing issue in the AI market. Many companies promoted AI as a tool that would cut costs, speed up work, and reduce the need for labor. Yet real-world use is showing a more complex picture.
Microsoft is not the only company facing this problem.
Why Microsoft and Uber Face a Reality Check on Infrastructure Spending?
Fortune, citing The Information, reported that Uber burned through its full 2026 budget for AI coding tools in only four months. Uber CTO Praveen Neppalli Naga said the rapid spending followed internal programs that pushed teams to compete on AI use.
That raises a hard question for businesses: What happens when the cost of running AI grows faster than the value it creates?
AI tools rely on powerful computing systems. Every prompt, code request, and automated task consumes processing power. At small scale, the cost may seem manageable. At enterprise scale, the bill can rise fast.
This creates pressure for companies that want workers to use AI every day.
If businesses spend heavily on AI software, cloud services, and data center capacity, they may look for ways to control those costs. Some may cap usage. Others may limit access to selected teams or specific tasks. Companies may also push workers toward cheaper in-house tools, as Microsoft appears to be doing.
The financial impact can spread beyond tech teams.
Higher AI costs can affect software pricing, enterprise subscriptions, and hiring plans. They also challenge one of the biggest claims around artificial intelligence: that machines will replace large numbers of workers because they cost less.
In some cases, the numbers may not support that idea.
NVIDIA executive Bryan Catanzaro summed up the issue in a blunt way. For his team, he said compute costs exceed employee costs.
That statement cuts to the centre of the AI business model. If the infrastructure needed to run advanced systems costs more than human labour, companies must rethink where and how they use AI.
Future demand may push costs even higher.
Navigating the New Era of AI Economics
Goldman Sachs projects that agentic AI systems could increase token use by 24 times by 2030, reaching about 120 quadrillion tokens per month. At the same time, Gartner warns that falling token prices may not lead to lower enterprise costs. Advanced systems use far more tokens for each task, which can offset price declines.
In simple terms, cheaper tokens do not always mean cheaper AI.
There is also an energy issue.
AI can help power companies predict electricity demand, manage grids, and support renewable energy systems. But AI systems also consume huge amounts of electricity and water. Data centers need constant power and cooling. As demand grows, communities may face grid strain, rising utility costs, and pressure on local resources.
Security, misuse, and broader social risks remain part of the discussion as well.
None of this means companies will abandon AI. The technology still offers clear benefits in coding, research, automation, and workflow support. But the current trend suggests a more cautious approach.
Instead of unlimited access, firms may focus on targeted deployment. They may approve AI use only for tasks that save clear amounts of time or improve output in measurable ways.
Microsoft’s decision may reflect that shift.
The AI race is no longer just about building stronger models. It is also about making them affordable to use every day. For companies betting billions on artificial intelligence, that may become the real challenge.
