Tokens are the basic units of text or code that AI models process, and the cost of generating them can reach thousands of dollars per month for a single high-power user.
This shift reflects a growing industry-wide effort to curb the high costs of AI infrastructure, even as Microsoft’s overall revenue continues to beat market expectations.
By moving away from "tokenmaxxing"—a term used to describe maximizing AI consumption without regard for cost—Microsoft aims to ensure that its massive investments in the technology yield proportional productivity gains.
The company joins other major firms like Amazon and Adobe in throttling employee access to prevent wasteful spending on expensive computing resources that do not always move the needle for the business.
To manage these costs, Microsoft is making OpenAI’s GPT-5.6 the default model for internal work because it is more affordable than other high-end alternatives.
While the company maintains its goal of becoming an "AI-first" organization, the new guidelines emphasize "impact per token" rather than sheer volume.
These changes affect internal developers using tools like GitHub Copilot, who must now balance their technical workflows against specific budgetary constraints as the company treats AI access with the same discipline as any other critical business resource.