Microsoft cuts monthly AI budgets from $100K to about $10K for employees

Microsoft once let some of its employees burn through as much as $100,000 a month on AI. For most staff, that ceiling has now dropped to about $10,000.

The cuts landed in the company’s Cloud & AI division. They come alongside a broader pullback in what Microsoft spends internally on Anthropic’s Claude models.

One zero goes missing

The new per-employee caps knock the monthly allowance from $100,000 to approximately $10,000. The stated driver is concern over excessive usage.

One case reportedly stood out. A single employee spent $28,000 on AI tokens in a 28-day stretch.

Microsoft’s response has gone beyond the cap itself. Internal tracking of daily AI spending has reportedly intensified.

Executives are also pushing departments to stick to dedicated “token budgets.” Each team effectively gets an allowance, and someone is now checking whether it is being respected.

The company is also paying closer attention to which models staff choose.

The Claude bill gets a haircut

Microsoft’s internal spending on Claude was projected to top $1 billion a year as of early 2026. That figure has been cut by more than a third, to below $670 million annually.

Two forces are behind the trim. The first is simple: token costs for Claude run high.

The second is strategic. Microsoft is steering employees toward its own AI tools, chiefly GitHub Copilot, which runs on OpenAI models.

Thanks to its partnership with OpenAI, Microsoft pays no usage fees for those models. Claude, by contrast, comes with a meter attached.

Microsoft is not alone in this. Meta has made a similar move, cutting its internal Claude Code users by 50%.

Anthropic is not exactly sweating

Microsoft’s payments to Anthropic have not collapsed. Customer-facing use of Claude through Microsoft products keeps growing, helping keep Microsoft’s overall payment commitments to Anthropic close to prior peak levels.

Microsoft is buying less Claude for its own workers while selling more Claude to its customers. The money flowing to Anthropic has shifted channels rather than dried up.

What this means

For investors, the clearest signal is that even the biggest AI spenders are now treating internal usage as a cost line that needs managing.

The competitive angle is also worth watching. GitHub Copilot gains a structural edge inside Microsoft because it runs on models the company can use without usage fees.

For Anthropic, the picture is mixed rather than grim. Losing a chunk of internal usage at Microsoft and Meta is a real dent in one category of demand. But steady growth in customer-facing Claude usage through Microsoft’s platforms suggests enterprise clients still want access to the model.

Watch whether other large tech companies announce or report similar caps on employee AI spending. Watch, too, whether customer-facing Claude usage on Microsoft products keeps climbing fast enough to offset the internal reductions. And keep an eye on GitHub Copilot’s adoption numbers, since Microsoft now has every incentive to make its in-house option the default.