Microsoft has told its engineers that maximising AI token usage is not a company goal and set new spending limits on internal tools.
Jay Parikh, an executive vice president at Microsoft, sent an email to staff stating that the firm is updating its guidance to manage token spend with the same discipline applied to other critical resources. The message explicitly warns employees against “tokenmaxxing,” a term for pushing AI usage to its limits regardless of value. Parikh wants staff focused on outcomes that move the needle for customers and the business, rather than simply consuming more compute.
GitHub Copilot, an AI coding assistant owned by Microsoft, is the primary tool in question. Parikh noted that the company is accelerating its use of the tool but requires everyone to be aware of how they consume tokens. He stated that the firm is moving toward OpenAI‘s GPT-5.6 model because it is cheaper to use than other options. As of July 2026, Microsoft divisions will have an “AI token budget target.” Employees will be able to track their individual spending, though no specific target value has been shared yet.
The guidelines indicate that some engineers currently spend between hundreds of dollars and a few thousand dollars a month on tokens. Parikh said further restrictions may be placed on specific decisions as the company monitors overall spend.
This policy shift comes despite strong financial results. Microsoft’s latest earnings report shows revenue, operating income, and net income all increased for the year, beating Wall Street expectations. The change is not driven by a cash shortage. It aligns with moves at other organisations including Amazon, Adobe, Atlassian, and Citi, which are also trying to throttle employee AI use.
One Microsoft employee, who asked to remain anonymous because they were not authorised to speak to the press, told me the move feels like an admission that hosts of AI infrastructure cannot afford their own products. “It’s very telling, to me, that a company which has invested so heavily in AI and subsidized so much AI inference is now guiding its own employees to curb spending,” the person said. They added that if this is even partially true, it raises questions about how the companies selling these tools can manage their own costs.
Parikh stressed that Microsoft does not want to slow down progress toward becoming “AI-first.” He wrote that the company is not optimising for fewer tokens, but rather for more impact per token. Microsoft did not immediately respond to a request for comment.
What it means
For developers and data scientists, the change means a shift from free-for-all experimentation to managed consumption. Teams will face hard caps on how many tokens they can use for training, testing, or generating code. This forces a focus on efficiency: writing shorter prompts, reusing models instead of training new ones, and auditing what tools are actually necessary. The era of unlimited internal compute for AI experiments is effectively over.
