Microsoft Implements Token Budgets to Curb Internal "Tokenmaxxing"

Microsoft is introducing internal AI token budgets and cost-control measures to curb "tokenmaxxing," a trend where employees excessively call AI models beyond what is necessary. Under the new guidelines, Microsoft will set department-level token budgets starting July 2026 and has set a lower-cost model as the default. This shift highlights a growing industry-wide concern over the soaring costs of generative AI, prompting tech giants to transition from unrestricted experimentation to strict resource management. It signals that even well-funded AI leaders must prioritize actual productivity and return on investment over raw usage metrics. Some Microsoft engineers have reportedly run up monthly token bills of hundreds or thousands of dollars, prompting the company to make a lower-cost model the default and phase out external tools like Anthropic's Claude Code. Similar budget overruns have affected other tech giants, such as Uber, which exhausted its entire 2026 AI coding budget in just four months.