A Microsoft employee managed to burn through $28,000 in AI tool usage over a single 28-day window, prompting the tech giant to start reining in internal compute budgets. The eye-popping number came to light after workers added a new column tracking monthly AI usage to an internal spreadsheet normally used to share salary and bonus info.
As reported by Gadget Review, the self-reported data offers a wild look into how high Microsoft employee AI token spending can climb when staff take corporate encouragement to use AI a little too literally.
A massive gap between median costs and extreme outliers
Out of Microsoft’s global workforce of over 223,000 employees, around 350 US workers voluntarily submitted their monthly AI numbers. While the overall median sat at a modest $300 per month, the spending gaps across different departments were massive.
The CoreAI group logged the highest median at around $975 per month, with individual maximums hitting $16,000. Security averaged $526 with spikes up to $10,000, while Microsoft AI averaged $490. Cloud and AI logged a $325 median with top spenders reaching $15,000, Experiences and Devices averaged $250; Azure sat at $241; and Customer and Partner Solutions averaged $134—despite being the home team of the worker who logged the $28,000 record bill. In some of those same teams, lower-end usage dropped down to just tens of dollars.
The rise of “tokenmaxxing” and vanity metrics
The root of these massive bills comes down to a trend known inside tech circles as “tokenmaxxing.” After Microsoft spent years incentivizing employees to use Copilot and AI features aggressively, some workers started gaming internal tracking dashboards by deliberately firing off massive prompts, heavy context windows, and automated queries just to boost their visible usage numbers.
Similar behavior has popped up across the industry, including at Amazon. However, Microsoft leadership is now drawing the line. In an internal memo, CoreAI Executive Vice President Jay Parikh explicitly warned staff that “tokenmaxxing is not what we are optimizing for.” He told teams to focus instead on actual business outcomes rather than racking up compute for vanity.
Tighter tracking and shifting default models
To keep costs from spiraling, Microsoft is now monitoring token value much more closely to flag individual high spenders. The company is also shifting internal default workloads over to OpenAI’s GPT-5.6 Sol while actively discouraging staff from using third-party options like Anthropic’s Claude for coding tasks.
While self-reported data from 350 workers doesn’t reflect the whole company, the leak highlights a bigger challenge facing big tech: after telling everyone to use AI for everything, setting clear financial guardrails is quickly becoming a top priority.
The post One Microsoft Worker Spent $28,000 on AI in 28 Days, Prompting Internal Audit appeared first on Android Headlines.
