Microsoft Employee Burns $28,000 on AI in a Month, Prompting Crackdown on ‘Tokenmaxxing’

Microsoft’s internal employee database has revealed anomalous AI usage: one specialist burned through about $28,000 worth of compute and tokens in 28 days. The incident has exposed the growing Silicon Valley phenomenon of “tokenmaxxing” (tokenmaxxing), prompting the company to impose strict personal budgets for AI work.

A $28,000 anomaly against a $300 median

The spending data came to light thanks to an internal spreadsheet where employees in US divisions anonymously share salaries and productivity metrics.

According to an investigation by Business Insider, among the 350 employees who disclosed their data, the median spend was just $300. The record holder, who spent $28,000, exceeded the company’s standard by 93 times. Notably, the employee is in the sales and customer solutions division (Customer and Partner Solutions), where the average AI spend was the lowest at $134.

According to Livemint, five-figure bills were recorded in several departments:

How you can spend $28,000 on AI in four weeks

In a corporate setting, such a bill doesn’t mean endless manual chatting with a bot.

The astronomical spend is driven by autonomous coding agents: they continuously scan codebases with millions of lines, keep massive documentation sets in context, run tests in parallel, rewrite code, and repeatedly call expensive flagship models with maximum reasoning effort.

The ‘tokenmaxxing’ phenomenon and management’s response

In Big Tech corporate culture, the term tokenmaxxing has emerged — artificially inflating the volume of tokens used to demonstrate high engagement with AI to management.

This imitation of intense activity drew a sharp response from leadership. As The Register reports, Jay Parikh, executive vice president of CoreAI, sent an official message to employees:

“Tokenmaxxing is not what we optimize for. The company cares about results per unit of cost, not burning compute for pretty dashboard numbers.”

Burning tokens doesn’t help your career

Journalists compared AI usage metrics with employee performance and found that there is no direct correlation between token spend and career advancement. Workers with thousands of dollars in AI costs did not receive higher performance ratings, bigger bonuses, or faster promotions.

To curb the uncontrolled growth of server infrastructure bills, Microsoft has implemented a set of measures:

• launching personal dashboards to monitor individual AI spending;
• allocating fixed token limits for each team;
• mandating that routine internal tasks use the more cost-effective GPT-5.6 model.

The precedent clearly illustrates a new phase in corporate AI: tech giants are moving from calls to “use AI everywhere” to strict financial audits of its real business value.