A leaked Microsoft spreadsheet shows one employee spent $28,000 on AI in 28 days against a $300 median
Business Insider's review of an internal pay-transparency document found a 90x gap between typical and outlier AI usage inside Microsoft -- weeks after a CoreAI memo already pushed back on employees running up token bills without checking whether the work behind them was any good.
On August 24, 2026, Business Insider reported it had reviewed an internal spreadsheet Microsoft employees use to voluntarily and anonymously compare pay -- one that gained a new self-reported column this year, "AI $ Usage Per Month." Of about 350 US employees who filled that field in (out of nearly 600 total entries, against Microsoft's roughly 223,000-person global headcount), the median reported spend was about $300 over a 28-day window. One person, in the Customer and Partner Solutions organization, reported $28,000 in the same period.
The spread underneath the median
Business Insider's own breakdown shows the variance wasn't confined to one outlier: CoreAI's 14 respondents reported a $975 median -- more than triple the company figure -- with individual entries as high as $16,000, while Cloud + AI's 107 respondents ran as high as $15,000 and Security's 12 respondents up to $10,000. The outlet is explicit about the data's limits: self-reported, voluntary, not comprehensive, and -- per its own analysis -- showing no meaningful correlation between an employee's reported AI usage and their bonus, raise, or promotion odds. A Microsoft spokesperson said the usage-tracking tool itself is "still in early testing."
A memo that came first, not after
The specific numbers are new, but Microsoft's concern about them isn't: on August 4, three weeks before Business Insider's story ran, CoreAI EVP Jay Parikh had already told engineers to default GitHub Copilot to OpenAI's GPT-5.6 Sol instead of the Claude models it had been auto-routing to, telling staff, according to CNBC's reporting, "Tokenmaxxing is not what we are optimizing for." Microsoft divisions had begun operating under formal AI token-budget targets the month before, in July 2026, though Parikh said CoreAI itself had not yet set hard per-employee caps.
This is Merit AC's own thesis playing out inside one of the world's largest AI spenders, without any of this site's own scoring infrastructure involved: a spend number nobody had systematically tracked turned out to vary 90-fold between a typical employee and an outlier, and the company's own response wasn't to celebrate high usage as engagement -- it was to question whether the extra tokens were producing outcomes worth the money. That's the exact distinction between spend and value this site's tracker exists to make legible, running here by memo and leaked spreadsheet instead of a dashboard.