Microsoft employees built their own pay spreadsheet, and it shows two companies

Around 600 Microsoft employees have entered their pay into a crowdsourced spreadsheet circulating internally, and the resulting picture is of a company that now compensates two quite different populations.

Base salaries run from $111,000 at level 59 to $280,000 at level 67, with Cloud and AI roles reaching $450,000, which is the kind of information Microsoft’s own employees have been asking for since it dropped the pay question from its internal survey.

Raises across the sample ran from 0% to 3.5%, which for most people is at or below inflation, and cash bonuses spanned $2,000 to $110,000.

Stock is where the company splits in two. Annual stock awards in the sample ranged from $10,000 to $1.4mn, with the top of that range concentrated in Cloud and AI, and equity of that size changes what a job is rather than what it pays.

The pattern reported alongside the raw numbers is that engineers on generative AI work receive larger initial grants and more frequent out-of-band refreshers than colleagues at identical levels in conventional software groups. Level, in other words, no longer tells you what someone earns.

That is a meaningful break with how Microsoft has traditionally worked. The level system exists precisely so that a level 65 in one organisation is comparable to a level 65 in another, and a six-figure divergence in annual vesting at the same level empties the framework of much of its meaning.

Xbox illustrates the other end. Base salaries there cap around $218,000 in the sample, well below what the same seniority commands in Cloud and AI, at a company where the games division was until recently a strategic priority worth $69bn in acquisitions.

Out-of-band refreshers are the mechanism doing most of the work here, and they are worth explaining. They are additional stock grants made outside the annual review cycle, typically to retain someone a competitor is likely to approach, which means they flow to whoever the market is currently bidding for rather than to whoever performed best.

The people on the losing side of this are not marginal. Engineers maintaining Windows, developer tools, and established cloud services produce the cash flow funding the AI infrastructure spending, and several of them are now being asked to add generative features to their products without a corresponding change in what they are paid.

None of that is unique to Microsoft. Anthropic has been paying the industry’s highest salaries and its chief executive has been visibly surprised that people take them for the money, and the whole sector has spent two years bidding for the same few thousand researchers.

What is specific to Microsoft is the contrast with everything else happening to its payroll. The company has run layoffs, offered its first voluntary retirement programme, and been part of a broader Big Tech pattern of converting payroll into AI capital expenditure.

A 0% raise reads differently in that context. It is not the same message when it arrives in a year of record infrastructure spending and million-dollar packages two floors away.

Pay transparency laws have made some of this visible anyway. Several US states now require salary ranges in job postings, which means employees can often see what a new hire is being offered for their own role, and the gap between that and their last raise is not hard to calculate.

The sample deserves the usual caution. Six hundred submissions represent well under 1% of Microsoft’s roughly 223,000 employees, the data is self-reported and unverified, and people with unusually good or unusually bad outcomes have more reason to fill in a spreadsheet than people in the middle.

Microsoft has not commented. Crowdsourced pay data of this kind tends to appear when employees stop trusting the official channels, and a company that removed the pay question from its own survey has fewer grounds than most to object to staff answering it themselves.