Microsoft Will Pay a Quarter of Its Staff to Leave by New Year's Eve. It Says AI Has Nothing to Do With It.

Microsoft is offering roughly 60,000 employees the chance to take severance and leave by 31 December. Its chief people officer announced it in a blog post on Monday. The company's position is that this has nothing to do with artificial intelligence.

That position is not obviously wrong. It is also not the whole of what Microsoft has said on the subject this year.

The Voluntary Exit Offer is open to staff in the United States, Canada, India and the United Kingdom at level 70 and below, which on Microsoft's internal scale runs up to and including some people with vice president titles. They have until early December to decide, and anyone who accepts leaves on 31 December. About 60,000 people are in scope, according to a person familiar with the matter who spoke to CNBC. Microsoft reported 223,000 full-time employees at the end of June, down from 228,000 a year earlier, so the offer reaches something like 27% of the company.

The terms are unremarkable, which is the point. Microsoft describes the package as equivalent to its standard severance: base pay scaled by tenure and level, continued stock vesting, healthcare for a fixed period, career transition services. Reporting puts the ceiling near 39 weeks of base pay for most US employees. There is no age or tenure test, and performance is not a factor. Unlike April's program, there is no extended healthcare. This is not a sweetener, which is why the company objects to the word buyout. It is the severance you would get if you were selected for a layoff, made available on request.

Then the sentence that has travelled further than the memo. Microsoft told The Register that its language about "changing needs" is not a euphemism for AI.

Six months earlier, on the April earnings call, Satya Nadella was asked where the money for all this AI spending would come from. His answer named a line item.

"IT budgets are going to have to be reshaped by a combination of business outcomes making their way into IT budgets and maybe reallocation from other line items on the income statement, like OpEx."

Operating expense. At a software company, the largest operating expense is people.

Both statements are on the record. They are not strictly contradictory. A company can fund AI out of opex in one breath and offer a voluntary exit in another for reasons that have nothing to do with it. But they describe the same dollar from opposite ends, and the distance between the two descriptions is the most interesting thing Microsoft has said all year.

Who is excluded is the actual announcement

Microsoft has declined to publish the list of participating organizations. The lists in circulation come from people familiar, not from the company, and they differ in the details.

The shape is consistent across them. Out: the data center organization, sales, marketing, commercial business, security, parts of corporate and legal. In, at least partly: AI experiences, Copilot, CoreAI, Windows and Devices, Xbox, finance and human resources. Reuters has the eligible list including AI experiences, Copilot, finance, HR, Xbox and Windows and Devices. The Verge adds CoreAI, and puts cloud and AI, MicrosoftAI and corporate affairs in the partially participating column.

Read as an org chart, that says something specific. The two functions most protected are the ones that produce capacity and the ones that produce revenue. The functions most exposed are the back office and the product organizations sitting on top of the build.

Finance and HR are not random inclusions. They are the two departments enterprise agent software has been marketed against for three years, the ones every vendor demo quietly promises to compress. Xbox sits in More Personal Computing, where full-year revenue slipped 1% to $54.1 billion and the June quarter fell 4%, and where Microsoft took a gaming impairment charge. If you were designing an exit offer to land where growth is weakest, this is roughly what it would look like.

The data center organization being outside the offer is the detail worth sitting with. The part of Microsoft that has become a physical asset is not being asked to leave. The part that administers it is.

The arithmetic refuses to behave like a cost program

This is where the story stops being about a memo and starts being about a spreadsheet.

The Register applied Google's 2025 experience to the eligible pool. Google's equivalent offer drew between 3% and 5% participation, so 60,000 eligible employees become somewhere between 1,800 and 3,000 departures. Microsoft told GeekWire it expects a much smaller share than the 30% its April program drew, and says it has no target.

Wells Fargo went the other way. In a note published the same day, the bank modelled a one-third acceptance rate and concluded the cost actions could produce a 2 percentage point annualized benefit to operating margin. It kept an Overweight rating and a $725 price target.

That is a six-fold disagreement about the size of the program, from people paid to be right about it. The honest position is that nobody knows, including Microsoft, because the program depends on 60,000 private decisions made between now and December.

What can be sized is the order of magnitude. Microsoft guided a $900 million one-time charge for the April program, which covered about 8,750 eligible employees at level 67 and below, of whom roughly 30% accepted. Divide one by the other and you get something north of $340,000 per departure. That is my arithmetic on two disclosed numbers, and it almost certainly overstates the average here, because April's pool was older, longer-tenured and US-only while this one adds Canada and India and drops the age test.

Apply it as a ceiling. At 3,000 departures the charge lands near $1 billion. At Wells Fargo's 20,000 it lands around $7 billion.

Now put those next to the capital program. Microsoft expects roughly $175 billion of capital expenditure in calendar 2026, after a lease accounting change moved the headline down from $190 billion without changing the spending itself. It has disclosed $329.1 billion of data center leases that have not yet commenced, with terms starting between fiscal 2027 and fiscal 2033. Quarterly capital spending is guided above $50 billion for the September quarter alone.

A charge of one to seven billion dollars is not funding a buildout measured in the hundreds of billions. It cannot be. Set it against the $155.2 billion of operating income Microsoft booked in fiscal 2026 and it is a rounding error at one end of the range and a decent quarter's growth at the other. Whatever else the Voluntary Exit Offer is, it is not a source of capital.

It is a margin lever. Which raises the question of what margin it is defending.

A 25-year building and a nine-week decision

On the same July call where Amy Hood put calendar 2026 capex at $175 billion, she made a smaller change with a longer tail. From fiscal 2027, Microsoft will depreciate data centers and office buildings over 25 years instead of 15.

Her reasoning was operating history and expected use of the assets. The mechanical effect is to spread the same concrete and steel across an extra decade, lowering the annual depreciation charge and flattering every margin comparison that follows.

Two-thirds of the quarter's capital spending went to short-lived assets, mostly GPUs and CPUs. Those do not get 25 years. Microsoft is stretching the life of the shell while the silicon inside depreciates at the pace the industry actually sets. That is the whole tension in one accounting note.

It also makes the exit offer legible. As the asset base matures, depreciation becomes the binding constraint on operating margin. Microsoft has already guided fiscal 2027 operating margin down less than a point while carrying quarterly capex above $50 billion, guided operating expense growth in the mid to high single digits, and promised positive free cash flow. Those three things have to reconcile somehow.

Capital spending is contracted. Leases are contracted. Depreciation is arithmetic. Payroll is none of those things, which is why it moves when the others cannot. Nine weeks from memo to final day is not a long time to change your mind about a career, but it is a very short time to change a cost base.

Hood told investors in April, six months before this offer existed, that she expected headcount to decline year over year in fiscal 2027. Total headcount was already down 2% in the June quarter while revenue grew 18%. The direction was set before the memo. The memo is the mechanism.

Microsoft's case is better than the headlines allow

It would be easy to write this as a company automating its own staff out of the building. The evidence for that is circumstantial, and Microsoft has a coherent alternative account.

The April program worked by the company's own measure. More than 30% of eligible employees took it. Coleman told GeekWire in July that the uptake reduced the size of the layoffs Microsoft announced that month, which is a real thing to have achieved: 4,800 people lost their jobs in July anyway, but the implication is that the number would otherwise have been higher. Microsoft also says it heard from employees who would have taken the April offer had they been eligible, which is a legitimate reason to widen the criteria and drop the age-plus-tenure test.

There is no target and no quota. Nobody is being selected.

The company is not shrinking in every direction either. It redeployed more than 4,000 people into new roles over the past year. In July it stood up Frontier Co, an engineering organization deploying 6,000 industry specialists to work alongside customers. Revenue divided by disclosed headcount works out to about $1.49 million per employee, up from roughly $1.24 million a year earlier, which is what a company looks like when it grows faster than it hires.

The exclusion list cuts against the harshest reading too. A company hollowing itself out to pay for servers would not protect sales and marketing. Those are the people who convert capex into revenue. Leaving them untouched is what you do when you are guarding the top line, not dismantling it.

Microsoft's spokesperson told GeekWire the offer is not a sign the company needs fewer employees, and that it is not a replacement for layoffs, which remain part of the toolkit.

The trouble with that account is not that it is false. It is that it is unfalsifiable. Every explanation on offer fits both a company re-mixing its workforce and a company beginning a long, quiet reduction, and Microsoft has declined to say which one this is. The Register asked directly whether AI was behind it. The answer was no. The same company's chief executive, six months earlier, said AI spending gets paid for out of operating expense. Both can be true. Only one of them shows up in the headcount line over the next eight quarters.

What Wednesday is actually testing

Microsoft reports fiscal first-quarter results after the close on Wednesday 28 October. Consensus sits near $4.71 in adjusted earnings on $90.63 billion of revenue, according to TipRanks. The numbers that matter for this story are elsewhere in the release.

Watch capital spending against the guided $50 billion. Watch whether the fiscal 2027 operating margin guide still holds at less than a point of decline. Watch free cash flow, which came in at $19.64 billion in the June quarter, down 23%, and which management has promised will stay positive. Watch the commercial remaining performance obligation, which reached $678 billion, up 84%, and which is the strongest single argument that the buildout has buyers.

The severance charge will not appear until the December quarter, because the departures happen on 31 December. The savings will not appear until the March quarter after that. What lands on Wednesday is forward guidance, and forward guidance is where a company tells you what it intends to do to its cost base.

There is a version of this where the exit offer is exactly what Microsoft says it is: a benefits program that learned from April, widened eligibility, and will produce a few thousand departures and a modest margin cushion. There is another where 27% of the company was offered the door two days before an earnings call at which management has to defend a $175 billion capital plan, and the offer is the first turn of a ratchet that becomes annual.

Both versions fit what is on the record today. The tell will be whether a similar offer appears in the first half of 2027, and whether the excluded organizations are still excluded when it does.

For anyone selling hardware into this buildout, the useful question is not whether the capex number goes up on Wednesday. It is which part of that number is underwritten by a 25-year depreciation assumption and which part is underwritten by a nine-week decision. Microsoft just showed that it will move its people faster than it will move its buildings. Suppliers modelling 2027 and 2028 capacity against a hyperscaler's capital plan are, in effect, lending against the 25-year side of that split while the customer manages the nine-week side.

At DMC we work with hardware companies whose order books are set by exactly this kind of decision, made in an earnings call and then paid for across seven years of lease commencements. If you are stress-testing a 2027 capacity plan against a customer's capex guide, let's talk.