Microsoft reported fiscal third-quarter results on the evening of April 29 and gave Wall Street the cleanest possible read of the AI thesis. Per the company’s Q3 FY26 press release and the CNBC writeup:
- Revenue: $82.89 billion, up 18% year-over-year (15% in constant currency).
- Operating income: $38.4 billion, up 20%.
- Diluted EPS: $4.27, up 23% on a GAAP basis.
- Azure and other cloud services: +40% year-over-year, ahead of the 38–39% Street consensus.
- Total AI annualized run rate: $37 billion, up 123% year-over-year — that line includes Azure AI, Copilot family, and revenue from model builders running on Microsoft infrastructure.
- Microsoft 365 Copilot: crossed 20 million paid seats, the first time the company has put a hard subscriber number on the productivity-side AI business.
- Q3 capital expenditure: $31.9 billion, up 49% year-over-year.
- Q4 capex guidance: stepping up to >$40 billion in a single quarter as more capacity comes online.
- FY26 full-year capex outlook: approximately $190 billion, up 61% year-over-year, with Amy Hood specifically calling out a $25 billion incremental impact from higher GPU and component prices.
That is the unambiguously bullish half of the call. The other half — the half that this site exists to read — is the operating-leverage sentence Hood put on the tape during forward guidance, lifted from the Motley Fool transcript and the Investing.com call summary:
“We continue to evolve how we operate to increase our pace and agility, and therefore, we expect headcount to decrease year over year.”
Total company headcount already declined year-over-year in Q3, per the same call. The future-tense version is the new disclosure.
How this stitches into the buyout we wrote about last week
Six days ago, on April 23, Microsoft opened the first voluntary retirement program in its 51-year history — roughly 8,750 U.S. employees inside a Rule-of-70 eligibility window, notifications dated May 7, 30-day acceptance, everything booked inside fiscal Q4 ending June 30.
At the time we wrote that piece, the framing question was: what’s the funding source. The answer in the press release sourcing was the AI capex line. The unanswered question was: what’s the operating-leverage story Microsoft is going to put behind it on the earnings call.
That story came in last night, and it is the cleanest version possible. Revenue +18%. AI ARR +123%. Headcount down YoY. Headcount guided down again next year. A board sitting in a Q3 strategy meeting at any other Fortune 500 company looking at that triple-stack now has a fully ratified template — Microsoft did this, the market accepted this, the financial press described this as a “clean beat,” and the only thing the stock dipped on was the upward revision of capex guidance, not the downward revision of headcount.
The 8,750 from last week is no longer the news. It is the floor. The guide is that number plus next year’s reductions on top.
The capex line is doing the rhetorical work
The line that Wall Street is actually watching is the capex one. Q3 came in at $31.9B — a tick below the $37.5B figure that was being cited in the buyout-week reporting, because finance-lease delivery timing slipped some hardware into Q4. The combined Q3 + guided-Q4 + FY26 outlook lands at about $190B for the fiscal year, up 61%. Two-thirds of that is short-lived assets — GPUs and CPUs that depreciate over 4–6 years. One-third is long-lived data-center infrastructure with a 15-year-plus depreciation runway.
The structural read for the headcount line: when 60–70% of incremental capital is going into 4–6-year-life GPUs that need to earn back their depreciation inside this 4-year window, the only way the operating margin holds is for the labor cost line to shrink in roughly the same proportion that the depreciation line grows. That’s not a corporate values statement; that’s an income-statement identity. Hood’s “headcount to decrease year over year” sentence is the income-statement identity made explicit.
What the 20-million Copilot number actually means
The single newest data point on the call was the 20 million paid Microsoft 365 Copilot seats figure. Microsoft has been notoriously coy about publishing a Copilot subscriber number for the past 18 months — the last hard disclosure we got was around 1 million paid seats in late 2024. The 20× growth in 18 months is the line that lets Hood guide capex up another $25B without spooking the Street.
It also matters for our beat in a quieter way. 20 million paid Copilot seats means 20 million seats inside enterprises whose CIOs have already signed off on AI in the productivity workflow. That installed base is the substrate the next 12 months of AI-attributed layoff press releases will run on top of. Every Q2 board meeting where a CHRO is asked “what’s the AI substitution path for this 200-person ops team,” the answer “we already pay $30/seat to Copilot for them, we just need to consolidate the work the agent already does” is now the easy answer.
The one bear point
The market did not love this print. Stock closed down ~4% in after-hours. The reason was the Q4 revenue guide of $86.7–87.8B, midpoint below the $87.5B Street consensus, combined with the capex step-up to >$40B per quarter. The bear read: AI revenue growth is real but is decelerating from the 123% rate at the same time capex is accelerating. That gap is the gross-margin compression risk every hyperscaler is running into right now.
The bull read: the gross-margin compression is the input to the headcount line going down. Trading lower-margin GPU revenue against lower labor cost is exactly what the buyout program funds.
Both reads are correct. The sentence “AI revenue is decelerating from 123% growth and headcount is going down to defend margin” is what the next year of this story actually looks like.
What LostJobs is watching
- The Q4 print and headcount disclosure on July 30. Microsoft typically discloses end-of-FY headcount in the 10-K filing two weeks later. The June 2025 number was ~228,000 globally and ~125,000 U.S. The June 2026 number is the one that confirms whether “headcount to decrease year over year” was a quarter-or-two thing or a multi-year reset.
- Whether other Q1-2026 mega-cap earnings calls copy the Hood guidance language. Alphabet reports tonight, Apple and Amazon later this week, Meta last week. If even one of the next three reports a similar “headcount to decrease YoY while AI revenue grows triple-digit” pairing, the sentence becomes the FY26 megacap consensus framing rather than a Microsoft-specific line.
- Whether the May 7 buyout-notification day produces leaks on uptake. The 30-day window closes around June 6. If uptake clears the ~5,000 mark out of 8,750 eligible, the “voluntary” program will have done most of the work the company needed without ever having to issue a layoff press release. That outcome is the Microsoft-pioneered template the rest of the Fortune 500 will copy in H2 2026.
The dry coda
The single most precise sentence in the entire Q3 release wasn’t on the income statement. It was Nadella’s positioning line, paraphrasing his prepared remarks: we are at the beginning of one of the most consequential platform shifts that will change the entire tech stack as agents proliferate and become the dominant workload.
Translate that out of investor-relations English: the workload that used to be a headcount line is becoming an Azure line, and we are the company that bills for both halves. The 20 million Copilot seats are the front-end of that translation. The Hood headcount guidance is the back-end. The 18% revenue beat is what it looks like when both halves work at the same time.