Microsoft FY26 Q4: Copilot Momentum and Capex Optics

Microsoft’s FY26 Q4 earnings, reported July 29, 2026, delivered $90.0 billion in quarterly revenue, up 18% year over year, with Azure crossing $100 billion in annual revenue for the first time and Microsoft 365 Copilot passing 30 million paid seats. By most measures it was the strongest print of the AI-capex earnings season so far — and the stock rose roughly 7–8% in after-hours and next-day trading, per multiple outlets.

But the recap cycle around this quarter carries two numbers Microsoft never said on the call. A “$37 billion Copilot AI ARR, up 123%” headline that actually dates to the April Q3 disclosure, and a “$255–260 billion FY27 capex guidance” that traces to a pre-earnings analyst preview, not to anything CFO Amy Hood guided. Both are already circulating in coverage as if they were July 29 disclosures.

This post does three things: reconciles what Microsoft actually said against what got repeated, runs the Copilot seat math that most recaps skipped, and explains — in plain English — why Microsoft’s reported capex numbers now look smaller even though the company says it is not spending a dollar less. Every figure is sourced to the official press release, the earnings-call transcript, or a named secondary source.

The headline totals, per Microsoft’s press release: Q4 FY26 revenue of $90.0 billion, up 18% year over year (+17% in constant currency); GAAP net income of $35.8 billion, up 31%; diluted EPS of $4.81 GAAP (+32%) and $4.74 non-GAAP (+23% adjusted for the OpenAI investment impact). The full fiscal year, ended June 30, 2026, closed at $331.8 billion in revenue, up 18%, with operating income of $155.2 billion (+21%) and net income of $133.7 billion (+31% GAAP).

The cloud engine did the pulling. Azure and other cloud services revenue grew 43% year over year in the quarter (the exact rate came on the call; the written release states the milestone), and Satya Nadella opened with the marquee line: Azure surpassed $100 billion in annual revenue, up 41% for the full fiscal year. Microsoft Cloud revenue was $59.3 billion in Q4, up 27% year over year, and passed $214 billion for the full year — with, per Hood, nearly 90% of it coming from customers outside of frontier model companies.

By segment: Productivity and Business Processes at $37.8 billion (+14% YoY), Intelligent Cloud at $39.3 billion (+32%, +31% constant currency), More Personal Computing at $12.9 billion (−4% YoY). Commercial remaining performance obligation — contracted future revenue — grew 84% year over year to $678 billion; excluding OpenAI commitments, RPO still grew 25%. A $3.2 billion gain on Microsoft’s investment in Anthropic contributed to the quarter, one of several discrete items worth 27 cents of EPS. Cash flow from operations was $55.4 billion, up 30% YoY, while free cash flow came in at $19.6 billion — down year over year, reflecting the capex ramp covered in section 05. Microsoft returned $10.2 billion to shareholders in the quarter and over $43 billion for the full year; total headcount declined 2% year over year.

The print landed in the middle of a heavy earnings week: Meta reported the same day with its own capex story spooking the street, a week after Alphabet’s Q2 print raised its capex guidance. Microsoft was the one of the three rewarded for spending more — roughly +7–8% after hours and into the next session, per TradingKey and other outlets (individual reports vary within that range).

Two numbers dominate the recap coverage of this quarter, and neither appears in the July 29 call. We checked the full official transcript end to end for both before writing this post — the same discipline we applied to OpenAI’s 10-million-agent-users claim, and it pays off again here.

The $37 billion “Copilot ARR.” Microsoft’s AI business crossing a $37 billion annualized revenue run rate, up 123% year over year, is real — but it is the company’s Q3 FY26 disclosure from the April 29, 2026 call, for the quarter ended March 31, 2026, as CIO Dive reported at the time. On the Q4 call, Microsoft did not restate a consolidated AI-ARR figure at all. The only “$37 billion” in the Q4 transcript is an unrelated coincidence: Hood’s Q1 FY27 revenue guide for the Productivity and Business Processes segment ($36.7–37 billion). Several recap outlets are nonetheless carrying the April number in July copy.

The $255–260 billion FY27 capex “guidance.” That range comes from a pre-earnings analyst preview — published before July 29 as an expectation. What Hood actually guided: FY27 capital expenditures “will grow year-over-year given demand signals across our portfolio,” calendar-year 2026 capex of approximately $175 billion, and Q1 FY27 capex of over $50 billion. No FY27 total dollar range was disclosed on the call.

Sources: official Microsoft FY26 Q4 earnings-call transcript (July 29, 2026) and press release; the $37B / +123% attribution to Q3 FY26 per CIO Dive (April 30, 2026); the $255–260B provenance per TradingKey’s pre-earnings preview. Confirmed rows quote the transcript directly.

The genuine Q4 Copilot headline is the seat count. Microsoft 365 Copilot passed 30 million paid seats in the quarter, and the sequential shape matters more than the level: net paid-seat adds more than doubled quarter over quarter. Against the roughly 20 million paid seats disclosed for Q3 FY26 (quarter ended March 31, 2026, per CIO Dive’s coverage of the April call), that is at least 50% seat growth in a single quarter — computed on the ~20M Q3 base.

Now the denominator most recaps skip. Microsoft-ecosystem trade analysis puts the total Microsoft 365 commercial seat base around 464 million — which makes 30 million paid Copilot seats roughly 6.5% penetration of the installed base, even after a doubling quarter. Read one way, that is a slow burn roughly three years after the product’s March 2023 announcement. Read the other way — the way the rest of this print supports — it is an enormous runway attached to an installed base Microsoft already owns, at the moment the depth-of-deployment signals turned up: the number of enterprise customers deploying Copilot to the majority of their information workers grew nearly 75% quarter over quarter.

The flagship deployment Nadella cited makes the depth story concrete: NHS England is rolling Copilot out to 505,000 clinicians and staff — described on the call as the largest healthcare deployment of its kind — after a trial found it saved staff an average of 43 minutes per day. And capacity is being built ahead of that demand: Microsoft added 31 new data centers across five continents in Q4 alone (88 for the full year), added another gigawatt of capacity in the quarter, and says it remains on track to roughly double overall capacity in two years. Monetization is following capacity, not the other way around. For what those seats actually do once deployed, see our breakdown of Copilot Cowork’s enterprise agent workflows.

Here is the two-quarter view that single-quarter recaps miss — and the thing that makes the unrefreshed ARR figure interesting rather than merely pedantic. Set the April disclosures next to the July ones and the pattern is stark: Microsoft stopped stating the consolidated AI run-rate number precisely in the quarter when every proxy for it accelerated.

Q3 column per CIO Dive’s coverage of the April 29, 2026 call; Q4 column per the official July 29, 2026 transcript. Derived cell: seat growth of at least +50% QoQ computed as (30M − 20M) ÷ 20M on the disclosed bases; all other change cells restate call-disclosed rates with their comparison periods named.

The revenue line kept pace with the seats. Nadella’s formulation on the call: “Copilot revenue accelerated over 60% quarter-over-quarter” — a rate that follows GitHub Copilot’s June 2026 shift to usage-based billing, so part of the acceleration is monetization-model change rather than pure volume. The surrounding usage stats came thick: GitHub Copilot reached 50 million users out of GitHub’s 225 million total (with no prior comparable stated on this call, so no growth rate can honestly be attached to it); over 90% of the Fortune 500 use GitHub; and, per Nadella, “one in three pull requests on GitHub now involves an agent.”

The governance-side numbers matter just as much for anyone deploying this at enterprise scale: Microsoft Purview has now audited over 50 billion Copilot interactions for compliance, up nearly 360% year over year, and Microsoft says it exposes over 650,000 MCP actions across sales, finance, supply chain, HR, and customer service so agents act with the same permissions and audit trails as human users. Foundry customers at a trillion-token annualized run rate grew 4x year over year. LinkedIn’s hiring tools added their own datapoint — recruiters at over 20,000 companies use them, with Recruiter seats up 140% quarter over quarter. Nadella also said Microsoft plans to bring its Copilot experiences, including Code, together in one “super app” this quarter — an on-the-record statement of intent, distinct from the earlier unconfirmed “unified Copilot app” rumor cycle. For how the largest early adopters govern this footprint, see our analysis of KPMG’s Agent 365 deployment.

The quarter’s most under-covered mechanic is an accounting change that makes Microsoft’s reported capital-expenditure numbers look smaller without the company spending less. Effective the start of FY27, Microsoft is extending the estimated useful life of its data centers and office buildings from 15 to 25 years — “reflecting our operating history and expected use of these assets,” in Hood’s words on the call.

The chain of consequences, as Hood laid it out: the change “affects only the timing of future depreciation and is expected to have a minimal benefit to FY27 operating income. The greater impact is on capital expenditures, as more of our future datacenter leases will shift from finance leases to operating leases” — and finance leases count toward reported capex while operating leases do not. Then the punchline: “Outside of this useful life impact, our calendar year 2026 CapEx investment expectations remain unchanged. However, the shift from finance to operating leases adjusts our expectation to approximately $175 billion.”

Decode that: the CY2026 capex figure fell roughly $15 billion against the prior ~$190 billion expectation purely from reclassification. The servers, chips, and buildings are all still being bought. Get the direction right — this change makes reported numbers look lower than underlying spend; it is not, as some coverage has implied, a driver of any increase.

The actual spending trajectory, for scale: Q4 FY26 capital expenditures were $41 billion, including the impact of higher component pricing, with roughly two-thirds of it going to short-lived assets — CPUs and GPUs. Full-year FY26 capex came to $115.9 billion. Q1 FY27 alone is guided at over $50 billion including the lease-reclassification impact — more than a quarter of FY26’s full-year total in a single quarter, before the rest of the FY27 ramp.

Strip the preview numbers out and Microsoft’s forward guidance on AI investment reduces to three statements. One qualitative: FY27 capital expenditures “will grow year-over-year given demand signals across our portfolio.” Two quantitative: calendar-year 2026 capex of approximately $175 billion (unchanged in substance, reclassified in presentation), and Q1 FY27 capex of over $50 billion. If a big round FY27 number is needed, the honest form is: expected to grow well past FY26’s $115.9 billion, with the specific total undisclosed — outside commentators have floated $255–260 billion, unconfirmed by Microsoft.

The interesting analytical question is why the disclosure pattern changed. In April, Microsoft volunteered a consolidated AI-ARR number with a growth rate attached. In July — with seats up 50%+, Copilot revenue accelerating over 60% QoQ, and RPO up 84% — it offered the proxies but not the refreshed total. Companies generally stop refreshing a metric either when it has become unflattering or when it is about to be reframed into new reporting lines. Nothing in this print suggests the former; the depth of segment-level AI commentary hints at the latter. Either way, the burden now sits with readers: the most-quoted “Microsoft AI” number in circulation has a April 29 date stamp on it, and treating it as current gets staler every quarter it goes unrefreshed.

Our forward read: expect the FY27 capex narrative to be structurally noisier than FY26’s, because the lease reclassification permanently decouples reported capex from real infrastructure spend. Analysts who track “capex as disclosed” will systematically understate Microsoft’s AI buildout relative to peers that keep finance-lease-heavy structures — which may make Microsoft’s spending look more disciplined than it is in any cross-company comparison. If the AI trade turns on capex anxiety, as it briefly did for Meta this same week, Microsoft has quietly bought itself better optics for the identical underlying behavior.

Four practical readings, depending on which seat you occupy. The common thread: this print rewards teams that treat vendor numbers as dated disclosures and platform roadmaps as swappable components — a stance Nadella himself endorsed on the call when asked about model dependency: “We are very, very clear about the architectural design of the platform, which is you’ve got to keep your harness separate from the model... that means any given model at any given time is swappable.”

The build-or-buy question underneath all of this — when 30 million seats of horizontal Copilot make sense versus a purpose-built agent system on your own data — is one we work through in detail in our agentic CRM buy-vs-build framework. And if this quarter is the prompt to figure out where agents actually fit in your operation — which workflows, which model, which governance — that scoping exercise is exactly what our AI transformation engagements are built for: an audit of where agentic capability pays back first, before any platform commitment.