Business
Know the Business
Microsoft is two businesses stapled together: a software-economics seat franchise (M365, LinkedIn, Dynamics, Windows commercial) that throws off ~$70B of segment operating profit at 58% margins, and a consumption-cloud-plus-AI franchise (Azure, server, OpenAI partnership) currently absorbing the largest enterprise-software capex regime change in history — $64.6B of capex in FY2025, up from $20.6B in FY2021. The seat business is what an investor is actually buying; the cloud build is what the market is currently arguing about, because that capex is converting Microsoft from a 33%-FCF-margin compounder into something closer to a 25%-FCF-margin infrastructure operator while D&A more than doubles. The most likely error today is treating the headline P/E as if the underlying franchise still earns the old returns on capital — ROIC has compressed from 37% in FY2021 to 24% in FY2025 even as operating margins expanded.
Revenue FY25 ($M)
Operating Income ($M)
Free Cash Flow ($M)
Capex ($M)
Operating Margin
FCF Margin
ROIC
Share Price (May 15)
1. How This Business Actually Works
Microsoft sells two units of value: a paid commercial seat and a billable cloud consumption hour. Each generates revenue on a different clock and at a different margin, and almost everything else is a feeder for one of the two.
The commercial seat is the engine. A Microsoft 365 E5 enterprise seat lists around $57/user/month, renews three years at a time under Enterprise Agreements, and once provisioned costs Microsoft almost nothing to deliver one more month of it — gross margin sits in the high-80s. Multiply ~415M+ paid M365 commercial seats (Q2 FY26 disclosure, growing ~6% YoY) by mid-single-digit ARPU growth (driven by Copilot, security, compliance upsell) and the result is a stable, predictable, software-economics annuity. LinkedIn (+12% Q3 FY26 GAAP / +9% cc), Dynamics 365 (+22% / +17% cc), and the cloud portion of M365 all sit in this layer. The Productivity & Business Processes segment earns ~$70B of operating income on 58% margins.
Cloud consumption is the second engine and the one re-rating the business. Azure bills compute hours, storage, network egress, and AI tokens. Gross margins are 60-70% at scale and structurally lower than software seats, because real silicon (Nvidia GPUs especially), power, cooling, and DRAM depreciate against every dollar of revenue. As AI inference mix has grown, Microsoft Cloud's blended gross margin has slipped to ~69% — the first material compression in years. Azure grew 40% in Q3 FY26 and the AI revenue run-rate hit $37B, +123% YoY. Commercial RPO — the contracted backlog — sits at $627B, up 99% YoY, the strongest forward visibility any large-cap software franchise has ever disclosed.
Where the incremental dollar of profit comes from: an additional Microsoft 365 Copilot add-on (~$30/user/month) at 80%+ contribution beats almost any other dollar Microsoft can capture, which is why the company has bundled Copilot aggressively into its E3/E5 stack and why management talks about "agentic" workloads — every new agent is a new SKU on an existing seat. The cloud capex build is, in part, a defensive move to make sure those AI-attached seat upgrades can actually run.
Mental model: P&BP is the bond (steady annuity, software margins). Intelligent Cloud is the call option (growth engine, lower margin, eating most of the capex). MPC is the residual (low-growth, mostly Windows OEM and Gaming). At ~$70B of segment operating profit, P&BP alone would value at $1.4-1.7T at peer software multiples; the rest of the market cap is the cloud bet.
Bargaining power. Microsoft is the system of record for almost every Fortune 500 IT department — Active Directory, Outlook, Excel, Teams, and increasingly Azure all sit on the same identity layer. Switching costs are not measured in months; they are measured in years and tens of millions of dollars of services work. The only real countervailing forces are (a) Nvidia, which captures the largest single slice of every new AI dollar, (b) regulators (EU DMA already forced Teams to unbundle in the EEA), and (c) AWS and Google Cloud, which remain credible substitutes at the IaaS layer.
2. The Playing Field
At this scale there are only five real comparables, and each one isolates a different question about Microsoft's economics. GOOGL is the only company that competes across cloud, productivity, AI models, and ads simultaneously. AMZN owns the only larger cloud (AWS). ORCL is the cleanest enterprise-software pure-play. META is the AI-capex comparator. AAPL is the platform-economics comparator.
All figures from latest reported fiscal year. Microsoft FY ends June; Apple FY ends Sep; Oracle FY ends May; others Dec.
Apple sits in the top-left corner because it spends ~3% of revenue on capex; the hyperscalers cluster on the right because they own the AI build. Microsoft sits in the middle of the build — high capex, but ROIC still meaningfully above Oracle and Amazon. Meta is the cleanest read-across: similar capex intensity, similar ROIC, very different valuation.
What the table reveals:
- Microsoft has the best margin/cash-flow combination among the cloud trio. 45.6% operating margin and 25.4% FCF margin both beat GOOGL (32% / 18%) and AMZN (11% / 1%). Only Meta — not in cloud — matches it.
- Microsoft trades at a higher multiple than peers with the same or better economics. P/E 36.5 vs META 28.1 and GOOGL 29.0. Some of that is mix (P&BP is genuinely scarcer than ads), some is the perceived AI-leadership premium via OpenAI.
- ROIC has compressed but Microsoft is still the second-best capital allocator in the group after Apple — and Apple's ~48% ROIC is partly a buyback artifact (a small equity base flatters the ratio). On a fairer pre-leverage basis Microsoft and Meta are the cleanest compounders.
- Oracle is the cautionary tale. Same enterprise-software DNA as Microsoft, much smaller cloud, higher capex intensity (37%), and FCF has gone negative as it chases hyperscaler-class AI capex without hyperscaler-class scale. Microsoft is the only enterprise-software incumbent that has actually crossed into hyperscale and is funding it from cash flow.
What "good" looks like in this peer set: sustained high-teens or better revenue growth, operating margin in the 35-45% band, FCF margin in the 20-30% band, ROIC above 20%, capex/revenue eventually settling below 25% as AI infrastructure depreciates. Microsoft is currently on all four — but capex intensity has overshot to 23% and the question is whether it normalises or keeps climbing.
3. Is This Business Cyclical?
The seat business is almost not cyclical. The cloud business is supply-cyclical today and demand-cyclical normally. Both are far less cyclical than they look.
History is the cleanest argument. Microsoft's revenue grew through 2008-2009 (FY09 dipped only 3%), grew through the 2015-2016 PC trough, grew through COVID, and grew through the 2022-2023 cloud "optimization" cycle. ~80%+ of commercial revenue is contracted or recurring, and Enterprise Agreements average 2.5-3 years of duration. What does move is the rate of growth — and the place to watch it is consumption-cloud growth and commercial bookings.
Revenue dipped only once in 18 years (FY09, -3%). Operating income compressed in FY15 (Nokia writedown) and FY18 (US tax reform), but operationally has compounded at ~13% over the period. Two consecutive years of revenue decline have never happened.
Where the cycle actually hits. When IT budgets tighten, the visible damage is not contract cancellations — it is Azure consumption growth decelerating as customers right-size their cloud spend, and bookings duration shortening as CIOs hesitate to commit three years out. The 2022-2023 mini-cycle compressed Azure growth from 50%+ to high-20s; revenue still grew 7%. There were no cancellations of M365 seats at any meaningful scale.
Today the cycle is inverted. The binding constraint is supply (GPUs, power, datacenter shells) not demand. Microsoft has said for six straight quarters that AI demand outruns supply. The risk is therefore not a normal recession; it is overbuilding — committing capex to a 5-6 year depreciation curve for AI gear at a moment when foundation-model architectures, custom silicon, and pricing are all moving. The commercial-seat side absorbs a normal recession easily; the AI cloud side could absorb a demand miss badly.
The cyclical question that actually matters: if Azure AI growth decelerates below ~25% before the $200B+ in cumulative FY24-FY26 capex finishes depreciating, FCF margins compress further (sub-20%) and ROIC follows. This is the bear thesis, and it is a real one. The seat business keeps paying the dividend regardless.
4. The Metrics That Actually Matter
Forget the headline P/E. The five numbers below explain ~80% of the variance in fair value for this company.
Operating margin still rising (mix toward cloud + scale), but capex intensity has doubled, FCF margin has fallen ~800bps, and ROIC has fallen ~1,400bps from peak. The four lines together are the AI-capex story in one chart: more profit, more capital, less return per dollar of capital.
Why these metrics matter more than the usual ratios. P/E and P/sales hide the capex regime change because they live above the depreciation line — D&A jumped from $14B in FY22 to $34B in FY25 and is set to keep climbing. EV/EBITDA compounds the same problem by excluding the very thing (the AI build) the multiple should be pricing. The honest valuation lens for Microsoft today is normalized FCF after a maintenance-level capex assumption — the only number that survives the difference between AI capex being a real asset or a depreciation overhang.
5. What Is This Business Worth?
Value here is determined by one question: what is the normalized free-cash-flow yield on $3.1T of market cap once the AI capex regime stabilizes? Everything else is decoration.
The right lens is earnings power × reinvestment runway, with a sum-of-parts cross-check because the seat franchise and the cloud build deserve different multiples. SOTP is not the right primary lens — there are no listed subsidiaries, no holdco structure, no separate currencies — but the segments are economically different enough that valuing them as one homogeneous "tech business" leaves real information on the table.
Segment value cross-check. A defensible separation is to value P&BP at high-teens to low-20s EV/operating income (software-seat economics), Intelligent Cloud at high-20s to low-30s on near-term operating income but with explicit credit for the capex-supported growth runway, and More Personal Computing at high-single-digit operating-income multiples (consumer-cyclical, low growth).
Range sums roughly to $2.7T - $3.5T versus a current market cap of ~$3.1T. The number itself is not the point — the point is that the seat franchise alone covers most of the market cap, which means investors today are paying a thin premium for the AI optionality. Whether that is cheap or expensive depends entirely on the Azure decel question.
Bottom line on valuation. Microsoft looks "cheap" relative to its own quality only if Azure stays above 25% growth for three more years and capex intensity normalizes below 18% by FY28. It looks expensive on every other path. The headline 36x P/E is misleading in both directions: it understates the seat franchise's defensibility and overstates the cloud business's near-term cash-conversion.
6. What I'd Tell a Young Analyst
Stop watching the P/E. Watch four things. First, Microsoft Cloud gross margin % — the moment it stabilizes or reaccretes, the AI capex bear case loses its central evidence. Second, Azure constant-currency growth — anything below 25% on a sustained basis is the leading indicator of capex overbuild. Third, capex/depreciation ratio — when it crosses back through 1.0x, the FCF margin starts climbing again. Fourth, Commercial RPO growth — the most underweighted disclosure in software; it loads multi-year AI commitments before they show up in revenue.
The market is probably underestimating two things. First, the quality of the seat franchise — Productivity & Business Processes alone runs at 58% operating margin with double-digit growth and would clear $1.4T+ of standalone enterprise value on a software multiple. Second, the defensiveness of the consumption-cloud business in a normal recession — the 2022-2023 mini-cycle showed Azure decelerates, it does not shrink, because the underlying workloads (identity, storage, line-of-business apps) are not discretionary.
The market is probably overestimating one thing. The simple "AI capex earns back as AI revenue" syllogism. Microsoft is committing capital with a 5-6 year useful life against an architectural landscape (custom silicon, frontier models, inference efficiency) that has moved faster than that in each of the last three years. The downside is not zero — it is a few years of FCF margin in the high teens while D&A catches up.
What would genuinely change the thesis. A clean Azure miss to sub-25% growth on demand (not supply) would. A sustained Microsoft Cloud gross margin below 65% would. A regulatory forced separation of the OpenAI economic interest would. A material erosion in M365 Commercial seat growth — currently mid-single-digit and accelerating with Copilot — would, because the entire valuation framework rests on the seat franchise carrying the cloud build.
Everything else — Windows OEM cycles, Xbox content swings, Surface launches, Bing share, even individual quarterly EPS beats — is noise. The signal is whether the largest enterprise-software capex experiment in history is paying back, and that question has a finite answer somewhere in the FY27-FY28 P&L.