Microsoft Corp (MSFT)

Thesis verdict: holds, with conditions — AI demand has genuinely turned into a record backlog of long-term contracts, but massive on-balance-sheet capex and off-balance-sheet lease commitments are relentlessly eating cloud gross margin; monetization must outrun the rigid climb in depreciation and power costs.

Microsoft's business model is clear and enormous: it sells public-cloud compute infrastructure to enterprises and individuals (billed on actual consumption), office collaboration and productivity SaaS tools (subscriptions per seat, monthly or yearly), and personal computing devices and gaming content (outright hardware/software purchases or memberships). Money flows into Microsoft first from corporate IT budgets and consumers' pockets worldwide, and is then poured on a massive scale into global data centers, AI servers and power facilities, forming a huge productive capital base.

In the just-completed FY26, Microsoft's business breaks into three segments:

The first is Intelligent Cloud, 41.5% of revenue and 36.7% of operating income. More importantly, it delivered a full 62.9% of the company's year-over-year revenue increase for the year — unquestionably Microsoft's growth engine today. The core contracts attached to this segment are very heavy: commercial remaining performance obligations (RPO, simply put, the total value of contracts customers have signed but Microsoft has not yet delivered) formed by customers reserving future compute, plus Microsoft's own enormous data-center lease and capital commitments. The capex surge reflected in news such as "Over $50B in FY27 Q1 capex as Microsoft (NASDAQ: MSFT) reorganizes around AI" (2026-09-02), the gross-margin damage noted in the earnings MD&A, and the hard fact that RPO jumped 84% year over year all show this business sits in the eye of the AI build-out and power-resource storm.

The second is Productivity and Business Processes, 42.2% of revenue, 54.0% of operating income and 38.3% of the increase. This is the biggest bedrock of Microsoft's profit, covering enterprise software such as Office 365 and Teams; its driver is net revenue retention (NRR — how much more existing customers pay each year) and seat upgrades. Paid M365 Copilot seats have passed 30 million, proving that AI monetization at the software layer has taken a substantive step, even as competitors such as Salesforce's Agentforce close in.

The third is More Personal Computing, 16.3% of revenue but only 9.3% of operating income, with a −1.2% contribution to the increase — growth has stalled. The segment is constrained by hardware inventory turnover and device depreciation; the report shows Xbox hardware revenue fell 29% year over year, alongside "Watch Microsoft's Xbox to Cut 3,200 Jobs as Part of Reorganization" (2026-07-06), a large layoff of 3,200 staff — a shrinking business within the base.

Placing Microsoft in the layered value stack (the distribution of power across the software industry chain, from bottom-layer infrastructure to top-layer entry points):

• Layer five (gatekeeper): Microsoft Entra (formerly Azure AD) and Purview govern identity, access control and data compliance for enterprises worldwide — an underlying security privilege that is almost impossible to replace.

• Layer four (orchestration): M365 Copilot and Teams. These are not surface-level UI plug-ins but are embedded deep in enterprises' daily systems of record — email, weekly reports, cross-department meetings — with migration costs too high to bear.

• Layer three (core data assets): the Microsoft Graph and LinkedIn's professional network form a unique relationship-chain asset.

• Layer two (heavy-asset plant): Azure's underlying data centers. It must be noted that raw compute and storage carry a natural risk of commoditized price cuts and price wars; Microsoft's answer is to bind layer-two Azure tightly to layer-four Copilot and Teams, so that underlying compute does not become a utility with no pricing power.