Microsoft's $50B FY2025 capex absorbed most of operating cash. The Research Desk argues the value-creation arithmetic justifies the spend, with Azure compounding revenue at 33% and operating margin at 47%.
Microsoft's FY2025 capital expenditure ran at approximately $50 billion, against operating cash flow of $122 billion and free cash flow of $71.6 billion. The capex absorbed nearly half of operating cash, which is a pace not seen at Microsoft outside of two prior generation-defining buildouts: the original Windows datacentre footprint of the late 1990s and the initial Azure infrastructure investment of 2014-2017. The current cycle is the third generation, focused on AI compute capacity for Azure OpenAI, Co-Pilot, and the broader enterprise-AI service set.
The headline number triggers two predictable bear concerns. The first is that the capex spend will not generate adequate ROI. The second is that the FCF compression in the buildout phase damages the multiple. The Research Desk view, which we develop in detail below, is that both concerns are overstated and that the FY2026-FY2027 cash flow inflection from the maturing capex cohort will materially exceed consensus.
The core arithmetic. Azure's revenue grew at approximately 33% in FY2025, contributing $130 billion to the consolidated revenue line. Capacity utilisation has been running at near 90%, meaning the constraint on Azure growth is supply, not demand. Each $10 billion of capex deployed adds roughly $4-5 billion of incremental revenue capacity at gross margins above 60%. The economic return on the capex programme is therefore in the 25-35% IRR range, comfortably above the 9.5% cost of capital we use for high-quality enterprise software platforms.
This article works through the capex-cycle analysis in five sections: the capacity-utilisation dynamics, the AI-revenue attribution, the Co-Pilot economics, the competitive-position read versus AWS and Google Cloud, and the cash-flow-inflection forecast through FY2028. Microsoft target: $510 within 12 months. We are buyers at current levels.
Microsoft's capex-to-revenue ratio has moved through three distinct regimes over the last 20 years. Pre-2014, the ratio averaged 8-10% of revenue, reflecting steady investment in datacentre footprint and product-development capacity. From 2014 to 2020, as Azure scaled, the ratio climbed to 14-16%. From FY2023 onward, the AI-driven capex cycle has lifted the ratio toward 18-19% of revenue.
The FY2025 capex of $50 billion represents 17.7% of revenue. That is the highest absolute ratio Microsoft has produced since the pre-Y2K datacentre buildout cycle, but it is not without precedent. The ratio compares to Amazon's capex-to-revenue of approximately 18% in FY2025 (driven by AWS and the fulfilment-network buildout) and Google's of roughly 15% (driven by TPU and datacentre infrastructure for Gemini and the broader AI-product set). Microsoft is therefore not running an outlier capex programme; it is running a capex programme commensurate with the others in the hyperscaler peer set.
The more interesting historical anchor is Microsoft's own FY2014-FY2017 capex cycle. During that earlier Azure buildout, capex ran at 13-15% of revenue, and FCF compressed by roughly 8-10% during the heaviest investment phase. By FY2018-FY2020, as the capacity matured, FCF inflected upward by 60% over a three-year window. The current capex cycle has a similar shape but on a larger scale. The implication is a similar FCF inflection through FY2027-FY2028 if the Azure revenue trajectory holds.
The consensus model has not yet fully absorbed this. FY2026 FCF consensus sits at approximately $80 billion, with FY2027 at $95 billion. Our model points to FY2026 at $84 billion (slightly above consensus, driven by capex moderation) and FY2027 at $108 billion (well above consensus, driven by both capex moderation and the Azure revenue compounding). The gap to consensus is the cleanest single source of FY2027 upside in the multi-year model.
TickerXray Report
Get the complete Microsoft report with all 12 quantitative models, AI-generated investment thesis, and real-time data.
Azure's revenue growth rate of 33% in FY2025 came in despite multiple disclosed quarters of capacity-supply constraints. Senior Microsoft executives, including Satya Nadella and Amy Hood, have repeatedly indicated on earnings calls that customer-demand orders for Azure AI services exceed available capacity. The capex programme is sized to close that gap rather than to push demand-creating supply.
The capacity-supply constraint is the cleanest single signal that the AI-driven Azure compounding is operationally durable. Demand-creating supply scenarios (where capacity is built ahead of demand and the demand follows) are riskier; supply-constrained demand scenarios (where demand exceeds capacity and capacity is built to match) are lower-risk because the revenue is contractually visible at the time of capacity commitment.
Azure's customer cohort that signed multi-year capacity commitments through FY2025 represents roughly $86 billion of contracted revenue runway, captured on the balance sheet's deferred revenue and remaining-performance-obligation lines. That contracted runway grows roughly $10-12 billion per quarter on net new bookings. Each capacity-completion milestone unlocks a slice of that contracted revenue, which is the engine of the FY2026-FY2027 revenue compounding.
The buyer mix on Azure AI workloads has shifted noticeably over the past four quarters. Enterprise-IT-led workloads (financial services, healthcare, retail) now contribute approximately 55% of new Azure AI commitments, up from 35% two years ago. The remaining mix is split between AI-native customers (model training, foundation-model providers) and government / public-sector. The diversification is favourable for the structural durability of the revenue line, because enterprise IT contracts have multi-year commitment structures whereas AI-native customer contracts can be more variable.
Microsoft's AI-related revenue stack across the Productivity-and-Business-Processes segment and the Intelligent-Cloud segment has emerged as a meaningful contributor in FY2025. Co-Pilot for Microsoft 365, priced at $30 per user per month for the standard tier, has scaled to an estimated 12-15 million paid seats by end of FY2025, contributing approximately $4-5 billion of annualised revenue. The seat-attachment rate within enterprise-365 customers continues to climb, with some Fortune 100 deployments now running at 60-70% seat coverage.
GitHub Co-Pilot has scaled separately. Paid seat count exceeded 5 million at end of FY2025, with the enterprise tier at $39 per user per month and the individual tier at $10. Annualised GitHub Co-Pilot revenue sits at roughly $1.5-1.8 billion. Adoption of GitHub Co-Pilot Workspace, the more advanced agent-style product launched in mid-FY2025, has been tracking ahead of pre-launch expectations.
Azure OpenAI service revenue, captured separately within the Azure consumption line, contributed roughly $7-9 billion to the FY2025 Azure revenue figure. The growth rate on this line is materially faster than the broader Azure rate, with sequential quarter-on-quarter growth exceeding 15-20% through FY2025. Foundation-model API consumption has been the largest individual contributor.
Layered together, the addressable AI-revenue stack contributing to Microsoft's FY2025 income statement runs at approximately $15-18 billion annualised. The Research Desk model points to that figure clearing $35 billion by FY2027, supported by Co-Pilot seat expansion, Azure OpenAI consumption growth, and the introduction of new AI-product variants across Power Platform and Dynamics. That trajectory is the single largest contributor to Microsoft's revenue compounding through FY2028.
The cloud-infrastructure market continues to expand at roughly 22-25% annually, supported by enterprise digital-transformation spend, AI-workload migration, and the broader application-modernisation tailwind. The market-share dynamics across the three hyperscalers tell a more nuanced story than the headlines suggest.
AWS remains the largest hyperscaler by revenue, contributing approximately $115 billion of run-rate revenue at end of FY2025 (calendar Q4 2025). The growth rate has slowed from peak levels but stabilised near 17-19% in recent quarters. AWS continues to dominate the legacy-application-migration and the steady-state enterprise-IT cloud-consumption market. The franchise is the most mature and the most diversified.
Microsoft's Azure ran at approximately $130 billion of FY2025 revenue (including non-cloud Intelligent Cloud components), with the pure-cloud subset at roughly $95 billion. The growth rate of 33% on the broader Azure line and 30%+ on the pure-cloud subset is the fastest among the three hyperscalers. The AI-workload mix at Azure has been the highest of the three, supported by the OpenAI partnership and the deeper enterprise-software integration.
Google Cloud has narrowed the gap meaningfully. FY2025 revenue ran at approximately $58 billion at growth rates of 30%+. The enterprise-customer base has expanded materially, supported by Gemini-driven AI capabilities, BigQuery's analytic platform, and the increasing TPU adoption. Google Cloud is the cleanest beneficiary of an AI-first hyperscaler choice for net-new workloads.
The net market-share read is that Azure has been the share-gainer over the past four quarters relative to AWS, with Google Cloud also gaining at a slightly slower pace. The driver is the AI-workload mix shift. Microsoft's competitive position within the cloud-infrastructure market is structurally favourable for the next 24-36 months, supporting the revenue trajectory underlying our model.
Three engines drive the Microsoft growth trajectory through FY2028. The first is Azure consumption compounding, driven by AI-workload demand and the contracted-runway dynamics described above. We model Azure revenue growth at 30-32% in FY2026, decelerating modestly to 25-28% in FY2027 as the law of large numbers sets in, and 22-25% in FY2028. The compounding effect produces $260+ billion of Azure revenue by FY2028.
The second engine is Co-Pilot and the broader AI-product set. Seat expansion continues at 25-30% annually as enterprise customers move from pilot deployments to organisation-wide rollouts. The pricing power on Co-Pilot remains strong; the FY2025 introduction of Co-Pilot Pro and the Co-Pilot Studio extension has produced incremental revenue per seat without seat-count attrition. Total AI-related revenue should clear $50 billion by FY2028.
The third engine is the productivity and business-processes segment continuing its mid-teens growth pace. Office 365, Dynamics, and LinkedIn each contribute steady high-single-digit to mid-teens growth. Business processes (Dynamics 365, Power Platform) have been the surprise upside contributor, supported by AI-feature integration that lifts pricing power. The segment as a whole runs at 14-16% growth through FY2028.
Layered together, the consolidated revenue compounding rate sits at 13-15% annually through FY2028, against the FY2025 base of $282 billion. By FY2028, revenue should clear $410 billion. EPS growth, supported by operating margin holding at 47-48% and the buyback contribution, lands at 14-16% annually. That is best-in-class compounding for any large-cap technology franchise at this scale.
Three risks would force the Research Desk to reduce conviction. First, an Azure capacity-utilisation deceleration where new bookings slow below 15% on a sequential basis. The current data shows the opposite trend, with sequential bookings accelerating, but the metric requires monitoring.
Second, an AI-related capex compression where the broader hyperscaler cluster decides that incremental compute investment has reached diminishing returns. Microsoft's FY2026 capex guidance, when it is provided at the Q2 FY2026 call, will be the clearest signal. A meaningful step-down from the $50 billion FY2025 base would imply either operational discipline (positive) or demand deceleration (negative). The interpretation matters more than the number.
Third, a major regulatory action that compresses Microsoft's competitive moat, particularly within the Office 365 enterprise productivity bundle. Antitrust scrutiny in Europe and the UK has been a persistent feature of the regulatory landscape, with various inquiries reaching different stages. None has produced a structural-revenue-impacting outcome to date, but the risk is real.
Microsoft is the highest-quality compounding franchise in the global large-cap technology cluster, with the cleanest combination of revenue growth, operating margin, and capital allocation discipline. The capex programme is value-accretive, the FCF compression is cyclical rather than structural, and the FY2026-FY2028 inflection is well-supported by the operational data.
Fair value, anchored on a 25x forward multiple applied to FY2027 EPS of $20.50, lands at $510. Today's $425 print offers 20% upside on a 12-month view. The bull case to $560 requires Azure revenue growth to hold above 30% through FY2027. The bear case to $370 requires the FY2026 capex line to compress below $40 billion, which would imply demand deceleration rather than discipline.
The trade is to own the franchise here. The compounding has not finished. The capex cycle is the cost of the next compounding wave, not the end of the current one. Microsoft remains the cleanest expression of large-cap technology value creation. Constructive.
For portfolio managers running benchmark-aware mandates, Microsoft is one of the few large-cap names where the active overweight conviction holds across multiple cycle scenarios. The franchise compounds in a deflationary technology cycle (operating-leverage on software-and-cloud), in an inflationary capex cycle (capacity-tightness pricing power on Azure), and in a recessionary scenario (sticky-recurring-revenue base across the enterprise customer set). That cycle-resilience is rare at this scale.
The pattern across decade-scale technology compounding cycles is consistent. The franchise that aligns its capex programme to the dominant compute paradigm of the next decade compounds value fastest. Microsoft has done this with Azure-AI-Co-Pilot in a way that the operational data confirms quarter after quarter. We are constructive at $425 and remain so through $470. Trim only above $530.
Full forensic analysis of Microsoft
+ 6 more models included
150,000+ stocks covered
Global coverage across 60+ exchanges. Every report includes all 12 quantitative models and AI analysis.
View plansEvery report runs 12 quantitative models and generates an AI investment thesis. From Piotroski scores to manipulation detection -- get the full picture in seconds.
12 forensic models
Piotroski, Altman, Beneish, DuPont & more
AI investment thesis
Synthesized outlook on every stock
Manipulation detection
Spot red flags before they hit the news
150,000+ tickers
Global coverage across 60+ exchanges
Expected return
Forward return projections for every stock
Real-time data
Live prices, insider trades, news sentiment
Free accounts get 1 report per month. Pro gets unlimited.
FY25 capex hit $64.6 billion, up 45% year over year. Operating income hit $128.5 billion, up 17%. The capital efficiency on AI infrastructure spend is the bull case.
Microsoft has never offered a voluntary employee buyout. It just did. The Capital Desk reads the signal as capex prioritisation, not cost distress.
The FY2026 capital expenditure guide has shifted higher than our prior model. Operating margin has expanded faster than our prior model. The net of the two is a revised fair value range.