Microsoft (MSFT) is a 50-year-old software company headquartered in Redmond, Washington that generated $281.7 billion in revenue in fiscal 2025 and is now operating at a $332 billion annualized revenue run-rate as of the quarter ended March 31, 2026. The company employs 228,000 people, holds $112 billion in cash and investments, has $50 billion in long-term debt, and just reported that its commercial backlog of contracted future revenue jumped to $627 billion, nearly double the figure from a year earlier. Operating cash flow is running near $170 billion annualized, and Microsoft is currently spending money on physical infrastructure at a pace that more closely resembles a power utility than a software company. The question this post takes on is whether the cash flows the existing software business generates are large enough to justify what management is doing with them.
What Microsoft Sells in 2026
The scope of Microsoft’s reach is best described by the market positions its products actually hold:
Operating systems. Microsoft Windows runs on 60.8% of all desktop and laptop computers worldwide as of March 2026, according to StatCounter data. macOS runs on 14.4%, desktop Linux on 3.2%, and ChromeOS on 1.6%. About 19.7% of traffic registers as “unknown” operating systems, which absorbs everything from privacy-blocking browsers to obscure distributions. Windows is installed on an estimated 1.4 billion active devices globally. In PC gaming specifically, where users tend to maintain newer hardware, Valve’s Steam Hardware Survey reports Windows at roughly 96% of gaming PC market share.
Productivity software. Microsoft has stated that 90% of the Fortune 500 are now using Microsoft 365 Copilot, and that daily active users of M365 Copilot increased 10x in the year leading up to early 2026. Microsoft 365 Consumer has 89 million subscribers. The commercial seat count is much larger and is sold per-seat under multi-year contracts.
Professional networking. LinkedIn is the only large-scale professional social network in the world. It is the dominant platform for corporate hiring, B2B marketing, and white-collar professional content.
Public cloud. Azure is the second-largest public cloud platform globally, behind Amazon Web Services. Microsoft does not disclose Azure revenue separately from the broader server and cloud line, but third-party estimates put Azure’s IaaS and PaaS market share at roughly 25%, with AWS at roughly 30% and Google Cloud at roughly 12%.
Developer infrastructure. GitHub hosts the majority of the world’s open-source code and is the default code repository for most enterprise software development.
Gaming. The Activision Blizzard catalog includes Call of Duty, World of Warcraft, Candy Crush, and Diablo. Microsoft’s first-party studios add Minecraft, Halo, Forza, and the rest of the Xbox Game Studios portfolio.
Healthcare AI. Nuance’s Dragon Medical software is used by most US physicians for clinical voice dictation, although Microsoft does not disclose specific share or user-count metrics.
Search. Bing operates the world’s second-largest general search engine by volume, but holds a small single-digit share of global consumer search traffic.
For SEC reporting, all of that surface area collapses into three segments:
| Segment | FY2025 Revenue | FY2025 Operating Income | Operating Margin |
|---|---|---|---|
| Productivity & Business Processes | $120.8B | $69.8B | 57.8% |
| Intelligent Cloud | $106.3B | $44.6B | 42.0% |
| More Personal Computing | $54.6B | $14.2B | 25.9% |
| Total | $281.7B | $128.5B | 45.6% |
The Productivity segment alone produces a higher operating profit than every public software company in the world other than Microsoft itself. Apple, Google, and Amazon are larger by market cap, but Microsoft has the most profitable software business that has ever existed. Understanding how it got there matters for understanding what the next decade looks like.
From Altair BASIC to a $282 Billion Revenue Company
Microsoft was founded in 1975 by Bill Gates and Paul Allen in Albuquerque, New Mexico. The two had been classmates at Lakeside School in Seattle and were fascinated by the January 1975 cover of Popular Electronics, which featured the Altair 8800, the first commercially successful microcomputer kit. Allen and Gates wrote a BASIC interpreter for the Altair and licensed it to MITS, the Albuquerque company that made the machine. They moved to Albuquerque to be near their first customer.
Two facts about that founding deal still shape Microsoft today. The first is that Gates and Allen retained the rights to the BASIC code rather than selling it outright. The second is that they licensed the same code to anyone else who wanted it. Selling software as a recurring license to multiple buyers, rather than as a custom build for one customer, was a structural decision that Microsoft has applied to almost every product it has ever sold.
In 1979, the company moved to Bellevue, Washington to be closer to Gates’ family in Seattle. The decision that changed everything came in 1980, when IBM was building the original PC and needed an operating system. Microsoft did not have one. Gates bought a system called QDOS from a small Seattle company for $50,000, modified it, and licensed it to IBM as PC-DOS. The deal Microsoft negotiated was non-exclusive, which meant Microsoft could sell the same operating system as MS-DOS to every IBM-compatible PC maker that emerged. By the end of the decade, virtually every personal computer in the world ran Microsoft software.
The 1980s built the franchise. Excel arrived in 1985. Windows 1.0 launched in November 1985. Microsoft went public on March 13, 1986 at $21 per share. Word for Windows came in 1989. The Office bundle, which combined Word, Excel, and PowerPoint into a single SKU sold at a discount versus buying them separately, launched in 1990 and became the dominant office productivity suite within five years.
The 1990s were the decade of Windows and Office. Windows 95 was released in August 1995 and sold seven million copies in the first six weeks. Internet Explorer was bundled with Windows starting that year, eventually triggering the United States v. Microsoft antitrust case that consumed the company from 1998 to 2001. The case did not break Microsoft up, but it forced changes in how the company bundled products and behaved toward competitors. It also marked the end of the Gates era as CEO. Steve Ballmer took over in January 2000.
The Ballmer years (2000-2014) are the part of Microsoft’s history that gets the most criticism. The company missed the smartphone (Windows Phone failed despite multiple relaunches), missed search (Bing was launched in 2009 and never seriously challenged Google), missed social (multiple attempts, all failed), and was slow to cloud computing. But the period also produced two products that became enormous: Xbox launched in 2001 and Azure launched in 2010, originally branded Windows Azure. Both took years to reach scale, but both became core businesses.
Satya Nadella became CEO in February 2014 and changed almost everything about how Microsoft operated externally. Office became Office 365 (now Microsoft 365), a subscription product instead of a perpetual license. Windows stopped being the strategic center and became one of many products. Microsoft made peace with Linux, contributed to open-source projects, and shipped its software on iOS and Android. Azure went from being a side project to being the most important product in the company. The acquisitions Nadella approved tell the strategy: Mojang (Minecraft) in 2014 for $2.5 billion, LinkedIn in 2016 for $26.2 billion, GitHub in 2018 for $7.5 billion, Nuance in 2021 for $19.7 billion, and Activision Blizzard in October 2023 for $75.4 billion, the largest acquisition in tech industry history.
The OpenAI relationship started small. Microsoft made a $1 billion investment in OpenAI in 2019, partly in cash and partly in Azure compute credits. A second investment of $10 billion was reported in early 2023, after the public release of ChatGPT in November 2022. Microsoft’s total funding commitment to OpenAI now stands at $13 billion.
Microsoft turned 50 years old in 2025. The company that started with two people writing a BASIC interpreter for a hobbyist computer kit now has 228,000 employees and ships products in 190 countries. Among megacap technology companies, Microsoft has been the most consistent at reinventing what its core products are while keeping the same customer relationships and cash-generating engine intact.
Inside the Three Segments
Microsoft’s revenue is now growing 18% year-over-year on a $300 billion-plus base, which is unusual at this scale. The drivers vary by segment.
Productivity & Business Processes ($120.8B in FY2025)
This segment is built on Microsoft 365 Commercial, the corporate version of the Office productivity suite, which generated $87.8 billion in revenue in fiscal 2025. The product is sold per-seat to corporate customers on multi-year contracts. Microsoft 365 Commercial seat count grew 6% in the most recent quarter, while average revenue per seat is rising as customers add Copilot, the AI assistant Microsoft built on top of OpenAI’s models and started selling in late 2023.
LinkedIn produced $17.8 billion in revenue in fiscal 2025, growing 12% in Q3 FY2026. The platform monetizes through Talent Solutions (recruiter subscriptions for hiring), Marketing Solutions (advertising), Premium Subscriptions, and Sales Navigator. LinkedIn is the only large-scale professional social network in the world, and it has been a steady high-margin contributor since the 2016 acquisition.
Dynamics, Microsoft’s ERP and CRM software, generated $7.8 billion in fiscal 2025 and grew 22% in Q3 FY2026. It competes directly with Salesforce, Oracle, and SAP in markets where those incumbents have been entrenched for decades. Dynamics is small relative to its competitors but growing faster.
Microsoft 365 Consumer (the personal version of Office, Outlook.com, OneDrive cloud storage, and the consumer Copilot product) added $7.4 billion in fiscal 2025 and grew 33% in Q3 FY2026. Consumer subscriber count reached 89 million.
Intelligent Cloud ($106.3B in FY2025)
This is Azure plus the on-premises server business plus Nuance plus enterprise services. Azure grew 34% in fiscal 2025 and accelerated to 40% in the quarter ended March 31, 2026. That acceleration is the single most important data point in the most recent earnings release. A multi-hundred-billion-dollar revenue line accelerating its growth rate is rare, and it happens here because of two reinforcing dynamics: enterprise customers migrating workloads from on-premises servers to the cloud, and AI workloads (both OpenAI traffic and Microsoft’s own AI services) running on Azure infrastructure.
The on-premises server business (Windows Server, SQL Server, System Center) is mature and growing slowly, but it produces high margins and serves as a hook into the Azure cloud migration sale. GitHub, Nuance, and Microsoft’s enterprise consulting services round out the segment.
Azure has roughly half the cloud market share of Amazon Web Services but is closing the gap. The acceleration to 40% growth, while AWS reported a slower growth rate, is the most concrete sign that the gap is narrowing. Microsoft’s structural advantage is its ability to bundle Azure with Microsoft 365 sales for the same enterprise IT buyer.
More Personal Computing ($54.6B in FY2025)
This is the part of Microsoft that looks the most like a traditional consumer technology business. Windows generates revenue from OEM licenses (the version of Windows that ships pre-installed on PCs), commercial licensing, and Surface device sales. Together these were $17.3 billion in fiscal 2025. This business is mature and PC-cycle dependent. In Q3 FY2026, Windows OEM and Devices revenue declined 2%.
Gaming generated $23.5 billion in fiscal 2025. Microsoft’s gaming business now combines Xbox hardware sales, first-party game studios (the existing Microsoft studios plus Activision Blizzard, ZeniMax, and Mojang), and the Xbox Game Pass subscription service. The Activision Blizzard acquisition closed in October 2023, and the integration is still ongoing. Q3 FY2026 was weak: Xbox content and services revenue declined 5%, and Xbox hardware revenue fell 33%. The gaming console cycle is at a low point, and Microsoft has explicitly guided that Game Pass and content sales are the long-term focus.
Bing and the search advertising business produced $13.9 billion in fiscal 2025, growing 12% in Q3 FY2026. Microsoft has used the OpenAI partnership to reposition Bing around AI search, but the underlying market position remains a distant second to Google.
The strategic question with this segment is whether Microsoft eventually carves it up or sells parts of it. Activision Blizzard was bought just three years ago, so a near-term divestiture is unlikely. But this segment generates only 11% of segment operating income on 19% of segment revenue, and a more focused enterprise software and cloud company would arguably trade at a higher multiple.
Reinventing the Commercial Engine
The structure of Microsoft’s sales force has been quietly redesigned twice during the Nadella era, and a third change happened in October 2025 that has been underweighted in most outside coverage of the company.
The first redesign was in 2017. Judson Althoff, who had joined Microsoft in 2013 from Oracle to run North America, was given responsibility for the global commercial business. He retrained roughly 10,000 of Microsoft’s salespeople and hired about 3,000 new technical employees to write code alongside customers. The big organizational change was splitting the salesforce by industry vertical (manufacturing, retail, financial services, education, government, healthcare) rather than just by geography. Within each vertical, Microsoft built first-party intellectual property tailored to industry workflows. The second change was the consolidation of the resulting structure into Microsoft Customer and Partner Solutions, or MCAPS, which became Microsoft’s primary growth engine and the unit through which all enterprise sales motion flowed.
The third change is the most consequential. On October 1, 2025, Satya Nadella promoted Althoff to “CEO of Microsoft’s Commercial Business,” consolidating sales, marketing (under CMO Takeshi Numoto), and operations under Althoff in a single reporting line. This is the first time Microsoft has used a “CEO of” title for a function that is not a separately incorporated business like LinkedIn or GitHub. In Nadella’s own announcement, the explicit purpose is to free Nadella and his engineering leaders to focus on “datacenter buildout, systems architecture, AI science, and product innovation.”
In context, the Althoff promotion is a clean signal of how Microsoft is splitting itself internally. Nadella is becoming a full-time chief technologist of an AI infrastructure and product company. Althoff is becoming the operating CEO of the rest. The four executive promotions Microsoft made under Althoff in early 2026 (Deb Cupp as Chief Revenue Officer for Global Enterprise, Nick Parker as Chief Business Officer for Worldwide Sales, Ralph Haupter as Chief Revenue Officer for Small and Medium Business, Mala Anand as Chief Customer Experience Officer) round out the new commercial leadership team.
Two things are worth flagging about this structure. First, the unification of marketing, sales, and operations under one leader is unusual at this scale. Most $300-billion-revenue companies have a CMO and a CRO who report independently to the CEO. Microsoft is putting them under one person who reports to Nadella. Second, this is the most direct organizational acknowledgment so far that the AI buildout is the company’s defining decade-long project, and that running the existing software business is now a separate full-time job from running the AI investment cycle.
What Microsoft Is Doing With Nuance
Microsoft acquired Nuance Communications in April 2021 for $19.7 billion, the second-largest acquisition the company had ever made at the time, behind only LinkedIn. Nuance was a Boston-area company best known for Dragon, the speech recognition software that runs on most personal computers, and for its dominant position in clinical voice dictation. Most of the doctors in the United States used Nuance’s Dragon Medical One for dictating notes into electronic health records.
The deal looked expensive when it closed. Nuance’s annual revenue at the time was roughly $1.4 billion, which made the purchase price more than 14x revenue. The strategic logic was that Microsoft was buying a foothold in healthcare, a sector that represents 17% of US GDP, with a customer relationship into nearly every large hospital system in the country.
Three years later, the strategic logic is starting to play out. In 2023, Microsoft launched DAX Copilot, an ambient AI tool that combined Nuance’s microphone tech with Microsoft’s Azure OpenAI service. DAX would listen to a doctor-patient conversation, draft a structured clinical note in real time, and file it directly into Epic, the dominant electronic health record system. By 2024, DAX was integrated into Epic Haiku, the mobile EHR app most US clinicians use, and was being deployed across major health systems including UNC Health, Vanderbilt, Mercy, and Stanford Health Care.
In March 2025, Microsoft announced Dragon Copilot at the HIMSS healthcare conference. Dragon Copilot consolidated Dragon Medical One (dictation), DAX Copilot (ambient listening), and generative AI into a single subscription product. In October 2025, Microsoft extended Dragon Copilot to nurses, with documentation workflows specifically designed for nursing care. The product is currently available in the United States, Canada, the United Kingdom, Austria, France, Germany, and Ireland, with more geographies announced.
The metric Microsoft cites in its own materials is that DAX Copilot saves clinicians five minutes per patient encounter and adds 13 appointment slots per provider per month. A 2024 internal Microsoft survey of 879 clinicians across 340 healthcare organizations using DAX Copilot reported a 50% reduction in documentation time and a 70% reduction in feelings of burnout. These are management-supplied numbers and should be read with appropriate skepticism, but the underlying customer adoption is substantial. Microsoft does not break out healthcare revenue separately in its segment reporting.
Whether the $19.7 billion was a good purchase is impossible to evaluate with precision from the outside without segment disclosure. What the SEC filings do tell you is that Nuance is now packaged inside the Intelligent Cloud segment, alongside Azure, GitHub, and the on-premises server business. The acquisition added roughly $9 billion of goodwill and $4 billion of intangible assets to Microsoft’s balance sheet at the time of closing, and there have been no impairments since. The strategic outcome being aimed at is for healthcare to become a vertical-specific Azure customer, with Dragon Copilot serving as the customer hook and the rest of Azure (data services, agents, AI) following on.
The pattern matches the rest of Microsoft’s M&A playbook. Buy a market-leading customer relationship in a specific vertical (LinkedIn for professional networking, GitHub for developer workflows, Activision for gaming content, Nuance for healthcare), then layer the entire Microsoft stack into that customer base. Of those four large acquisitions, LinkedIn and GitHub are clearly working. Activision is uncertain. Nuance is on a similar trajectory to the LinkedIn and GitHub model, although it is two to three years behind in commercial maturity.
Microsoft and OpenAI
Microsoft’s relationship with OpenAI is the largest strategic AI bet the company has made, although as the next section discusses, it is no longer the only one. The SEC filings now provide more detail on the OpenAI structure than they did a year ago.
The structure as currently disclosed: Microsoft holds an equity method investment in OpenAI Global LLC. Microsoft’s total funding commitment to OpenAI is $13 billion. Microsoft holds intellectual property rights to OpenAI models. OpenAI’s API for third-party developers is exclusive to Azure (developers cannot access GPT models through AWS or Google Cloud directly). Microsoft has a right of first refusal on OpenAI’s new compute capacity. There is a reciprocal revenue-sharing arrangement, the precise terms of which are not disclosed.
A clarification on the exclusivity. The “OpenAI runs only on Azure” framing is partially out of date. OpenAI’s third-party API for outside developers remains exclusively hosted on Azure, which is the most commercially significant piece of the relationship. But OpenAI itself has signed multi-billion-dollar compute commitments with Oracle and CoreWeave in 2024 and 2025, and the company’s first-party use of compute is no longer Microsoft-only. The Azure exclusivity that matters today is the part where outside enterprises building applications on top of GPT models do so on Azure infrastructure.
In fiscal 2025, Microsoft recognized $4.7 billion in losses in “Other income (expense), net” that reflected its share of OpenAI’s losses under equity method accounting. That number tells you OpenAI was burning cash at a high rate in 2025. In the first nine months of fiscal 2026, Microsoft recognized a $5.9 billion pre-tax gain ($0.60 of EPS after-tax) from what the filings describe as the “OpenAI Recapitalization.” This is a non-cash accounting gain reflecting the markup of Microsoft’s stake when OpenAI restructured its corporate form. In the third quarter of fiscal 2026 alone, Microsoft’s equity method losses from OpenAI shrank to $19 million, down from $583 million a year earlier. OpenAI’s economics are improving as its revenue scales.
Microsoft also extended its partnership with OpenAI in October 2025 and again in April 2026. The terms of those extensions have not been disclosed in detail, but the disclosure that they happened, combined with the structural exclusivity on the third-party API, suggests Microsoft has secured the relationship through whatever transition OpenAI is making toward independent commercial operation.
Microsoft’s “equity and other investments” line on the balance sheet rose from $15.4 billion at June 30, 2025 to $33.7 billion at March 31, 2026, a jump of $18.3 billion in nine months. Most of that increase reflects the markup of the OpenAI stake during the recapitalization. Secondary market valuations of OpenAI in early 2026 imply meaningful additional appreciation that has not yet been recognized on Microsoft’s books.
The skeptic’s view of the OpenAI partnership is that Microsoft has tied itself to a counterparty that is burning cash, has unclear governance, and may eventually try to renegotiate terms. The optimist’s view is that Microsoft has locked up the most valuable AI laboratory in the world for less than the cost of one of its other recent acquisitions, while making its core Azure cloud business strategically essential to the AI revolution.
Adding Anthropic to the Stack
If you read Microsoft’s AI strategy through 2024, OpenAI was the only meaningful bet. That changed on November 18, 2025, when Microsoft, Nvidia, and Anthropic announced a three-way strategic partnership at Microsoft’s Ignite conference. The headline numbers are large. Microsoft committed up to $5 billion of equity investment in Anthropic. Anthropic committed to purchase $30 billion of Azure compute capacity, with the option to scale up to one gigawatt of contracted capacity. Nvidia committed up to $10 billion of equity investment in Anthropic and a partnership to optimize Claude on Nvidia’s Grace Blackwell and Vera Rubin systems.
The structural meaning of this announcement is that Microsoft pivoted from a single-model AI strategy to a multi-model one. Anthropic’s Claude models (Sonnet 4.5, Opus 4.1, Haiku 4.5) are now available across Microsoft’s product surface:
- Microsoft Foundry, Azure’s developer-facing AI platform, where Anthropic is one of the model options enterprise developers can deploy via standard Azure billing and contracts.
- GitHub Copilot, where Microsoft had already added Anthropic and Google models alongside OpenAI’s in 2024.
- Microsoft 365 Copilot, where admins can enable Claude as an alternative model for individual queries. Default enablement began rolling out to commercial tenants in January 2026, with full geographic rollout (excluding EU, EFTA, UK, and government clouds) expected by March 2026.
- Copilot Studio, Microsoft’s tool for building enterprise AI agents.
In Satya Nadella’s own words from October 2025: “As AI becomes more capable and agentic, models themselves become more of a commodity.” That is a major position shift. A year earlier, OpenAI’s models were positioned as Microsoft’s distinctive AI moat. The November 2025 announcement reframes the moat as Azure infrastructure plus Microsoft 365 distribution, with the model layer now treated as a competitive battleground that Microsoft can hedge across providers.
The commercial implications. Anthropic’s $30 billion Azure compute commitment represents a guaranteed revenue stream for Azure infrastructure spread over multiple years. Microsoft’s $5 billion equity stake in Anthropic gives it economic exposure to Anthropic’s success, although without the IP rights and right of first refusal that Microsoft has on OpenAI. Anthropic is, importantly, the only frontier AI lab with active distribution on all three major clouds (Azure, AWS, and Google Cloud), so Microsoft is not locking Anthropic’s commercial relationships down the way it tried to with OpenAI.
In March 2026, Microsoft launched Copilot Cowork, a multi-step agentic workflow product within Microsoft 365 that was built explicitly using Anthropic’s technology. The product is priced at $30 per user per month and is the first marquee Microsoft 365 AI feature where Anthropic, not OpenAI, is the underlying model. Microsoft has also publicly disclosed using Anthropic’s Claude Code coding agent across its own internal engineering teams.
Beyond the OpenAI and Anthropic relationships, Microsoft has been quietly developing its own frontier models in-house. The MAI-1 model line, which Microsoft began testing in late 2025, is the first major attempt at proprietary frontier AI capacity not built on top of an external partner. That work signals what the long-term strategy probably is: keep the OpenAI relationship as the deepest commercial partnership, the Anthropic relationship as the credible alternative that prevents single-vendor lock-in, and Microsoft’s own MAI models as eventual insurance against both partners.
The clearest read on Microsoft’s AI strategy as of April 2026: OpenAI remains the largest single relationship and the only model with full IP rights flowing back to Microsoft. The Anthropic deal is the second pillar, structured to give Anthropic significant Azure consumption while preserving Anthropic’s independence. MAI-1 is in-house insurance. The bet underneath all three of those positions is that the model layer commoditizes faster than the infrastructure layer does, and that Microsoft owns more of the infrastructure layer than anyone else.
Building the AI Grid
The most consequential change happening at Microsoft right now is the rate at which the company is building physical infrastructure. The data is in the filings.
| Fiscal Year | Property & Equipment Additions | Year-End Net P&E |
|---|---|---|
| FY2022 | ~$24B | ~$74B |
| FY2023 | $28.1B | $95.6B |
| FY2024 | $44.5B | $135.6B |
| FY2025 | $64.6B | $205.0B |
| FY2026E (annualized from Q3) | ~$106B | ~$280B |
In four years, the property and equipment on Microsoft’s balance sheet will have nearly quadrupled. Most of this is data centers. The 10-K discloses $32.1 billion in construction commitments and another $92.7 billion in future operating and finance lease commitments that have not yet commenced. Add $110 billion in purchase commitments for data center equipment, and Microsoft has committed to roughly $235 billion in additional infrastructure spending beyond what is already on the balance sheet.
Three things matter about this CapEx cycle.
First, the spending is being funded out of operating cash flow, not borrowed. Microsoft generated $136.2 billion in cash from operations in fiscal 2025 and is on track for roughly $170 billion in fiscal 2026. The company has $112 billion in cash and investments and $50 billion in debt. Even at the elevated CapEx run-rate, Microsoft is producing positive free cash flow and continuing to return cash to shareholders through buybacks and dividends.
Second, the gross margins of the cloud business are being compressed by the buildout. Microsoft Cloud gross margin was 69% in fiscal 2025 and dropped to 66% in the quarter ended March 31, 2026. Three points of margin on $54.5 billion in quarterly cloud revenue is about $1.6 billion per quarter of margin pressure. Management has been clear that this is the price of building infrastructure ahead of demand. The bet is that the demand materializes faster than the depreciation does.
Third, the depreciation schedule on this infrastructure is what drives whether the bet pays off. Servers and data center equipment typically depreciate over 4-6 years. Buildings and land last longer. If demand for AI compute is sustained for a decade, the assets being built today will throw off cash for years after they are fully depreciated. If demand is concentrated in a 3-4 year window and then plateaus, Microsoft will have built more than it needs.
The skeptical read is that Microsoft is replicating the dot-com era telecom buildout. The optimistic read is that AI compute is a real, durable demand source and Microsoft is building the infrastructure that the next decade of computing will run on. The filings do not resolve which is right. They do, however, document that Microsoft has the cash flow to fund the buildout without borrowing, and that the contracted revenue backlog supports the investment thesis.
That contracted backlog is the most striking single number in the most recent quarterly filing. Microsoft’s commercial Remaining Performance Obligation, which represents contracted future revenue not yet recognized, was $375 billion at June 30, 2025. By March 31, 2026, it was $627 billion. That is a 67% increase in nine months. Some of this represents customers signing longer contracts. Some of it represents new AI-related deals, including the $30 billion Anthropic commitment. Either way, $627 billion in contracted future revenue is a strong signal that the demand the CapEx is being built against actually exists.
Cash Returns: A Decade of Capital Discipline
Microsoft has been one of the most disciplined large-cap capital returners in the public markets. The pattern is consistent: pay a steadily growing dividend, repurchase shares to offset stock-based compensation, and use the rest of the cash flow for accretive M&A and infrastructure investment.
| Fiscal Year | Dividends Paid | Share Repurchases | Total Returned |
|---|---|---|---|
| FY2025 | $24.1B | $13.0B | $37.1B |
| FY2024 | $22.3B | $17.3B | $39.6B |
| FY2023 | $19.8B | $22.2B | $42.0B |
| FY2022 | $18.1B | $32.7B | $50.8B |
| FY2021 | $16.5B | $27.4B | $43.9B |
The repurchase pace has slowed since fiscal 2022, which corresponds with the start of the AI CapEx cycle. Microsoft is choosing to spend more on infrastructure and slightly less on buybacks. Dividends, however, have been raised every single year, reflecting management’s confidence in the durability of the cash generation. The most recent annualized dividend is $3.32 per share, with the most recent quarterly increase to $0.83.
Microsoft authorized a new $60 billion share repurchase program that commenced in April 2025. As of March 31, 2026, $44 billion of that authorization remains. The buybacks are not opportunistic in the value-investing sense (Microsoft does not time them based on price), but the steady removal of shares is meaningful. Diluted share count was 7.43 billion at the end of fiscal 2025 and is roughly flat over the most recent two years because buybacks have offset stock-based compensation issuance.
Stock-based compensation was $12.0 billion in fiscal 2025, which is real economic dilution that the buyback program has to overcome before any net reduction in share count happens. The $13 billion of buybacks in fiscal 2025 essentially neutralized the SBC, with a small net reduction in share count. This is the core capital allocation pattern at Microsoft right now: the dividend grows, the share count stays flat, and the rest of the cash flow funds infrastructure and M&A.
Risks That Could Derail the Thesis
I have already covered the biggest ones (the AI CapEx bet and OpenAI counterparty risk). A complete view also includes the following.
AI infrastructure demand could disappoint. If the demand for AI compute plateaus or shifts to lower-margin uses, the depreciation expense on $200+ billion of property and equipment will outpace the revenue. The contracted backlog of $627 billion mitigates this risk, but only somewhat. Contract cancellations or repricing on renewal would weigh on the long-term return on invested capital.
Hyperscaler competition is intensifying. Amazon Web Services is still the largest cloud provider. Google Cloud is closing in on Azure in some workloads. Oracle is winning AI infrastructure deals at scale. Beyond the public cloud, large customers are building their own AI infrastructure. The 40% Azure growth rate could compress.
Antitrust and regulatory pressure is rising. The European Commission has multiple ongoing investigations into Microsoft’s bundling of Teams with Microsoft 365 (resulting in unbundling commitments) and the OpenAI partnership. The UK’s Competition and Markets Authority approved the Activision deal but with concessions. The US FTC has filed and lost antitrust challenges against Microsoft’s recent M&A. Future deals will face higher scrutiny. The OpenAI partnership itself is under regulatory review in multiple jurisdictions.
Cybersecurity risk is elevated. In November 2023, a Russian state-sponsored actor accessed Microsoft email accounts of senior employees. The breach was disclosed in early 2024 and prompted Microsoft to make security its top corporate priority. The Cyber Safety Review Board, a US government body, criticized Microsoft’s security culture in a 2024 report. A repeat incident could damage customer trust at the worst possible time.
Gaming integration is incomplete. The Activision Blizzard acquisition closed in October 2023 for $75.4 billion. Three years later, gaming revenue is declining. The Activision deal added $51 billion of goodwill to Microsoft’s balance sheet. A goodwill impairment is not an immediate risk, but if the gaming segment’s revenue trajectory does not reverse, an impairment becomes a real possibility within the next 2-3 years.
OpenAI partnership concentration. Microsoft’s AI strategy has diversified meaningfully with the November 2025 Anthropic deal, but OpenAI remains the deepest single commercial relationship and the only one with full IP rights. OpenAI is a non-public organization with unusual governance and a recent history of executive turbulence and corporate-form changes. The October 2025 and April 2026 partnership extensions suggest the relationship is stable. A counterparty whose corporate structure has changed twice in three years is harder to model than a long-term commercial supplier.
Long-running IRS transfer pricing dispute. The IRS has been challenging Microsoft’s Ireland-based transfer pricing arrangements for tax years 2004-2013 since 2007. In September 2023, the IRS issued Notices of Proposed Adjustment seeking $28.9 billion in additional tax plus penalties and interest. Microsoft is contesting through the IRS Independent Office of Appeals. The dispute has been pending for 19 years, which suggests it is more of a structural background liability than a near-term catalyst, but a ruling that goes against Microsoft would require a meaningful one-time cash payment. The timing of any resolution is unknown.
Tariffs and supply chain. Microsoft’s hardware supply chain (Surface devices, Xbox consoles, data center equipment) runs through China and Southeast Asia. The 10-K specifically calls out tariff and trade policy risk as a factor that could increase costs and pressure margins. This is a manageable risk for the company, but it is more material now than it was three years ago.
Valuation
The user request for this analysis was explicit about not using the current stock price or analyst opinions in the valuation. What follows is a fundamentals-only build, using the most recent reported financials and a few different valuation lenses.
Setting the Earnings Baseline
The first thing to settle is what Microsoft’s normalized earnings per share actually is. The reported diluted EPS for the trailing nine months of fiscal 2026 is $13.14, compared to $9.99 in the prior-year period. Annualizing the most recent quarter at $4.27 per share gives $17.08 in run-rate EPS.
Adjustments to that figure:
- The OpenAI Recapitalization gain added $0.60 per share in the first nine months of fiscal 2026. Subtract this for a normalized number. Run-rate EPS adjusted for the recap: roughly $16.50.
- The OBBBA tax law lowers Microsoft’s effective tax rate starting fiscal 2027. The 10-K describes the change as moving the GILTI rate to 14%. Microsoft’s foreign income before tax was $54.4 billion in fiscal 2025. A 4-percentage-point reduction on that base is $2.2 billion of additional after-tax income, or roughly $0.30 per share.
- Underlying revenue growth at 18% with operating leverage suggests EPS could grow to $19-20 in fiscal 2027 with a tailwind from the tax change.
I will use $16.50 as a clean trailing baseline and $19.00 as a reasonable forward estimate.
P/E on Forward Earnings
| Multiple | Trailing $16.50 | Forward $19.00 |
|---|---|---|
| 22x (low end of large-cap tech) | $363 | $418 |
| 28x (Microsoft’s historical average through cycles) | $462 | $532 |
| 35x (premium for AI exposure and growth) | $578 | $665 |
A 28x multiple on forward earnings gives roughly $530 per share. That multiple is justified if you accept that Azure’s 40% growth is partially sustainable, that the contracted backlog converts to revenue, and that the AI CapEx is an investment rather than a recurring expense.
EV / Free Cash Flow
Microsoft generated $71.6 billion of free cash flow in fiscal 2025. The growth CapEx is depressing this number. If you assume “maintenance CapEx” runs at roughly $25-30 billion per year (which approximates the pre-AI-buildout level adjusted for inflation), then normalized FCF on the current revenue base would be roughly $115 billion.
| Method | FCF | Multiple | Equity Value | Per Share |
|---|---|---|---|---|
| Reported FCF | $71.6B | 35x | $2.51T + $62B net cash = $2.57T | $346 |
| Reported FCF | $71.6B | 40x | $2.86T + $62B = $2.92T | $393 |
| Normalized FCF | $115B | 30x | $3.45T + $62B = $3.51T | $472 |
| Normalized FCF | $115B | 35x | $4.03T + $62B = $4.09T | $551 |
The honest range here is $400 to $550, depending on how much credit you give the “normalized” FCF concept.
Sum of the Parts
Treating each segment as an independent business and applying segment-appropriate multiples:
| Segment | Annualized OI | Multiple | Value |
|---|---|---|---|
| Productivity & Business Processes | $85B | 28x | $2.38T |
| Intelligent Cloud (Azure-led, 40% growth) | $55B | 35x | $1.93T |
| More Personal Computing | $15B | 15x | $0.23T |
| Operating Businesses Subtotal | $155B | $4.54T | |
| OpenAI and Anthropic equity stakes (carrying value plus markup) | ~$85B | ||
| Net cash and investments | $62B | ||
| Less: IRS dispute reserve | ($30B) | ||
| Estimated Equity Value | $4.66T | ||
| Per share (7.43B shares) | $627 |
Sum of the parts produces the highest valuation because it allows each segment to be valued at its own appropriate multiple without corporate overhead. It is best used as an upper-bound sanity check.
Discounted Cash Flow
Assumptions:
- Revenue grows 18% in fiscal 2026, 16% in 2027, decelerating to 10% by fiscal 2030 and 5% by fiscal 2035.
- FCF margin expands from current depressed level (around 21% reported) to 32% by 2030 as CapEx normalizes.
- Terminal growth rate of 4% (high, but defensible for a company with this kind of moat).
- Discount rate of 8.5% (Microsoft’s WACC is debatable, but this approximates a reasonable input).
The output of that DCF is roughly $4.1 trillion in equity value, or $552 per share. The most sensitive assumption is the rate at which CapEx normalizes. If it stays at 30%+ of revenue indefinitely, the DCF drops to roughly $440 per share. If CapEx falls back to 12-15% of revenue by 2029, the DCF approaches $620 per share.
Bear Case: Azure Decelerates and CapEx Returns Disappoint
In this scenario, Azure’s growth rate falls below 25% within four quarters as enterprise AI demand normalizes. The CapEx cycle continues, but the assets do not earn their cost of capital because utilization is lower than planned. Cloud gross margin compresses below 60%. Either OpenAI or Anthropic (or both) renegotiates terms in a way that costs Microsoft margin or share. Goodwill impairments hit the gaming segment. The result is an EPS path that flattens around current levels rather than expanding.
| Assumption | Value |
|---|---|
| Trailing EPS (no growth) | $16.50 |
| Multiple | 22x (low end of large-cap tech) |
| Per share | $363 |
Base Case: Demand Materializes and CapEx Normalizes by 2029
This is the most likely path given current evidence. Azure stays in the 25-35% growth band for several more years before slowing. The contracted backlog of $627 billion converts into recognized revenue at expected margins. CapEx peaks in fiscal 2027-2028, then begins moderating in fiscal 2029 as the build-ahead position normalizes. The OBBBA tax change provides a $0.30+ EPS tailwind starting fiscal 2027. Microsoft’s commercial reorganization under Althoff produces modest productivity gains. Forward EPS reaches $19 by fiscal 2027.
| Assumption | Value |
|---|---|
| Forward EPS (FY2027) | $19.00 |
| Multiple | 28x (Microsoft’s historical multiple through cycles) |
| Per share | $532 |
A cross-check using the EV/FCF method assuming normalized FCF of $115 billion at a 35x multiple plus $62 billion of net cash gives $4.09 trillion of equity value, or $551 per share. The DCF using the assumptions above gives $552 per share. All three base case methods cluster between $530 and $555.
Bull Case: Microsoft Owns the AI Infrastructure Layer
In this scenario, the multi-model AI strategy works as intended. Azure remains the default enterprise AI infrastructure provider. The $235 billion of pending datacenter commitments runs at high utilization with paying customers. Microsoft 365 Copilot adoption accelerates as Anthropic and OpenAI integrations differentiate Microsoft’s offering from Google Workspace. Forward EPS reaches $20+ by fiscal 2027, and the market awards a premium multiple for the demonstrated AI infrastructure moat.
| Assumption | Value |
|---|---|
| Forward EPS (FY2027) | $20.00 |
| Multiple | 32x (premium for confirmed AI infrastructure moat) |
| Per share | $640 |
A cross-check using sum-of-the-parts logic (applying segment-specific multiples to operating income, then adding the OpenAI and Anthropic equity stakes and net cash) gives $627 per share. Both bull case methods cluster around $625-640.
Summary
| Scenario | Method | Per Share |
|---|---|---|
| Bear | 22x P/E on $16.50 trailing EPS | $363 |
| Bear | Reported FCF at 35x with no margin recovery | $346 |
| Base | 28x P/E on $19 forward EPS | $532 |
| Base | EV / FCF at 35x normalized | $551 |
| Base | DCF with normalized CapEx | $552 |
| Bull | 32x forward P/E with AI premium | $640 |
| Bull | Sum of the parts | $627 |
The fair-value range I would put forward is $430 to $580, with a midpoint near $505.
The most important variable is not any single assumption in the model. It is what happens to the $235 billion of additional infrastructure spending Microsoft has committed to. If those data centers are running at high utilization with paying customers in three years, the bull case is conservative. If utilization is low, the bear case is too generous.
If you found this helpful, consider supporting my work. This blog has no paywalled content and reader support is how it stays that way. You can also subscribe to my mailing list to get new deep dives in your inbox.
What to Watch
The catalysts below are the events most likely to either confirm or invalidate the thesis over the next 18 months. You can download these catalysts as a calendar file to import into Google Calendar, Apple Calendar, or Outlook. The file includes specific dates for the dated events and quarterly placeholders for the recurring ones.
| Catalyst | Why It Matters | Timeline |
|---|---|---|
| Microsoft Build 2026 | Annual developer conference; expect MAI-1 progression and Copilot announcements | May 19-22, 2026 |
| Q4 FY2026 earnings | First fiscal year-end with full year of AI CapEx run-rate; full year RPO disclosure | July 30, 2026 |
| FY2027 GILTI rate transition | OBBBA tax change kicks in; $0.30+ EPS tailwind expected | July 1, 2026 |
| Q1 FY2027 earnings | First quarter under new GILTI rate; Anthropic Azure consumption begins to flow through | October 28, 2026 |
| Microsoft Ignite 2026 | Annual enterprise customer conference; expect new Azure AI infrastructure disclosures | November 17-21, 2026 |
| 2026 Annual Shareholders Meeting | Director elections, executive compensation vote, shareholder proposals | December 2026 |
| Q2 FY2027 earnings | Mid-year read on Azure growth and cloud gross margin | January 27, 2027 |
| HIMSS 2027 | Healthcare conference; expect Dragon Copilot expansion and customer adoption metrics | March 2027 |
| Q3 FY2027 earnings | Full year of Anthropic deal flowing through Azure consumption | April 28, 2027 |
| Azure growth deceleration threshold | Currently 40%; key threshold is whether it stays above 25% | Each quarterly report |
| Cloud gross margin trend | Currently 66%, down from 69%; further decline indicates CapEx outrunning demand | Each quarterly report |
| Buyback pace | $44B remaining on $60B authorization; pace signals management confidence | Each quarterly report |
| Activision/Gaming impairment risk | Xbox hardware down 33%; impairment becomes possible if no turnaround in 18-24 months | FY2027-2028 |
Sources:
- 10-K filed July 30, 2025 (fiscal year ended June 30, 2025)
- 10-Q filed April 29, 2026 (quarter ended March 31, 2026)
- 8-K filed April 29, 2026 (Q3 FY2026 earnings release)
- “Accelerating our commercial growth,” Satya Nadella, The Official Microsoft Blog, October 1, 2025
- “Microsoft, NVIDIA and Anthropic announce strategic partnerships,” The Official Microsoft Blog, November 18, 2025
- “Microsoft adds Anthropic AI model to Copilot assistant,” CNBC, September 24, 2025
- “Microsoft Dragon Copilot provides the healthcare industry’s first unified voice AI assistant,” Microsoft News, March 3, 2025
- “Microsoft extends AI advancements in Dragon Copilot to nurses and partners,” Microsoft News, October 16, 2025
- StatCounter Global Stats, Desktop Operating System Market Share Worldwide
- Usage share of operating systems, Wikipedia, accessed April 2026
Research and analysis conducted with AI assistance using SEC EDGAR filings as primary sources.