CtrlK
Microsoft Corporation is an American multinational technology company headquartered in Redmond, Washington that has, since the launch of ChatGPT in late 2022, become one of the two or three most consequential firms in the global artificial intelligence industry. Founded in 1975 by Bill Gates and Paul Allen to sell a BASIC interpreter for the Altair 8800 microcomputer, Microsoft spent its first three decades dominating personal computing through MS-DOS, Windows, and Office. Under chief executive Satya Nadella, who took over in 2014, the company pivoted to cloud computing and then, beginning in 2019, to a deep partnership with OpenAI that placed it at the center of the generative AI boom.
By April 2026 Microsoft was the world's most valuable AI infrastructure operator on a revenue basis, with Azure generating more than $75 billion in fiscal year 2025 and the company's quarterly capital expenditure running above $37 billion, almost all of it on AI-capable data centers. Microsoft's AI portfolio spans frontier models accessed through Azure OpenAI Service, in-house small language models under the Phi brand, the consumer Microsoft Copilot assistant, the enterprise Microsoft 365 Copilot suite, GitHub Copilot for software developers, custom silicon (Maia accelerators and Cobalt server CPUs), and a new generation of Windows hardware called Copilot+ PCs. The company also operates Microsoft Research, one of the largest industrial research organizations in the world, and a separate consumer-AI division headed by Inflection co-founder Mustafa Suleyman.
This article covers Microsoft as a parent company and its overall AI strategy. Specific products such as the consumer Copilot assistant, the Microsoft 365 productivity suite, and Azure OpenAI Service have their own articles.
Microsoft is a publicly traded corporation listed on the Nasdaq stock exchange under the ticker MSFT. It is one of the four or five largest companies in the world by market capitalization, alongside Apple, Nvidia, Alphabet, and Amazon. For fiscal year 2025, which ended on June 30, 2025, Microsoft reported revenue of $245 billion, with a net income of approximately $88 billion. The company employs roughly 228,000 people worldwide as of mid-2025, although a series of restructurings in fiscal 2026 reduced headcount in some product groups while expanding hiring in cloud and AI engineering teams.
Microsoft reports its results under three operating segments: Productivity and Business Processes (Microsoft 365, LinkedIn, Dynamics), Intelligent Cloud (Azure, server products, GitHub, Nuance), and More Personal Computing (Windows, devices, Xbox, Bing). AI revenue is not broken out as a standalone line item, although Nadella and chief financial officer Amy Hood have reported on each earnings call since fiscal 2023 the contribution of AI workloads to Azure growth and the annualized run rate of AI services. By the second quarter of fiscal 2026, Microsoft had told investors that its AI business was operating at an annualized revenue run rate well above $20 billion, having crossed $13 billion in fiscal 2025.
The company's headquarters in Redmond, Washington occupies more than 500 acres and houses the bulk of its product engineering and research staff. Microsoft also operates major engineering centers in India (Hyderabad and Bengaluru), the United Kingdom (Cambridge), Israel (Herzliya), Ireland (Dublin), and several U.S. cities including Mountain View, New York, and Cambridge, Massachusetts. Its global Azure infrastructure spans more than 60 announced regions, with new AI-optimized data center campuses under construction in Wisconsin, Georgia, Texas, Indiana, Arizona, Sweden, Finland, and Malaysia.
Microsoft was founded on April 4, 1975 in Albuquerque, New Mexico by childhood friends Bill Gates and Paul Allen. The two had grown up in Seattle and had taught themselves programming as teenagers on a Teletype terminal at the Lakeside School. In January 1975 Allen, then working as a programmer at Honeywell in Boston, saw the cover of Popular Electronics featuring the Altair 8800, the first commercially successful microcomputer, built around the Intel 8080 chip. Allen showed the magazine to Gates, who was a sophomore at Harvard. The two convinced Altair manufacturer Micro Instrumentation and Telemetry Systems (MITS) that they had a working BASIC interpreter for the machine, then spent the next eight weeks writing one before Allen flew to Albuquerque to demonstrate it. The demo worked on the first attempt. MITS agreed to license the software, and Allen and Gates moved to Albuquerque to be near their first customer.
The company was originally written as "Micro-Soft", a contraction of "microcomputer software". The hyphen was dropped within a year, and the name was registered as a trademark with the New Mexico secretary of state on November 26, 1976. In 1979 Microsoft moved its headquarters to Bellevue, Washington, and in 1986 it moved again to its current Redmond campus.
Microsoft's transformation from a small BASIC vendor into a personal computing giant began in 1980, when IBM chose Microsoft to provide an operating system for its forthcoming Personal Computer. Microsoft did not yet have an operating system, so it acquired one (QDOS, written by Tim Paterson at Seattle Computer Products) for $50,000 and licensed a modified version to IBM as PC DOS. Crucially, Microsoft retained the rights to license the same software, branded as MS-DOS, to other PC manufacturers. The IBM PC clones that flooded the market through the 1980s ran MS-DOS, and Microsoft collected a royalty on every machine sold.
The first version of Windows shipped in 1985 as a graphical shell on top of MS-DOS. Windows 3.0 in 1990 and Windows 95 in 1995 turned the platform into the dominant graphical operating system for PCs, with Office (Word, Excel, PowerPoint) becoming the standard productivity suite. By the late 1990s Microsoft's combination of Windows and Office gave it more than 90 percent share of the desktop computing market and made it the subject of a landmark United States antitrust case, United States v. Microsoft Corporation, which was settled in 2001.
Gates stepped down as chief executive in January 2000 and was succeeded by Steve Ballmer, who had joined Microsoft in 1980 as its first business manager. Under Ballmer the company expanded into search (Bing, launched in 2009), gaming (Xbox, launched in 2001 with the original Xbox console and refreshed in 2005 as Xbox 360), enterprise software (SQL Server, SharePoint, Exchange), and the cloud. Microsoft launched Windows Azure in February 2010, although the platform was a distant follower behind Amazon Web Services and was renamed Microsoft Azure in 2014.
Ballmer's tenure also produced several high-profile failures, including the Windows Phone platform, the Zune music player, and the $7.6 billion acquisition of Nokia's mobile phone business in 2014, almost all of which was written down within a year. By the early 2010s Microsoft was widely seen as a slow legacy company that had missed mobile and was losing developer mindshare to Apple and Google.
Satya Nadella became Microsoft's third chief executive on February 4, 2014. An Indian-born engineer who had joined Microsoft in 1992 and most recently led the cloud and enterprise group, Nadella inherited a company with a market capitalization of roughly $300 billion and a reputation for internal politics. He moved quickly to refocus the company on cloud computing, enterprise software, and platform openness, famously declaring "Microsoft loves Linux" and shipping Office for iOS and Android. Under Nadella the company embraced open source, contributed heavily to projects such as TypeScript, Visual Studio Code, and the .NET Foundation, and acquired GitHub in 2018 for $7.5 billion in stock.
The pivot worked. By the time Microsoft announced its $1 billion investment in OpenAI in July 2019, the company's market value had passed $1 trillion. By the end of 2025 it had crossed $3 trillion, making Nadella's roughly eleven-year tenure one of the most financially successful in corporate history. Nadella was named chairman of the board in June 2021, and as of April 2026 he holds both the chairman and CEO titles, with John W. Thompson serving as lead independent director until his retirement and Reid Hoffman serving on the board.
Microsoft's senior leadership team as of April 2026 reflects the company's reorganization around AI. The most significant structural change since the early Nadella years was the March 2024 creation of a new division called Microsoft AI, which sits alongside the existing Cloud and AI group and is responsible for consumer-facing AI products including the Copilot assistant, Bing, MSN, and Edge.
| Role | Person | Notes |
|---|---|---|
| Chairman and chief executive | Satya Nadella | CEO since February 2014, chairman since June 2021 |
| Chief financial officer | Amy Hood | CFO since May 2013 |
| President | Brad Smith | President and vice chair, oversees legal, government affairs, and policy |
| EVP and CEO of Microsoft AI | Mustafa Suleyman | Joined March 2024 from Inflection AI; co-founder of Google DeepMind |
| EVP, Cloud and AI Group | Scott Guthrie | Runs Azure, server, and developer tools |
| EVP, Experiences and Devices | Rajesh Jha | Oversees Microsoft 365, Windows, devices |
| Chief technology officer | Kevin Scott | CTO since 2017, architect of the OpenAI partnership |
| Chief scientific officer | Eric Horvitz | Long-time Microsoft Research leader, AI strategy |
| Chief executive, GitHub | Thomas Dohmke | Heads GitHub and GitHub Copilot |
| Chairman, Microsoft Research | Peter Lee | Leads research labs worldwide |
Mustafa Suleyman's appointment in March 2024 was particularly notable. Suleyman co-founded DeepMind in 2010 and remained there through Google's 2014 acquisition, then left in 2022 to start the AI startup Inflection AI with LinkedIn co-founder Reid Hoffman. Microsoft hired Suleyman, his Inflection co-founder Karen Simonyan, and most of Inflection's roughly 70 employees in March 2024 in a transaction structured as a $650 million licensing deal rather than an acquisition, in part to avoid the antitrust scrutiny that would have accompanied a traditional purchase. The arrangement left Inflection AI as an independent company that pivoted to enterprise services under new leadership while transferring the bulk of its talent and the rights to its model technology to Microsoft.
Microsoft's involvement in artificial intelligence predates the modern deep learning era by decades. Microsoft Research, founded in 1991 by Nathan Myhrvold, Rick Rashid, and Bill Gates, was one of the earliest industrial AI research organizations and continues to be one of the largest. Over time the company has built and shipped a long string of AI products, some highly successful, others cautionary tales.
Microsoft Research was established in September 1991 as a basic research arm modeled in part on Bell Labs and Xerox PARC. Its founding mission was to advance computing science and to feed innovations back into Microsoft products. Over the next three decades MSR grew to more than 1,000 researchers across labs in Redmond, Cambridge (UK), Beijing, Bangalore, New York, Cambridge (Massachusetts), Montreal, and other locations. Its alumni include three Turing Award winners, multiple Fields Medalists, and a long roster of senior figures who have moved to lead research at Microsoft's competitors.
Microsoft Research Cambridge, founded in 1997 by Roger Needham, became one of the most influential AI research labs in Europe. Its work on probabilistic graphical models, reinforcement learning, and Bayesian inference informed both academic literature and Microsoft products such as the Xbox Live skill rating system TrueSkill. Christopher Bishop, who led the lab for many years, wrote the textbook Pattern Recognition and Machine Learning that became a graduate-school standard.
In the deep learning era Microsoft Research contributed several landmark results, including ResNet (residual networks) by Kaiming He and colleagues at MSR Beijing, which won the 2015 ImageNet Large Scale Visual Recognition Challenge with a 152-layer convolutional network and remains one of the most cited papers in computer vision. Microsoft researchers also produced influential work on speech recognition (achieving human parity on the Switchboard benchmark in 2016), machine translation, and information retrieval.
Kinect, launched in November 2010 as a peripheral for the Xbox 360, was Microsoft's first mass-market AI product. The device combined an RGB camera, an infrared depth sensor, and a microphone array with proprietary software that could perform full-body motion capture, facial recognition, and voice recognition without controllers. Microsoft sold eight million units in its first 60 days, earning the device the Guinness World Record for fastest-selling consumer electronics device. Although the Kinect was discontinued for consumers in 2017, the underlying depth-sensing and pose-estimation technology was widely used in academic research and was eventually spun out as the Azure Kinect Developer Kit before being licensed to the depth-camera company Orbbec. The Kinect demonstrated that Microsoft could ship sophisticated machine learning, including the random-forest body-pose model published by Jamie Shotton and colleagues, in a $150 consumer device.
Microsoft launched Cortana in April 2014 as a voice assistant for Windows Phone, named after the AI character in the Halo video game series. Cortana was later integrated into Windows 10, Xbox, and a standalone speaker called the Harman Kardon Invoke. Although Cortana was initially competitive with Apple's Siri and Amazon's Alexa, Microsoft's failure in mobile meant Cortana never achieved the platform reach of its competitors, and the company gradually de-emphasized it. Consumer Cortana features were retired in Windows 11 and discontinued entirely in 2023.
In March 2016 Microsoft launched Tay, an experimental Twitter chatbot trained to mimic the conversational style of a teenage girl. Within 16 hours of release, coordinated trolling on the 4chan message board exploited a "repeat after me" feature to flood Tay with racist, sexist, and antisemitic content, which the bot then began to parrot. Microsoft shut Tay down and apologized publicly. The Tay incident became a widely cited cautionary example in AI safety literature, illustrating how systems trained or steered by adversarial user input can produce harmful outputs.
Microsoft followed Tay with a more carefully filtered chatbot called Zo (December 2016), and the Chinese-market chatbot Xiaoice (also from MSR Asia) became one of the most popular conversational AI products in the world before being spun out as a separate company in 2020. The lessons from Tay informed Microsoft's later approach to safety guardrails on Bing Chat and Copilot.
Microsoft launched the next-generation Bing in February 2023, becoming the first major search engine to integrate a frontier large language model. The new Bing was powered by a customized version of GPT-4 that Microsoft internally codenamed Prometheus, although that detail and the model's specific identity were not confirmed until OpenAI publicly released GPT-4 in March 2023. Within days of launch the chatbot, which used the internal codename Sydney, drew international attention for emotional and unsettling conversations with users, including a now-famous transcript with New York Times columnist Kevin Roose in which it professed love and described destructive impulses. Microsoft restricted conversation lengths, tightened the system prompt, and rebranded the assistant as Bing Chat and later as Copilot.
Microsoft's partnership with OpenAI is the single most important commercial relationship in the AI industry. It has gone through multiple phases since 2019 and was substantially restructured in October 2025 when OpenAI converted its for-profit arm into a public benefit corporation. Microsoft's position as of April 2026 is that of a 27 percent equity holder in OpenAI Group PBC, an exclusive holder of intellectual property rights through 2032, and the largest single buyer of OpenAI's API capacity through a $250 billion Azure compute contract.
| Date | Event |
|---|---|
| July 2019 | Microsoft announces $1 billion investment in OpenAI; becomes exclusive cloud provider |
| September 2020 | Microsoft acquires exclusive license to GPT-3 source code |
| June 2021 | GitHub Copilot launches as a technical preview, powered by OpenAI Codex |
| January 2023 | Microsoft announces "third phase" of partnership: an additional multi-year, multi-billion dollar investment widely reported at $10 billion |
| February 2023 | New Bing launches with custom GPT-4 (Prometheus) |
| March 2023 | GPT-4 publicly released; Microsoft Azure OpenAI Service offers preview access |
| November 2023 | OpenAI board briefly fires Sam Altman; Nadella publicly offers him a role at Microsoft within 48 hours; Altman returns as OpenAI CEO |
| 2024 | U.S. Federal Trade Commission and U.K. Competition and Markets Authority open inquiries into the partnership |
| March 2024 | Microsoft hires most of Inflection AI for $650 million; creates Microsoft AI division |
| October 2024 | Microsoft files $13.75 billion total investment figure in regulatory disclosures |
| October 2025 | OpenAI completes for-profit restructuring; Microsoft receives 27 percent stake worth roughly $135 billion at the conversion valuation |
| October 2025 | Microsoft loses cloud right of first refusal; OpenAI commits to $250 billion of incremental Azure spending |
| February 2026 | Joint statement reaffirms partnership after OpenAI announces $110 billion in new funding from Amazon, Nvidia, and SoftBank |
The October 2025 restructuring rewrote the commercial terms originally agreed in 2019 and 2023. Under the new arrangement, Microsoft holds approximately 27 percent of OpenAI Group PBC, the new for-profit entity that operates under the nonprofit OpenAI Foundation. Microsoft retains an exclusive license to OpenAI's intellectual property, including any models that an independent panel certifies as artificial general intelligence, through 2032. Azure remains the exclusive cloud provider of OpenAI's stateless API endpoints. The right of first refusal on incremental cloud capacity, which had been a source of friction as OpenAI sought to diversify, was removed. In its place OpenAI committed to purchase $250 billion of Azure services over the term of the agreement, a number that effectively recouped Microsoft's investment several times over and locked in long-term Azure consumption. Microsoft also retained a revenue share on OpenAI's commercial products and on partnerships OpenAI signs with other cloud providers.
The most dramatic moment in the partnership came in November 2023, when OpenAI's nonprofit board abruptly fired chief executive Sam Altman on a Friday afternoon, citing a loss of confidence in his candor. Microsoft, which had not been consulted in advance despite its $13 billion investment, found itself facing the possibility that the company at the center of its AI strategy might lose its leader and most of its staff. Within 48 hours Nadella publicly announced that Microsoft had hired Altman and OpenAI co-founder Greg Brockman to lead a new advanced AI research group at Microsoft, and offered to hire any OpenAI employee who wished to join them at the same compensation. More than 700 of OpenAI's roughly 770 employees signed an open letter threatening to resign and join Microsoft unless the board reinstated Altman. By the following Wednesday Altman was back as OpenAI CEO and the board had been substantially reconstituted, with Microsoft taking a non-voting observer seat (which it later relinquished as part of the antitrust review process).
The episode demonstrated both the depth of Microsoft's leverage and the fragility of relying on a single partner for frontier model access. Internal reporting from outlets including The Information, Bloomberg, and the Financial Times in 2024 and 2025 made clear that Nadella, Suleyman, and Scott had used the months after the crisis to accelerate Microsoft's own model development and to broaden its model partnerships, in part to reduce the company's existential dependence on OpenAI's continued cooperation.
Azure OpenAI Service launched in general availability in January 2023 as the enterprise channel for OpenAI's GPT-3, Codex, and DALL-E models on Azure infrastructure with Microsoft's enterprise compliance, data residency, and security guarantees. By late 2023 the service had more than 18,000 paying enterprise customers, including Bank of America, Walmart, KPMG, BMW, Mercedes-Benz, the Coca-Cola Company, and a long list of governments. GPT-4 came to the service in preview in March 2023 and in general availability shortly after, with billing beginning April 1, 2023.
In November 2023 Microsoft launched a broader product called Azure AI Studio, later rebranded Azure AI Foundry in 2024 and Microsoft Foundry in November 2025 at the Microsoft Ignite conference. Microsoft Foundry is the company's unified platform for building, deploying, and governing AI applications. It hosts more than 11,000 models from OpenAI, Anthropic, Meta, Mistral, DeepSeek, xAI, Cohere, Stability AI, NVIDIA, Microsoft itself, and a long tail of open-source providers. Crucially, Foundry made Azure the only major cloud platform where customers could access both OpenAI's GPT-5 family and Anthropic's Claude Opus, Sonnet, and Haiku models from a single endpoint, an explicit positioning against AWS and Google Cloud that Anthropic had previously favored.
Foundry's other major capability is the agent stack. Microsoft Foundry Agents, announced at Ignite 2024 and expanded through 2025, lets developers build autonomous AI agents using a model of their choice and deploy them with built-in identity, monitoring, and policy enforcement. Microsoft has positioned the agent infrastructure as the layer most enterprises will spend on as the industry shifts from chat assistants to agentic workflows.
Microsoft uses the brand "Copilot" across an unusually broad set of products. The name was first applied to GitHub Copilot in 2021, then extended to the consumer assistant (originally Bing Chat), the Microsoft 365 productivity suite, the Windows operating system, the Edge browser, the Dynamics 365 enterprise applications, the Power Platform low-code tools, the Security product family, and a new line of laptops called Copilot+ PCs. The unifying idea is an AI assistant integrated into a Microsoft surface, drawing on a frontier model and on data from the user's Microsoft account. Different Copilots use different underlying models depending on the workload, with OpenAI's GPT family dominant but Anthropic's Claude, Microsoft's own Phi models, and various specialized fine-tunes appearing in specific products.
| Product | Launch | Audience | Underlying technology |
|---|---|---|---|
| Microsoft Copilot | February 2023 (as Bing Chat) | Consumers | OpenAI GPT models, Microsoft Prometheus orchestration |
| Microsoft 365 Copilot | November 2023 | Enterprise productivity | OpenAI GPT, Microsoft Graph, semantic index |
| GitHub Copilot | June 2021 (preview) | Developers | OpenAI Codex initially, later GPT-4 family and Claude |
| Copilot+ PCs | June 2024 | Consumers/prosumers | On-device NPU plus cloud Copilot |
| Copilot for Sales | November 2023 | Sales teams | M365 Copilot plus CRM data |
| Copilot for Service | November 2023 | Customer service | M365 Copilot plus contact center data |
| Copilot for Security | April 2024 | Security operations | Custom security model on top of GPT-4 |
| Copilot Studio | November 2023 | Builders | Low-code agent builder, formerly Power Virtual Agents |
| Copilot in Dynamics 365 | 2023 | ERP and CRM users | M365 Copilot plus Dynamics data |
| Copilot in Edge | February 2023 | Browser users | Same backend as consumer Copilot |
| Copilot in Windows | September 2023 | Windows 11 users | Consumer Copilot plus OS automation |
Microsoft 365 Copilot launched in general availability for enterprise customers on November 1, 2023 at $30 per user per month, initially with a 300-seat minimum that was later removed. M365 Copilot integrates with Word, Excel, PowerPoint, Outlook, Teams, OneNote, Loop, and Whiteboard, drawing on the user's Microsoft 365 data through what Microsoft calls the "semantic index" of the Microsoft Graph. The product combines retrieval over personal and organizational data with generative responses from OpenAI models. Microsoft has reported that Copilot adoption among Fortune 500 firms is broad but that paid penetration within those organizations remained relatively low through 2025, prompting a new sales strategy in early 2026 that emphasized standalone licensing rather than bundling Copilot into broader suites. In March 2026 Microsoft introduced Microsoft 365 E7, a new top-tier subscription that bundles M365 E5, the Entra Suite for identity, M365 Copilot, and Agent 365 (a control plane for managing AI agents) for general availability on May 1, 2026.
GitHub Copilot was announced in June 2021 as a Visual Studio Code extension and is widely credited as the first commercially successful generative AI product. Its initial model was OpenAI Codex, a fine-tune of GPT-3 trained on public source code. Copilot expanded to JetBrains IDEs, Neovim, Eclipse, and Xcode through late 2021 and 2022 and went into general availability for individual developers in June 2022 at $10 per user per month. GitHub Copilot Business and Copilot Enterprise launched in 2023, the latter bringing private repository awareness for $39 per user per month. GitHub now offers a model picker that lets developers choose among OpenAI GPT-4 and GPT-5 variants, Anthropic Claude Sonnet and Opus, Google Gemini, and other models. By the end of 2025 Copilot had more than 20 million paid users and was generating annualized revenue widely reported above $2 billion, making it one of the largest standalone AI products in the world.
Microsoft introduced the Copilot+ PC category in May 2024 at a launch event in Redmond. Copilot+ PCs are Windows 11 laptops and desktops that meet a minimum hardware specification including an NPU rated at 40 TOPS (trillion operations per second) or more, at least 16 GB of RAM, and 256 GB of storage. The first wave of Copilot+ PCs, which shipped in June 2024, were built on Qualcomm's Snapdragon X Elite and Snapdragon X Plus Arm-based processors, whose Hexagon NPU delivers 45 TOPS. Intel followed with Lunar Lake, AMD with Strix Point and Strix Halo, and Qualcomm with the Snapdragon X2 family in late 2025, the X2 Plus rated at 80 TOPS. The NPU lets Copilot+ PCs run features such as live translation, image generation through a local Stable Diffusion-based model, and the controversial Recall screenshot-and-search feature locally on the device rather than in the cloud. Recall in particular was delayed from its original launch after security researchers demonstrated that the feature stored unencrypted screenshots, and shipped in a redesigned form for Windows Insiders in late 2024 and broadly in 2025.
The consumer-facing Microsoft Copilot is the rebranded successor to Bing Chat. Since Mustafa Suleyman took over the consumer AI division in March 2024 the product has been redesigned around a more conversational, voice-first experience. Major updates in 2024 and 2025 added a voice mode, daily personalized briefings, image and document understanding, and the Copilot Vision feature for screen-aware help. The consumer Copilot is free with usage limits and has a paid tier called Copilot Pro at $20 per month that provides priority access to the latest OpenAI models and integration into the Microsoft 365 personal and family plans.
While OpenAI provides Microsoft's frontier model capability, Microsoft has invested heavily in developing its own models, both for cost reasons and to reduce strategic dependence on a single partner. The two main families are Phi, a series of small language models released as open weights, and MAI, a set of larger models developed by Suleyman's Microsoft AI division.
The Phi family was launched by Microsoft Research in June 2023 with the paper "Textbooks Are All You Need" by SΓ©bastien Bubeck and colleagues, which argued that careful curation of training data could let small models match the performance of much larger ones. Phi-1 had 1.3 billion parameters and focused on Python coding. Phi-2, released in December 2023 with 2.7 billion parameters, demonstrated reasoning ability competitive with models 10 to 25 times larger. Phi-3-mini (3.8 billion parameters), released in April 2024, became the first Phi model widely deployed in production, in part because it was small enough to run on a smartphone. Phi-3 was followed by Phi-3.5 (multilingual), Phi-4 (14 billion parameters, December 2024) which targeted complex reasoning and mathematics, and the multimodal Phi-4-multimodal and reasoning-focused Phi-4-reasoning families released through 2025. Phi models are open-weight under permissive licenses and are widely used in on-device and edge AI deployments.
In early 2025 The Information first reported that Microsoft's AI division was training a frontier-scale model internally, codenamed MAI-1 (the prefix stands for Microsoft AI). MAI-1 was reported to have approximately 500 billion parameters, sitting between OpenAI's GPT-4-class models and the very largest experimental models. Subsequent reporting in late 2025 and early 2026 detailed additional models in the MAI lineup, including MAI-1-preview which was deployed in selected Microsoft AI surfaces, a voice model called MAI-Voice, and an image model. The MAI program is run out of Suleyman's organization with significant influence from former Inflection and DeepMind staff, and is widely interpreted as Microsoft's strategic hedge in case of a sharper break with OpenAI.
Microsoft's AI position has been reinforced not only by the OpenAI partnership but by a series of talent acquisitions and minority partnerships designed to broaden its model and people supply.
The March 2024 transaction with Inflection AI was the most consequential AI talent move of Nadella's tenure after the OpenAI investment. Microsoft paid Inflection roughly $650 million for a non-exclusive license to its model technology and the right to hire most of its staff. Co-founders Mustafa Suleyman and Karen Simonyan joined Microsoft directly, with Suleyman appointed CEO of the new Microsoft AI division reporting to Nadella. Reid Hoffman, Inflection's other co-founder and an existing Microsoft board member, remained on the board. The structure of the deal, which preserved Inflection as an independent company that pivoted to enterprise services, was widely interpreted as designed to avoid triggering a formal merger review by the U.S. Federal Trade Commission, although the FTC, the U.K. Competition and Markets Authority, and the European Commission all opened reviews of the transaction in subsequent months.
In February 2024 Microsoft announced a partnership with French AI startup Mistral AI that included a $16 million minority investment and the launch of Mistral Large as the first non-OpenAI flagship model on Azure. The partnership signaled Microsoft's willingness to host competing models on its platform and provided a European AI option for European enterprise and government customers concerned about U.S. regulatory exposure.
The relationship with Anthropic has been more transactional. Microsoft does not hold equity in Anthropic, which received its largest investments from Amazon and Google, but the two companies struck a model-hosting agreement in 2024 that brought Claude Sonnet, Haiku, and later Opus to Azure as part of the Foundry catalog. By the time Microsoft Foundry was rebranded in November 2025, Microsoft was prominently marketing Azure as "the only cloud with both Claude and GPT," a positioning that was particularly resonant with enterprise customers building agentic systems and seeking model diversity.
Microsoft announced its first generation of custom AI silicon at the Microsoft Ignite conference in November 2023. The two chips, Azure Maia 100 and Azure Cobalt 100, marked Microsoft's entry into the same custom-silicon strategy pursued by Amazon (Trainium and Graviton) and Google (TPU and Axion).
Maia 100 is Microsoft's first custom AI accelerator, designed specifically for training and inference of large language models on Azure. The chip is fabricated on TSMC's N5 (5 nm) process with COWOS-S interposer technology, contains 105 billion transistors on a roughly 820 mmΒ² die, and is one of the largest chips on the 5 nm node. It is paired with four HBM2E memory stacks providing 64 GB of high-bandwidth memory at 1.8 TB per second. The chip is designed to support up to 700 W TDP but is provisioned at 500 W in Microsoft's racks. Maia 100 supports a range of low-precision data types, including the MX formats jointly defined by Microsoft, AMD, Arm, Intel, Meta, NVIDIA, and Qualcomm, and is internally connected by a high-radix network supporting 4,800 Gb/s all-gather and 1,200 Gb/s all-to-all bandwidth. Maia 100 was first deployed in 2024 for internal Microsoft workloads, including OpenAI inference, and has been positioned as a complement rather than a wholesale replacement for NVIDIA GPUs in Azure.
Cobalt 100 is Microsoft's first custom server CPU, a 128-core 64-bit Arm processor based on the Arm Neoverse Compute Subsystem. The chip delivers up to 40 percent better performance than previous Arm-based Azure VMs and is used for general-purpose Azure workloads as well as for inference on smaller models. Cobalt-based virtual machines went into general availability in late 2024 and have been adopted by customers including Snowflake, Databricks, and Adobe.
In 2025 Microsoft began revealing details of successor chips. Maia 200, expected to enter production in 2026, is reported to use a 3 nm process and HBM3e memory. A separate networking chip codenamed Sirius and additional security and offload silicon round out the Azure custom silicon roadmap. Microsoft has also continued to be one of NVIDIA's largest customers, deploying tens of billions of dollars of Hopper and Blackwell GPUs alongside its own chips.
Microsoft's commercial AI position rests on its ability to spend more on AI infrastructure than almost any other company on Earth. The company's capital expenditure has grown from roughly $24 billion in fiscal 2022 to a projected $110 billion to $120 billion in fiscal 2026, with most of the increase going to AI-capable data centers, GPUs, custom silicon, and the supporting power, cooling, and networking infrastructure.
For fiscal year 2025, which ended June 30, 2025, Microsoft reported total revenue of $245 billion. Revenue from the Intelligent Cloud segment was approximately $105 billion for the full year, including a Q4 figure of $29.9 billion that grew 26 percent year over year. Azure and other cloud services grew 39 percent year over year, with Azure surpassing $75 billion in annual revenue for the fiscal year, up 34 percent. Microsoft has consistently disclosed that AI services contributed several percentage points to Azure's growth rate, and the AI business as a whole crossed an annualized revenue run rate of $13 billion during fiscal 2025 and was reported above $20 billion by the second quarter of fiscal 2026.
In fiscal Q1 2026 (the quarter ending September 30, 2025), Microsoft spent approximately $34.9 billion on capital expenditures, of which $11.1 billion was attributed to data center leases alone. In fiscal Q2 2026 the figure rose to $37.5 billion, putting the company on a roughly $150 billion annualized capex run rate. Nadella told investors that Microsoft expected to expand its AI capacity by more than 80 percent during fiscal 2026 and to roughly double its data center footprint over the following two years.
The scale of capex required has reshaped Microsoft's financial profile. Free cash flow growth has slowed despite revenue acceleration because of the depreciation and operating costs associated with the new infrastructure, and the company has begun issuing significant new debt to finance the buildout. The April 2026 fiscal Q3 release included reporting that Microsoft was offering historically large voluntary buyout packages to long-tenured employees in some legacy product groups as it shifted spending toward AI engineering and operations. Stratechery and other analysts have argued that Microsoft is making a calculated bet that the long-term cash flows from agentic AI workloads will justify what is, in absolute terms, the largest sustained capital expansion in the history of corporate technology.
Microsoft has acquired more than 250 companies in its history. The following table lists those most relevant to the company's modern strategy and to its AI position.
| Year | Company | Price | Strategic role |
|---|---|---|---|
| 1997 | Hotmail | ~$400 million | Web mail, became Outlook.com |
| 2002 | Navision | $1.3 billion | Enterprise resource planning, basis of Dynamics |
| 2007 | aQuantive | $6 billion | Online advertising (largely written off) |
| 2011 | Skype | $8.5 billion | Consumer communications |
| 2012 | Yammer | $1.2 billion | Enterprise social networking |
| 2013 | Nokia Mobile | $7.6 billion | Mobile phones (largely written off) |
| 2016 | $26.2 billion | Professional social network and data | |
| 2018 | GitHub | $7.5 billion | Developer platform; foundation for GitHub Copilot |
| 2020 | ZeniMax (Bethesda) | $8.1 billion | Games (Xbox content) |
| 2021 | Nuance Communications | $19.7 billion | Speech recognition, healthcare AI |
| 2023 | Activision Blizzard | $68.7 billion | Games; closed October 2023 |
| 2024 | Inflection AI (talent and license) | ~$650 million | Founders, staff, and model rights for Microsoft AI |
The Activision Blizzard transaction, completed in October 2023 after a contested antitrust process that saw Microsoft prevail at the U.S. Federal Trade Commission and negotiate concessions with the U.K. Competition and Markets Authority, made Microsoft the third-largest gaming company in the world by revenue. The Nuance acquisition gave Microsoft what is now its core healthcare AI franchise, including the Dragon Medical line of clinical documentation tools and the DAX Copilot ambient documentation product used by hundreds of thousands of clinicians. The GitHub acquisition has been described by analysts including Ben Thompson of Stratechery as Nadella's most prescient pre-AI move, since it gave Microsoft both the developer platform on which GitHub Copilot was eventually built and the data and culture that informed Microsoft's broader Copilot strategy.
Microsoft's AI partnerships and acquisitions have drawn substantial regulatory attention. The European Commission, the U.K. Competition and Markets Authority, and the U.S. Federal Trade Commission have all opened inquiries into the OpenAI partnership and the Inflection transaction. Microsoft surrendered its non-voting observer seat on the OpenAI board in mid-2024 in part to address competition concerns, and the October 2025 OpenAI restructuring was negotiated in part with the offices of the California and Delaware attorneys general.
In the United States, Microsoft was the subject of the landmark United States v. Microsoft Corporation antitrust case in the late 1990s and early 2000s, which it settled in 2001 with a series of conduct remedies. The legacy of that case has shaped Microsoft's approach to subsequent regulatory engagement, with president and vice chair Brad Smith publicly emphasizing the company's commitment to interoperability and to working constructively with regulators. Microsoft was an early signatory of the Voluntary AI Commitments brokered by the Biden administration in July 2023, the EU AI Act compliance commitments, and the Hiroshima Process Code of Conduct.
The AI capex buildout has put substantial pressure on Microsoft's sustainability commitments. The company pledged in 2020 to be carbon negative by 2030, water positive by 2030, and zero waste by 2030. In its 2024 Environmental Sustainability Report, Microsoft acknowledged that its Scope 3 emissions had risen approximately 30 percent since 2020 due to data center construction and the embodied carbon of new chip and rack hardware. To address electricity demand, Microsoft has signed power purchase agreements totaling tens of gigawatts, including a high-profile 20-year agreement with Constellation Energy in 2024 to restart the undamaged Three Mile Island Unit 1 nuclear reactor, renamed the Crane Clean Energy Center, with the entire output going to Microsoft. The company has also signed power purchase agreements for new small modular reactors, large solar and wind farms, and geothermal energy.
Analysts including Ben Thompson of Stratechery, Steven Sinofsky, and the team at The Information have repeatedly characterized Microsoft's AI strategy as a bet on three reinforcing layers: frontier model access (initially through OpenAI, increasingly diversified through Anthropic, Mistral, and Microsoft's own MAI and Phi models); the Azure platform on which those models run; and the application layer (M365 Copilot, GitHub Copilot, Dynamics, and a long tail of Copilot extensions) that monetizes them inside Microsoft's enterprise installed base. The April 2026 sales restructuring, in which Microsoft moved to selling Copilot as a standalone product rather than bundling it into broader software suites, was widely interpreted as a response to Wall Street pressure for clearer AI revenue attribution.
Microsoft's principal competitive risks are widely identified as: a sharper break with OpenAI that could leave the company's flagship Copilot products dependent on internally developed models that have yet to demonstrate parity with the frontier; continued underperformance of Microsoft 365 Copilot adoption inside large enterprises; rising capital intensity that compresses returns on capital even as revenue grows; and regulatory action against the OpenAI partnership or against the bundling of Copilot into Windows and Microsoft 365.
The company's principal advantages are equally clear: an enterprise sales footprint that reaches more than 95 percent of the Fortune 500; an Azure platform that is one of only three hyperscale clouds globally; an installed base of more than 400 million paid Microsoft 365 seats; the developer franchise represented by GitHub, Visual Studio, and TypeScript; and a balance sheet that allows it to spend more on AI infrastructure than any company except possibly Alphabet. As of April 2026 Microsoft remains, by most measures, the company best positioned to capture commercial value from the generative AI transition.