Microsoft

Updated June 2026

Microsoft is the enterprise AI platform leader, distributing AI through Copilot across Microsoft 365, GitHub and Azure while hedging across three model sources: a roughly 27% stake in OpenAI, Anthropic's Claude inside Copilot, and its own in-house MAI models.

Website →API →Pricing →

Microsoft is the world’s largest enterprise AI distributor, reaching users through Microsoft Copilot across Microsoft 365, Windows, GitHub and Azure. Its defining 2026 move is a three-way model hedge: Microsoft holds a stake in OpenAI valued at roughly $135 billion (about 27% on an as-converted basis), it has embedded Anthropic’s Claude inside Copilot, and it is now shipping its own in-house MAI models built by Microsoft AI. No other company sells frontier AI from three rival sources at once. Microsoft trades on Nasdaq as MSFT with a market capitalisation of around $2.8 trillion as of mid-June 2026, making it the world’s fourth most valuable public company (companiesmarketcap).

The strategic story is that Microsoft is loosening its dependence on OpenAI while keeping the relationship. An October 2025 recapitalisation, finalised in 2026, removed the contract’s AGI exit clause, let OpenAI use rival clouds, and — in Microsoft AI chief Mustafa Suleyman’s words — “set free” his division to pursue superintelligence with its own researchers and silicon, though he stressed it is “very early days” (VentureBeat).

Quick facts

History and founding

Microsoft was founded on 4 April 1975 by Bill Gates and Paul Allen in Albuquerque, New Mexico, and built its dominance on MS-DOS, Windows and Office. Satya Nadella became chief executive in 2014 and chairman in 2021, refocusing the company on cloud (Azure) and, from 2019, on AI through its partnership with OpenAI. Microsoft’s largest acquisitions shaped its current platform: LinkedIn for $26.2 billion (2016), GitHub for $7.5 billion (2018), Nuance for $19.7 billion (2022), and Activision Blizzard for roughly $69 billion (2023). In 2024 Microsoft hired most of the team from Inflection AI, including co-founder Mustafa Suleyman, to lead a new consumer-focused Microsoft AI division — the group now building the MAI models. Microsoft marked its 51st year in 2026.

Funding and partnerships

Microsoft has been self-funding since its 1986 IPO; the relevant “funding” story today is its investment in partners and infrastructure, not capital raised.

OpenAI (the central relationship). Microsoft first invested $1 billion in OpenAI in 2019 and expanded to a cumulative total of about $13 billion over subsequent rounds. A recapitalisation announced in October 2025 and carried into 2026 reset the terms (OpenAI, CNBC):

• Microsoft holds an investment in OpenAI Group PBC valued at approximately $135 billion, about 27% on an as-converted, diluted basis.
• Microsoft keeps a 20% revenue share from OpenAI through 2030, now subject to a total cap and independent of OpenAI’s technical progress.
• Microsoft’s licence to OpenAI’s models and products runs through 2032.
• The AGI clause was removed — the deal no longer dissolves if OpenAI declares it has reached artificial general intelligence.
• OpenAI can now serve its products on any cloud, though Microsoft remains its primary cloud partner and OpenAI ships first on Azure in most cases.

Anthropic (the new addition). Microsoft added Anthropic’s Claude to Copilot Studio in September 2025 and made Anthropic an official subprocessor on 6 January 2026, so Copilot Studio now offers Claude Sonnet 4 and Claude Opus 4.1 alongside OpenAI’s models (Microsoft Copilot blog). In March 2026, Microsoft launched Copilot Cowork, a multi-step M365 agent, built on Anthropic’s Claude rather than OpenAI’s models (Winbuzzer). In the EU, UK and EFTA, admin opt-in is still required to enable Anthropic models.

Infrastructure. Microsoft guided to roughly $190 billion of capital expenditure in FY2026, up 61% year on year, the bulk of it AI data-centre buildout (CNBC).

Product and model launch timeline

The flagship in-house model, MAI-Thinking-1, is a sparse mixture-of-experts reasoning model with roughly 1 trillion total parameters, about 35 billion active per token, and a context window up to 256,000 tokens (GeekWire). The seven-model Build 2026 family spans reasoning, code (MAI-Code-1, with a 5-billion-parameter MAI-Code-1 Flash live in VS Code and GitHub Copilot), image generation (MAI-Image-2.5), transcription (MAI-Transcribe-1.5, covering 43 languages) and speech (MAI-Voice-2, covering 15 languages). Suleyman calls them “the basic building blocks of a superintelligence” and frames the goal as “long-term self-sufficiency,” positioning Microsoft to become a self-sufficient frontier AI lab over time (eWeek).

Leadership

• Satya Nadella — Chairman and Chief Executive Officer; CEO since 2014, chairman since 2021.
• Mustafa Suleyman — CEO of Microsoft AI; DeepMind and Inflection AI co-founder; leads Copilot and the MAI model programme.
• Kevin Scott — Chief Technology Officer and EVP of AI; architect of the original OpenAI partnership.
• Amy Hood — Executive Vice President and Chief Financial Officer.
• Scott Guthrie — Executive Vice President, Cloud and AI Group (Azure).

Business and financials

Microsoft reported Q3 FY2026 revenue of $82.9 billion, up 18% year on year, beating estimates, with EPS of $4.27 (CNBC). The AI story sits inside Azure: Azure and other cloud services grew 40%, Azure passed $75 billion in annual revenue in FY2025 and is heading toward a $100 billion run rate by mid-2026, and Microsoft’s AI business surpassed a $37 billion annual run rate, up 123% year on year. Commercial remaining performance obligations — contracted future revenue — reached $627 billion, up 99%, an unusually large forward-demand signal (Futurum).

On distribution, Microsoft reported 420 million monthly active Copilot users in Q1 2026, up from 230 million a year earlier, and over 20 million paid Microsoft 365 Copilot seats as of Q3 FY2026. Adoption is broad but shallow: only about 4% of Microsoft’s 450 million Microsoft 365 commercial subscribers pay for the Copilot add-on, and third-party analysis puts active-usage conversion at roughly 36% of licensed users, versus around 83% for ChatGPT (Stackmatix). More than 120,000 custom Copilot agents had been deployed by enterprises by Q1 2026, and 64% of Fortune 500 companies have active Copilot deployments.

Competition and market position

Microsoft competes on two fronts. In cloud, Azure is the clear number two to Amazon Web Services and ahead of Google Cloud, and its 40% growth is the fastest of the three hyperscalers. In AI products, Copilot competes with Google’s Gemini-in-Workspace and with OpenAI’s ChatGPT Enterprise; Microsoft’s edge is distribution — Copilot rides an installed base of 450 million commercial Microsoft 365 seats that no rival can match.

Microsoft’s distinctive position is being multi-model by default. It is simultaneously OpenAI’s largest backer, a major Anthropic customer inside Copilot, and a builder of its own MAI models. That hedge reduces single-vendor risk, but it also means Microsoft does not yet own a frontier-leading model: on independent coding benchmarks the models doing the heavy lifting inside Copilot are OpenAI’s GPT-5.x and Anthropic’s Claude Opus 4.8, not MAI. See best AI for coding for where those models rank.

Controversies

• The OpenAI tension. Microsoft is OpenAI’s biggest investor while building rival MAI models and embedding Anthropic — a “frenemy” dynamic the 2026 recapitalisation managed but did not end. The removal of the AGI clause and the revenue-share cap were both concessions reflecting that friction (Spyglass).
• Capex and the AI-spend debate. The ~$190 billion FY2026 capex plan, up 61%, weighed on the stock even as revenue beat, reviving questions about AI return on investment (Global Data Center Hub).
• Layoffs amid record profits. On 23 April 2026 Microsoft offered voluntary retirement to about 8,750 US employees (a “Rule of 70” buyout, a first in its 51-year history) after a March 2026 hiring freeze; AI and Copilot teams were exempt, and the CFO signalled further headcount declines (Fortune).
• Security incident. In June 2026, attackers compromised Microsoft open-source developer tools to steal credentials from AI developers, a supply-chain attack reported by TechCrunch.
• The adoption gap. Critics note the distance between 420 million Copilot users and the much smaller share who actively rely on it, arguing Copilot’s reach is wider than its engagement.

Recent developments (2026)

• Seven MAI models at Build 2026 mark Microsoft’s most serious bid for model self-sufficiency, led by the MAI-Thinking-1 reasoning model.
• Copilot Cowork (March 2026) became the first flagship M365 agent powered by Anthropic’s Claude rather than OpenAI.
• The OpenAI recapitalisation closed the AGI clause, capped the revenue share, and freed both sides to work across clouds.
• Microsoft began moving its own engineers from Claude Code to GitHub Copilot CLI in May 2026, a notable internal dogfooding signal (Winbuzzer).

Where Microsoft excels

• Distribution. Copilot reaches 450 million commercial Microsoft 365 seats and 420 million monthly active users — unmatched enterprise reach.
• Model optionality. Customers can route Copilot and Copilot Studio agents to OpenAI, Anthropic or MAI models, reducing lock-in to any single lab.
• Developer surface. GitHub, GitHub Copilot, Azure AI Foundry and Copilot Studio give Microsoft the broadest AI developer footprint.
• Capital and infrastructure. ~$190 billion of FY2026 capex and a $627 billion order book fund an AI buildout few can match.

Where Microsoft falls short

• No frontier-leading model of its own. MAI is promising but not yet competitive with the best OpenAI, Anthropic or Google models on independent benchmarks; Copilot’s quality still depends on partners.
• Shallow engagement. Roughly a third of licensed Copilot users are active, far below ChatGPT’s conversion, and only about 4% of M365 commercial seats pay for Copilot.
• Strategic dependence. Despite the hedge, Microsoft’s most-used Copilot experiences still lean on OpenAI and Anthropic models it does not own.
• Cost scrutiny. The scale of AI capex has made investors more demanding about returns.

Developer resources

Microsoft’s AI developer stack centres on Azure AI Foundry (model catalogue, evaluation and deployment, hosting OpenAI, Anthropic, Meta and MAI models), Azure OpenAI Service, GitHub and GitHub Copilot, and Copilot Studio for building custom agents. Microsoft Research also publishes the open Phi family of small models. Tooling includes Visual Studio, VS Code, the Semantic Kernel and AutoGen agent frameworks, and the Model Context Protocol support now common across Copilot surfaces.

Frequently asked questions

Does Microsoft own OpenAI?

No. Microsoft holds an investment in OpenAI Group PBC valued at about $135 billion, roughly 27% on an as-converted, diluted basis, after the 2025–2026 recapitalisation. It is OpenAI’s largest investor and primary cloud partner but does not control the company, and the previous AGI exit clause has been removed.

What AI models does Microsoft Copilot use?

Microsoft Copilot is multi-model. It runs on OpenAI’s GPT-5.x models by default, can use Anthropic’s Claude (Sonnet 4 and Opus 4.1) in Copilot Studio and in features such as Copilot Cowork, and increasingly uses Microsoft’s own MAI models. Administrators in the EU, UK and EFTA must opt in to enable Anthropic models.

What are Microsoft’s MAI models?

MAI is Microsoft’s in-house model family, built by the Microsoft AI division under Mustafa Suleyman. At Build 2026 Microsoft unveiled seven MAI models spanning reasoning (MAI-Thinking-1, a ~1-trillion-parameter mixture-of-experts model with ~35 billion active per token and 256K context), code (MAI-Code-1), image generation (MAI-Image-2.5), transcription (MAI-Transcribe-1.5) and speech (MAI-Voice-2, 15 languages). Microsoft frames them as building blocks toward “humanist superintelligence” and long-term independence from OpenAI.

Is Microsoft bigger than OpenAI, Google or Anthropic in AI?

Microsoft has the largest AI distribution and a $37 billion-plus AI revenue run rate, but it does not yet have a frontier-leading model of its own; the strongest models inside Copilot are made by OpenAI and Anthropic. Google leads on owning a full frontier stack, and Anthropic and OpenAI lead on model capability, while Microsoft leads on enterprise reach.

How much is Microsoft worth?

Microsoft’s market capitalisation was around $2.8 trillion in mid-June 2026, making it the world’s fourth most valuable public company. It reported Q3 FY2026 revenue of $82.9 billion, up 18% year on year, with Azure growing 40%.

How many people use Microsoft Copilot?

Microsoft reported 420 million monthly active Copilot users in Q1 2026 and over 20 million paid Microsoft 365 Copilot seats. Active engagement is lower than reach: third-party analysis estimates only about 36% of licensed users are active, versus roughly 83% for ChatGPT.

What is Copilot Cowork?

Copilot Cowork is a Microsoft 365 agent, launched in March 2026, that carries out multi-step tasks across Microsoft 365 on a user’s behalf. It is built on Anthropic’s Claude — the same Claude Cowork engine Anthropic released to consumers — rather than on OpenAI’s models, making it the clearest example of Microsoft’s multi-model strategy.

Apps

• Microsoft Copilot