What Microsoft is in the AI world
Microsoft is one of the most powerful forces shaping how AI gets built and used — not primarily because it trains the world's most capable models, but because it controls the infrastructure, products, and developer platforms that bring AI to hundreds of millions of people. Think of it as the company that built the roads and the storefronts, and is now also starting to make some of the cars.
Its AI story has three overlapping chapters: a deep partnership with OpenAI, a newer relationship with Anthropic, and a growing push to build its own AI from scratch.
The OpenAI partnership: where it started
In 2019, Microsoft made a $1 billion bet on OpenAI and became its exclusive cloud provider. That meant OpenAI's models — including the GPT series — would be trained and served on Microsoft's Azure cloud. It was a foundational deal that gave Microsoft early access to frontier AI and gave OpenAI the computing power to scale.
That relationship deepened over the years. Today, GPT-5.6 is the default AI model inside Microsoft 365 Copilot — the AI assistant built into Word, Excel, PowerPoint, and Teams. That's one of the largest deployments of a frontier AI model to any single product suite in the world.
But the partnership has also evolved. In April 2026, Microsoft and OpenAI restructured their agreement. OpenAI gained the freedom to sell its models on competing clouds like Amazon Web Services and Google Cloud. Microsoft retained its status as OpenAI's primary cloud partner through 2032 but gave up an exclusive clause that had given it special access to OpenAI's most advanced systems.
The Anthropic relationship: hedging the bet
Even while partnered with OpenAI, Microsoft invested $5 billion in Anthropic — OpenAI's main rival — and committed to purchasing $30 billion of Azure computing capacity for Anthropic's use. Claude models (Sonnet, Opus, and Haiku variants) are now available through Microsoft Foundry, Microsoft's platform for enterprise AI deployment, and Claude has been integrated into Microsoft 365 Copilot's Agent Mode in Excel.
This might seem contradictory — backing two competitors at once — but it reflects a straightforward strategy: Microsoft wants to be the platform where AI runs, regardless of which lab wins the model race.
Building its own: the MAI family
The most significant recent shift is Microsoft's move to build AI models entirely in-house. At Microsoft Build 2026, the company unveiled the MAI family — seven models trained from scratch, not borrowed or adapted from OpenAI.
The flagship is MAI-Thinking-1, a reasoning model with a mixture-of-experts architecture (meaning it activates only part of its total parameters for any given task, making it more efficient). It scored 97.0% on a demanding math benchmark called AIME 2025, placing it among the top reasoning models available. Microsoft also released MAI-Image-2.5, a text-to-image model, and MAI-Transcribe-1.5 for audio.
Why build in-house? Reports indicate Microsoft is actively replacing OpenAI and Anthropic models in some of its own products — Excel, Outlook, Teams — with MAI models, as its discounted access to OpenAI's models is set to expire. Building your own AI is expensive, but it removes dependence on a supplier who is also, increasingly, a competitor.
GitHub Copilot and the developer platform
Microsoft owns GitHub, the platform used by over 100 million developers worldwide. GitHub Copilot — the AI coding assistant built into tools like Visual Studio Code — has become one of the most widely used AI products in the world. It has served as a distribution channel for multiple AI models, including Claude and GPT variants, and Microsoft has expanded it with a desktop app designed to manage multiple AI agents running in parallel.
Microsoft also co-developed the Agentic Resource Discovery (ARD) open specification with Google and others, a standard for how AI agents discover and use tools at runtime — a sign that Microsoft is thinking about the plumbing of the next generation of AI, not just the models themselves.
Safety, regulation, and the bigger picture
Microsoft has agreed to submit its models to NIST's TRAINS task force — a new U.S. government program that evaluates frontier AI for national security risks before public deployment. It is also a member of Project Glasswing, Anthropic's consortium for proactively patching software vulnerabilities discovered by AI. And it is co-developing an industry-wide jailbreak severity framework with Anthropic, Amazon, and Google.
Like other major tech companies, Microsoft has acknowledged that its surging AI infrastructure needs are causing it to miss earlier climate commitments, with data center energy consumption rising sharply.
Where it's heading
Microsoft's trajectory points toward becoming an AI platform company that is less dependent on any single model provider. It is building its own models, distributing models from multiple labs, and embedding AI into the productivity software used by most of the world's knowledge workers. The OpenAI partnership remains central — but it is no longer the whole story.
