A worker drafting a quarterly report in Word can now call on a model built by Elon Musk’s xAI without leaving the document. On September 12, Microsoft began rolling out Grok inside Copilot for its enterprise customers, putting the model alongside the OpenAI technology that has powered the assistant since its launch.
The move covers Word, Excel and PowerPoint, the three applications at the center of the Office suite that most of corporate America still runs on. Grok joins Claude, the model from Anthropic that Microsoft added in September 2025, and the OpenAI models that remain the default. Copilot Studio, Microsoft’s tool for building custom agents, had already previewed Grok 4.1 Fast to U.S. organizations earlier. Grok is hosted on Azure AI Foundry, the platform Microsoft built to manage multiple models under one roof.
The significance lies less in any single model than in the change of posture it represents. For the first years of Copilot, Microsoft leaned on a single supplier. Now it runs a shelf of models, each with its own strengths and prices, and lets customers pick. That is a concession to a market reality Microsoft’s own customers have been forcing on it: no single model wins every task, and buyers do not want to be locked to one vendor’s technology.
The OpenAI relationship is not ending. The two companies remain bound by a multibillion-dollar investment and a compute deal that runs through Microsoft’s Azure cloud. What has changed is that the exclusive lane has widened into a multi-lane road. Microsoft executives have described the strategy in plain terms: give customers a choice and let usage data decide which models get promoted.
For enterprise buyers, the choice brings new trade-offs. Each model carries its own cost per token, its own limits on how it handles confidential data, and its own compliance footprint. A company that once had one AI vendor to vet now has three or more, each with a different answer to the question of where data goes and who can see it. That is the governance burden the analyst community has flagged as Copilot’s model list grows.
Grok’s arrival also sharpens the competitive picture. xAI has been building data centers at a pace that rivals the biggest cloud providers, and Musk has positioned the model as an alternative to what he calls the cautious tone of OpenAI and Anthropic. Putting Grok inside the most widely used office software in the world gives it a distribution channel that would otherwise take years to build.
Microsoft’s own interest is straightforward. Copilot competes against Google’s Gemini inside Workspace, and against a wave of standalone AI assistants that live in a browser tab. The more capable models Copilot can offer, the harder it is for a customer to justify paying for a separate tool. Each new model is a reason to stay inside the suite.
The risk is that a wide shelf becomes a confusing one. Analysts said the more models Microsoft adds, the more it has to explain which one a customer should use for which job, and why the answer changes from quarter to quarter. If that explanation gets murky, the choice Microsoft touts as a feature starts to look like a tax on the customer’s attention.
There is also the question of who gets credit. When a model from xAI produces an answer inside Word, the customer may not know or care whose model did the work. But the model vendors care, and Microsoft must divide both the revenue and the credit in a way that keeps each partner invested without letting any one of them become indispensable again.
The backdrop is the long, exclusive courtship between Microsoft and OpenAI. Microsoft has invested about $13 billion in OpenAI since 2019 and built Copilot around the startup’s models when the assistant launched in 2023. That arrangement gave OpenAI a privileged place inside the most widely used productivity software on earth. It also left Microsoft exposed to the pace of OpenAI’s releases, and to the governance dramas that periodically ripple through the startup. Adding Grok and Claude is a hedge against both.
The multi-model approach mirrors what the broader industry is doing. Google runs Gemini inside Workspace but also lets customers use other models, and the enterprise software vendors that once sold one AI partner now sell several. Analysts said Microsoft is following the market rather than leading it, a comfortable position for a company that makes money on every model that runs on Azure, whoever built it.
For now, the addition is another sign that the AI race is being fought inside the applications people already open every day. Copilot began as OpenAI with a Microsoft wrapper. It is becoming a marketplace, and Grok is the newest stall in it.
