Copilot hits 20M paying users as Microsoft modularizes the product

Microsoft’s Copilot story just changed shape. For two years, the loudest critique was simple — nice demo, but is anyone really paying for this at scale? Microsoft answered that on April 29, saying Microsoft 365 Copilot now has more than 20 million paid enterprise seats, with usage intense enough that weekly engagement has reached Outlook’s level. But the more interesting part is underneath that number: Copilot is being rebuilt less like one chatbot and more like a traffic controller for many models. (microsoft.com) ### Why does 20 million matter? Because this is the first number that makes Copilot look less like an experiment and more like a real software business. Nadella said seat growth accelerated quarter over quarter, and the number of customers with more than 50,000 seats quadrupled year over year. That is not a handful of pilots. That is procurement, rollout, training, and budget approval happening inside giant companies. (microsoft.com) ### Which customers make that believable? The names are the kind that calm enterprise buyers. Microsoft highlighted Bayer, Johnson & Johnson, Mercedes, and Roche as customers with more than 90,000 seats each. It also pointed to Accenture, which now has more than 740,000 seats — Microsoft’s biggest Copilot win so far. Big logos do not (microsoft.com) on it. (microsoft.com) ### Are people actually using it? Microsoft says yes, and this part matters more than raw seat count. Copilot queries per user rose nearly 20% quarter over quarter, and Nadella said weekly engagement is now at the same level as Outlook. Basically, Microsoft is arguing that Copilot is becoming a habit, not shelfware. That claim is hard(microsoft.com)t the “buy a few licenses and see” phase. (microsoft.com) ### So what changed in the product? Agent mode is a big piece of it. Microsoft made agentic capabilities generally available and turned Agent mode into the default experience across Copilot and apps like Word, Excel, and PowerPoint. That means Copilot is no longer just answering prompts. It can take multistep actions inside documents and workflows — closer to delegated work than assisted typing. (techcrunch.com) ### Why is the multi-model move a bigger deal? Because it weakens the idea that enterprise AI will be won by one foundation model. Microsoft has been adding Anthropic models to Copilot Studio, including Claude Sonnet 4 and Claude Opus 4.1, alongside OpenAI models. In some Microsoft 365 Copilot ex(techcrunch.com)reasoning, another for throughput, and switch automatically if policy or availability demands it. (microsoft.com) ### Why would enterprises care about that? Because most companies do not want to bet their workflow stack on one model vendor’s strengths, pricing, or uptime. They want controls. They want admins to decide which models are allowed, where they run, and what happens if one is(microsoft.com)ult model if Anthropic is turned off. That is enterprise software logic, not chatbot logic. (microsoft.com) ### What becomes the moat then? Less raw model brilliance. More orchestration. If multiple frontier models are available inside the same product, the durable advantage shifts to context, permissions, audit trails, compliance, and how smoothly the system moves through Word, (microsoft.com) wants Copilot to live. (microsoft.com) ### Bottom line The 20 million figure is the headline, but the architecture is the real story. Microsoft is turning Copilot into a managed AI layer for work — one that can mix models, enforce policy, and stay embedded in the software companies already use. If that works, Copilot stops being “an AI feature” and starts looking like the control plane for enterprise AI. (microsoft.com)