Microsoft just put a hard number on the Copilot question that’s been hanging over the AI productivity space. CEO Satya Nadella told investors Wednesday that M365 Copilot has crossed 20 million paid enterprise seats, and that weekly engagement now matches Outlook, according to TechCrunch AI. That’s a direct shot at the lingering narrative that nobody actually uses the tool baked into Word, Excel, and Outlook.
TechCrunch AI reports the announcement came during Microsoft’s quarterly earnings call, where Nadella laid out specifics that go well beyond seat counts. The company has quadrupled the number of customers paying for more than 50,000 seats. Bayer, Johnson & Johnson, Mercedes, and Roche each run more than 90,000. And Accenture just signed for over 740,000 seats, which Nadella called “our largest Copilot win to date.”
The engagement story
Seat counts are easy to inflate. Usage isn’t. Nadella’s pitch was that people are actually opening Copilot, not just paying for it.
- Copilot queries per user up nearly 20% quarter over quarter
- Weekly engagement now at the same level as Outlook
- Agent mode rolled out as the default experience in Copilot, Word, Excel, and PowerPoint last week
“This is like a daily habit of intense usage,” Nadella said on the call. Morgan Stanley analyst Keith Weiss called the numbers “super impressive and I think way ahead of most people’s expectations.”
That last bit matters. Wall Street has been openly skeptical of enterprise AI ROI for most of 2026, with several analysts questioning whether Copilot subscriptions were quietly being shelved after pilots. A 20% jump in per-user queries is the kind of metric that’s hard to fake.
Why this matters
Two things stand out here. First, Microsoft is no longer married to OpenAI for Copilot’s intelligence. Nadella explicitly walked through the multi-model architecture: users get access to multiple models by default, with intelligent auto-routing inside agents. Microsoft 365 now supports Anthropic’s Claude, and Nadella framed the lineup as “critique and counsel” between models to generate better responses.
For anyone tracking the AI vendor landscape, that’s a real shift. A year ago, Copilot was effectively a wrapper around GPT. Today it’s a routing layer with model choice as a feature.
Second, agent mode is now the default. Microsoft made Copilot’s agentic capabilities generally available last week, letting the assistant take multi-step actions directly inside documents. Nadella’s framing: “You now have a new way to delegate and complete work using Copilot.”
This is the same pattern we’re seeing across the productivity stack. Anthropic’s Claude with computer use, Google’s Gemini agents in Workspace, and now Copilot agents as the default in Office apps. The chat box is becoming a fallback. Agentic delegation is the headline experience.
What to expect next
A few practical implications for AI buyers and builders:
- Procurement pressure is shifting. With six-figure seat deals at Bayer, J&J, Mercedes, and Roche, and a 740K-seat Accenture deal as the new ceiling, expect aggressive enterprise discounting and bundling from competitors. Google Workspace AI and Notion will need to respond.
- Model-agnostic Copilot changes the OpenAI calculus. If Microsoft is routing queries to Claude and others by default, OpenAI loses some of its captive distribution inside the Office install base. That’s a meaningful structural change.
- The “AI doesn’t get used” critique is harder to sustain. When weekly Copilot engagement matches Outlook, the burden of proof flips. Skeptics now have to explain why their organization is the exception.
The one caveat: Microsoft picks its metrics carefully. “Weekly engagement” doesn’t tell us session length, task completion, or whether users are running a single query a week or fifty. Expect competitors and analysts to start asking sharper questions about what engagement actually means.
For now, the headline number is hard to argue with. 20 million paid seats, growing usage, and agents as the default. More details at the original TechCrunch AI report.