Microsoft's Moat and Claude Sheets: Copilot Is the Target

When Anthropic launched Claude Sheets and Whiteboards on November 10, the headlines read it as a run at Office (Bloomberg: "Taking on Microsoft"). We don't think it dents Microsoft's moat. Claude is going after Copilot, the AI upgrade Microsoft wants every Office customer to pay for. That's a risk to Microsoft's growth. The things that lock companies into Microsoft, their logins, mail and files, are out of a beta spreadsheet's reach.

Anthropic's own post says the sheets have "simple spreadsheet functionality" for now. When work outgrows that, it points users to export to .xlsx or CSV, or to Claude for Excel, Anthropic's add-in that runs inside Microsoft's app. Claude Docs and Slides came out in September. It's a real suite and it overlaps Office. It's also a first version with no Excel import, no price of its own and no usage figures.

A company doesn't run on spreadsheets. It runs on Entra ID deciding who can sign in, Exchange holding the mail, SharePoint and OneDrive holding the files, and Teams holding the calls. A sheet made in Claude still has to be sent to a supplier who opens it in Excel, saved where compliance can find it, and opened by someone signed in with a Microsoft account. The grid below rates each company on three sources of lock-in. Strong means few rivals can match it, intact means it's real but others have it too, and weakened means it's eroding.

Microsoft is rated strong on the record and the bundle. Anthropic has no record of a customer's data to rate at all, which is the gap: Claude works on top of other companies' files. Google is the useful comparison, because Workspace overlaps Claude Sheets just as Office does and Google is also one of Anthropic's biggest compute suppliers. The software more exposed to Claude is the kind whose moat is the screen itself. Adobe's interface pillar is already rated weakened. Figma's is still intact but marked as narrowing, because Claude Design lets marketers and founders describe a prototype instead of learning the tool. Adobe and Atlassian also now open inside Claude, which makes them destinations as much as targets.

Microsoft and Anthropic are also tied together, in ways that cut both ways.

The strongest bear case doesn't need Office to die. It needs the new money to go elsewhere. Microsoft's plan for the next five years is that every Office seat gets an AI upgrade: Copilot, then the E7 premium tier and pricing that charges for usage on top of the seat. If workers start their day in a Claude window, Office becomes the place files get saved and the AI upgrade money goes to Anthropic. The bundle would survive and the growth wouldn't, much as Windows kept the desktop while the money moved to phones. The Microsoft stock page counts Copilot as part of the bundle, so splitting growth from moat this way is our reading, not the framework's.

Some of that shows already. Microsoft disclosed over 30 million paid Copilot seats in July, then on the October call said it wasn't updating the number, while E7 had "millions of paid seats". On the same call, paid Microsoft 365 commercial seats grew 5%, down from 6% to 7% through fiscal 2026 on the new definition, and finance chief Amy Hood blamed "weaker sales execution" in E3 and standalone plans. And the quarter's lift came from GitHub Copilot, which competes with Claude Code.

Our answer is that none of this is customers leaving. Revenue growth sped up in the same quarter, which is what a price-and-mix gain on a stable base looks like. Slower seat growth in a market where most office workers already have a seat is a limit on growth. A falling seat count would be a crack in the moat, and the two lines below are what we'd watch for it.

Since September, Microsoft counts GitHub and Security Copilot inside this revenue line and paid GitHub users inside the seat count, and it restated fiscal 2026 to match. Some of the recent speed-up is GitHub Copilot. We'll check both lines at the fiscal Q2 2027 report in late January, along with whether Microsoft gives a Copilot seat count again.