Microsoft Spent $13B on OpenAI, Then Built Cowork on Claude

Microsoft just shipped Copilot Cowork, the new agent-style feature in M365 Copilot that handles multi-step tasks across your Office apps. It runs on Anthropic’s Claude.

Microsoft — the company that put $13 billion into OpenAI and built its entire Copilot brand on GPT models — chose a competitor’s model for its most ambitious M365 feature.

Enterprise AI is going multi-model by default. Microsoft just made that official.

The reasoning is straightforward. Different models have different strengths. Claude’s extended thinking and instruction-following make it better suited for the kind of multi-step, context-heavy work that Cowork handles — managing projects across Teams, Excel, and Outlook simultaneously. Microsoft apparently decided that shipping a better product mattered more than loyalty to their $13B investment.

The pricing tells its own story. Copilot with Cowork runs $30/user/month. The new M365 E7 tier bundles everything — Copilot, Cowork, Security Copilot, the works — at $99/user/month. These numbers matter for enterprise planning because they set the ceiling on what “AI-augmented productivity” costs per seat. Budget accordingly.

What this means for your AI strategy:

• Single-provider commitments are increasingly risky. If Microsoft won’t go all-in on one model provider, your mid-size company probably shouldn’t either.
• Evaluate by task, not by brand. The right model for customer support might not be the right model for document analysis. Microsoft just proved this at scale.
• Watch the bundling. The E7 tier at $99/month signals where Microsoft thinks the market is heading — AI capabilities as a standard enterprise line item, not an experimental add-on.

The companies that locked themselves into a single AI provider last year are watching Microsoft, of all companies, demonstrate why that was premature.

If the company that owns a chunk of OpenAI is hedging its bets, what makes you confident enough to go all-in on one provider?