Microsoft's MAI Era Begins: Production Excel and Outlook Traffic Moves to First-Party Models

What Changed

Microsoft’s first-party model transition:

• Excel: Production traffic now routed to MAI models
• Outlook: Production traffic now routed to MAI models
• GitHub Copilot: Moving to MAI-Code-1-Flash
• Scale: Millions of enterprise users affected

The MAI (Microsoft AI) model family has been in development for years. Microsoft has invested billions in its own models, including hiring former Inflection AI leadership and researchers.

Why Now

Several factors drove the transition:

• Cost control: First-party models are cheaper than OpenAI API fees
• Model maturity: MAI models have reached production quality
• Strategic independence: Reducing reliance on a single vendor
• OpenAI’s turbulence: The Hugging Face incident and safety concerns accelerated plans
• Competitive positioning: Microsoft wants its own AI differentiation

Microsoft’s FY2026 Q4 results showed a $3.2 billion gain from its Anthropic investment against a $600 million write-down on OpenAI — evidence of the company’s hedging strategy across multiple AI labs.

MAI-Code-1-Flash

The coding model moving into GitHub Copilot:

• MAI-Code-1-Flash: A first-party coding model optimized for speed
• Positioning: Competitive with frontier coding models at lower cost
• Scale: GitHub Copilot has tens of millions of users

Implications

• OpenAI dependence: Microsoft’s shift reduces OpenAI’s largest revenue stream
• Model market: First-party models compete directly with frontier labs
• Enterprise AI: Microsoft is betting its models are good enough for its own products
• Industry signal: Every major platform is now building or buying alternatives to frontier labs

For the AI industry, Microsoft’s MAI transition is a watershed: the company that made OpenAI a household name is now confident enough in its own models to use them for its most important products.