MICROSOFT BUILDS SEVEN AI MODELS FROM SCRATCH AND SAYS ONE ALREADY BEATS OPENAI AT A TENTH OF THE COST Microsoft stood at its Build 2026 developer conference and told the world it no longer needs OpenAI to compete in artificial intelligence. The company’s new Microsoft AI Superintelligence team unveiled seven homegrown models, all trained from scratch on commercially licensed data with no distillation from other companies’ work. The flagship is MAI-Thinking-1, Microsoft’s first reasoning model, designed to compete directly with OpenAI’s o-series and Anthropic’s extended-thinking offerings. Alongside it came MAI-Image-2.5 for text-to-image generation, which landed at number three on the Arena AI leaderboard. MAI-Code-1 targets software development inside GitHub Copilot and VS Code. MAI-Transcribe-1.5 covers 43 languages with streaming on the way. MAI-Voice-2 brings expanded language options and new voice profiles. The commercial proof of concept came from McKinsey. Microsoft AI chief Mustafa Suleiman said the MAI models outperformed OpenAI’s GPT-5.5 on McKinsey’s enterprise workflows at ten times lower cost. That is not a rounding error. That is a business model shift. Microsoft framed the move as a bid for what Suleiman called long-term self-sufficiency. After years of writing checks to OpenAI and repackaging the results, the company is betting it can build better, faster, and cheaper on its own. The implications for OpenAI’s revenue concentration are not subtle. Keywords: Microsoft MAI models, MAI-Thinking-1, Microsoft AI Build 2026, Microsoft OpenAI competition