Microsoft’s annual developer conference, Build 2026, kicked off in San Francisco today with a keynote from CEO Satya Nadella — and the announcements signal something bigger than a product refresh. Microsoft is systematically building its way out of dependence on OpenAI.

The headline release is MAI-Thinking-1, Microsoft’s first dedicated reasoning model. What makes it notable is how it was built: the company says it did not use distillation, meaning the model was not trained on outputs from another AI system. That is a pointed distinction in an industry where many “reasoning” models are trained on outputs from OpenAI’s o-series or Anthropic’s thinking models. Microsoft built this one from scratch.

A Full Suite of Foundation Models

MAI-Thinking-1 is one piece of a larger model release. Microsoft is also unveiling:

• MAI-Image-2.5 and MAI-Image-2.5-Flash — updated vision models for multimodal tasks
• MAI-Voice-2 — a multilingual voice AI model targeting enterprise use cases
• MAI-Transcribe-1.5 — an updated speech-to-text model

The voice and transcription models are particularly relevant for enterprise customers who are deploying AI in customer-facing and internal communication workflows. Better transcription accuracy and multilingual voice generation capabilities directly expand what voice AI agents can do reliably.

Microsoft Scout and the Copilot Super App

Beyond models, Microsoft is previewing its consolidated Copilot experience. The current sprawl of Copilot products — Copilot for Microsoft 365, GitHub Copilot, Copilot Studio — is being brought together into a single interface. The new unified app will include a coding tab (GitHub Copilot), a collaboration tab (Cowork), and an always-on AI agent called Microsoft Scout — a preview version is expected by late summer.

Scout positions itself as a persistent AI agent that works across apps without needing a new prompt for each task. It is Microsoft’s answer to the growing category of autonomous agents that orchestrate work rather than respond to individual questions.

The OpenAI Independence Story

The MAI model family has been building since late 2025. Microsoft’s AI chief Mustafa Suleyman, a DeepMind co-founder, was hired in 2024 and now leads the MAI Superintelligence team. A March 2026 reorganisation shifted him from Copilot oversight to focus exclusively on frontier model development.

The strategic read is straightforward: Microsoft invested heavily in OpenAI but has watched the relationship become more complicated as OpenAI pursued its own enterprise products and direct distribution. Building MAI-Thinking-1 without distillation means Microsoft now has a frontier reasoning model that does not require a licence from, or a revenue share with, OpenAI.

For enterprise buyers, this is meaningful. It means Microsoft Azure customers will have access to reasoning capabilities that are native to the Azure stack, with no third-party dependency for compliance, data residency, or pricing.

What This Means for Business

Reasoning models are going mainstream. Until recently, extended-thinking AI capabilities were niche — expensive, slow, and mostly useful for technical tasks. Microsoft shipping a reasoning model as part of its standard platform means it is moving into everyday enterprise software. Expect to see it embedded in Copilot, in Azure AI Foundry, and in GitHub Copilot for code review and architecture tasks within the next twelve months.

Voice AI infrastructure is maturing fast. MAI-Voice-2 and MAI-Transcribe-1.5 are not standalone products — they are infrastructure that other AI workflows run on top of. Every voice agent, every meeting transcription, every call centre AI depends on this layer being accurate and affordable. Microsoft improving it raises the floor for what voice AI agents can deliver.

The enterprise AI landscape is consolidating around a few major platforms. Between Microsoft’s Build announcements, Google’s Gemini 3.5 family, and Anthropic’s enterprise push, businesses are approaching a moment where the foundational AI infrastructure question gets settled. The real competitive advantage will shift to how well you deploy these models — not which model you choose.

For organisations still figuring out their AI strategy, the question is no longer whether to use AI. The question is how fast to build the internal capabilities and workflows to extract value from tools that are becoming available at every layer of the enterprise stack.

---

Enterprise DNA helps business leaders navigate AI strategy, build AI literacy across teams, and deploy AI agents that deliver measurable operational results. Explore Omni Advisory to discuss your AI roadmap.