Microsoft AI

Microsoft AI (abbreviated MAI) is the artificial intelligence research and product division of Microsoft, established in 2024 and led by Mustafa Suleyman. It develops Microsoft's own MAI foundation models alongside consumer AI experiences including Microsoft Copilot, AI-enhanced Bing search and Windows on-device AI features.^[1]

History

Microsoft created the division in 2024 to concentrate its consumer AI work — including Bing, Edge and the Copilot app — under a single leadership structure, appointing Mustafa Suleyman, co-founder of DeepMind and Inflection AI, to lead it. While Microsoft's large-scale partnership with OpenAI continued in parallel, the new unit was tasked with building models in-house as well.^[1]

In 2025 the division released its first self-developed models under the MAI prefix, and by 2026 the MAI family had expanded considerably: a Build 2026 announcement unveiled seven new MAI models spanning language, speech and transcription, and the division published MAI-Thinking-1, a reasoning model trained from scratch by Microsoft AI. In October 2026 Microsoft AI released MAI-Voice-2.1, MAI-Voice-2.1-Flash and MAI-Transcribe-2-Streaming for speech and audio workloads.^[2][3]

Key Concepts and Technology

The MAI line reflects a strategy of vertical integration: rather than depending solely on a single external model provider, Microsoft develops models matched to its own product surfaces — voice interaction in Copilot, transcription in Teams and Outlook, on-device inference in Windows. The division's speech models are optimised for low latency, with the Flash tier positioned for high-volume, cost-sensitive use. Microsoft has also published internal usage policies for its models, requiring them to follow user objectives rather than pursue independently generated goals — an alignment posture consistent with broader industry practice.^[4]

Alongside model development, Microsoft's infrastructure arm operates the Azure Maia accelerator programme for AI training and inference, and Azure remains the cloud foundation for both Microsoft's own models and the OpenAI services it hosts.^[5]

Applications and Impact

Microsoft AI's products reach consumers through Copilot in Windows, Microsoft 365 and mobile apps; enterprises through Azure-hosted model APIs; and developers through the company's model catalogue. The existence of a capable in-house model family gives Microsoft negotiating leverage and redundancy in a market where frontier model supply is concentrated among a handful of labs, and it positions speech and transcription — interfaces with broad consumer reach — as a first battleground for MAI models.^[2]

>See Also

• Microsoft Copilot
• OpenAI
• Azure AI
• Google DeepMind
• Text-to-Speech