Microsoft has announced its first fully homegrown artificial intelligence models, marking a significant step in its long-term AI strategy and signalling a move to reduce reliance on OpenAI.

The two new systems, MAI-Voice-1 and MAI-1-preview, were unveiled by Microsoft AI (MAI) in a blog post as part of what the company describes as a “big ambition” to build specialised models optimised for consumer use.

MAI-Voice-1 is a speech generation model capable of producing a full minute of audio in under a second on a single GPU. Microsoft says this makes it one of the most efficient voice systems available today.

The model is already powering Copilot Daily, where an AI host narrates news updates, and podcast-style discussions designed to explain complex topics. It is also integrated into Copilot Labs, where users can customise the model’s speaking style and test scenarios such as storytelling or guided meditations.

“Voice is the interface of the future for AI companions,” Microsoft AI wrote in its blog. “MAI-Voice-1 delivers high-fidelity, expressive audio across both single and multi-speaker scenarios.”

Excited to share our first @MicrosoftAI in-house models: MAI-Voice-1 and MAI-1-preview. Details and how you can test below, with lots more to come⬇️ pic.twitter.com/LtL2YmzuTv

— Mustafa Suleyman (@mustafasuleyman) August 28, 2025

Text-based Foundation Model
Alongside its speech system, the company introduced MAI-1-preview, its first in-house text foundation model trained end-to-end.

The model was developed using around 15,000 Nvidia H100 GPUs, a figure far below some rivals such as xAI’s Grok, which used over 100,000 GPUs.

Microsoft says this efficiency-first approach is intended to prioritise high-quality data while avoiding unnecessary compute costs – a theme repeatedly emphasised by Mustafa Suleyman, head of Microsoft AI.

MAI-1-preview is now undergoing public testing on LMArena, a community benchmarking platform. The model will be rolled out to limited Copilot use cases in the coming weeks, with feedback from users and trusted testers shaping its development.

A Step Change in Microsoft’s Long-term AI Strategy?
Although Microsoft Copilot continues to rely heavily on OpenAI’s GPT models, the introduction of MAI-Voice-1 and MAI-1-preview highlights the company’s intent to gain greater independence in AI.

In its announcement, Microsoft said: “Not only will we pursue further advances here, but we believe that orchestrating a range of specialised models serving different user intents and use cases will unlock immense value.”

Recommended reading

• Scots Startup Wordsmith AI Raises $25M Series A
• Scots Legal Tech Firm Raises $5M Seed Funding
• New AI Guide Launched for Scottish Solicitors

Suleyman added that efficient training is crucial for sustainability at a time when industry observers are questioning the long-term costs of AI development.

Microsoft says it will continue to develop a multi-model strategy, combining different systems to address a wide variety of user needs rather than focusing solely on building larger models.

The company’s AI division described its latest work as “the first steps to making this vision a reality”, suggesting further updates will follow as its in-house portfolio expands.