Microsoft has invested more than $13 billion in OpenAI. Now it's building products that compete with it.
On 2 April, the tech giant released three new artificial intelligence models developed entirely in-house: MAI-Transcribe-1, a speech-to-text system; MAI-Voice-1, which generates natural-sounding speech from text; and MAI-Image-2, an image generator. All three are available through Microsoft's Azure AI Foundry platform, built by the MAI Superintelligence team led by Mustafa Suleyman, the DeepMind co-founder who became CEO of Microsoft AI in 2024.
The models are impressive on their own terms. MAI-Transcribe-1 supports 25 languages and transcribes audio 2.5 times faster than Microsoft's previous offering, at a starting price of just $0.36 per hour. MAI-Voice-1 can produce 60 seconds of natural audio in a single second of processing. MAI-Image-2 has debuted in the top three on the Arena.ai image generation leaderboard and is already rolling out across Bing, PowerPoint, and Copilot.
But the real story isn't the technology. It's the business strategy.
The $13 billion hedge
Until recently, Microsoft was contractually limited in its ability to build competing AI models — part of the deal that gave it exclusive cloud hosting rights for OpenAI's products. A renegotiation in 2025, coinciding with OpenAI's landmark $110 billion funding round and its march towards an IPO, freed Microsoft to develop its own foundational models while maintaining its role as OpenAI's primary cloud provider.
Forrester analyst Rowan Curran has framed the logic bluntly: as OpenAI moves towards IPO and potentially takes on different priorities, Microsoft needs to ensure it has alternatives it controls.
Dan Ives, Managing Director at Wedbush Securities, called it "the most complex corporate partnership in technology history. Microsoft needs OpenAI for Azure revenue today, but they're simultaneously building the capability to compete with OpenAI tomorrow."
Suleyman himself struck a diplomatic tone. "At Microsoft AI, we're building Humanist AI," he wrote in the launch announcement. "We have a distinct view when creating our AI models — putting humans at the centre, optimising for how people actually communicate, training for practical use."
What it means for Scotland
For Scottish businesses, the most immediately relevant model may be MAI-Transcribe-1. Microsoft has engineered it for resilience across regional accents and challenging audio conditions — conference rooms, call centres, and phone lines with background noise — all scenarios familiar to Scottish enterprises.
Anyone who has battled a voice assistant that cannot tell Dundee from Delhi will appreciate the significance. Scotland's public services, broadcasters, and contact centres have long grappled with speech recognition systems trained predominantly on standard American and southern English accents. A model that genuinely handles accent diversity at enterprise scale could unlock real productivity gains — from automated meeting transcription in council chambers to accurate call analytics in Glasgow's thriving financial services sector.
Independent testing by Artificial Analysis placed MAI-Transcribe-1 fourth globally for transcription accuracy, with a word error rate of 3.0% and processing speed of approximately 69 times real-time — making it the fastest model in the top five by accuracy.
Microsoft is already integrating the model into Teams and Copilot Voice mode, meaning Scottish organisations already using Microsoft 365 could see improved transcription without changing a thing.
What comes next
These three models are specialists — none is a general-purpose large language model to rival GPT or Claude. Suleyman has indicated Microsoft's frontier LLM is 12 to 18 months away, suggesting these releases are the opening move in a much larger campaign.
The financial firepower is certainly there. Microsoft's AI-related capital expenditure hit $37.5 billion in a single quarter at the end of 2025, with analysts projecting up to $120 billion for the full fiscal year. Azure cloud revenue grew 39% year-on-year, with AI services a significant driver.
For now, the message is clear: the company that bet $13 billion on OpenAI is no longer putting all its chips on one table. The question is whether OpenAI — and the rest of the industry — can keep pace.
