Mustafa Suleyman walked into Microsoft in 2024 with a clear job description: make the company less dependent on OpenAI. On Thursday, April 2, his team delivered the first tangible proof that they meant it — three production-grade AI models built entirely in-house, available immediately, and priced to undercut the company Microsoft has spent $13 billion backing.

The models are called MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, and they cover three of the most commercially lucrative AI capabilities in enterprise software: speech-to-text, voice generation, and image creation. All three are available now through Microsoft Foundry and a new MAI Playground.

MAI-Transcribe-1 handles speech-to-text across the top 25 most-used languages and runs batch jobs 2.5 times faster than Microsoft's own Azure Fast offering. Cost: $0.36 per hour. MAI-Voice-1 generates 60 seconds of audio in exactly 1 second, preserving speaker identity across long-form content, at $22 per million characters. MAI-Image-2 debuted in the top 3 on the Arena.ai leaderboard — the same benchmark that determines how the world ranks foundation models — starting at $5 per million text tokens and $33 per million image tokens.

These are not experiments. They are production models designed to be sold to the same enterprise customers who are already using OpenAI's Whisper, TTS, and DALL-E through Azure. Microsoft just put competing options on the same shelf.

The team behind this is called MAI Superintelligence, and it is five months old. Suleyman announced it in November 2025. Five months later, it has shipped three foundation models ready for enterprise deployment. That timeline is aggressive by any standard, and someone at the top of Microsoft set that pace on purpose.

The pricing is the tell. Microsoft didn't just build these models — it built them cheaper. At every tier, MAI pricing comes in below equivalent OpenAI offerings available on the same Azure platform. That is not a coincidence. That is a message to enterprise customers, and to OpenAI, about who holds leverage in this relationship.

My Opinion

I'll be direct: this is Microsoft hedging, and it's the smartest thing they've done in AI since the original OpenAI deal. For years, Microsoft's AI strategy was essentially "trust Sam Altman." That worked brilliantly — Azure AI revenue has exploded, Copilot is everywhere, and Microsoft became one of the most AI-exposed large-cap stocks on earth. But betting $13 billion on a single supplier, no matter how good, is not a strategy. It's a dependency.

Here's what bugs me about the coverage of this story: everyone's calling it a "shot at OpenAI" as if it's dramatic. It isn't. It's rational. Every serious technology company builds redundancy into its supply chain. Microsoft was unusually late to realize it needed to do this with AI. The fact that Suleyman's team shipped three production models in five months tells you exactly how much internal urgency there was to close this gap fast.

The relationship between Microsoft and OpenAI isn't over. OpenAI's frontier models remain more capable, and GPT-4o and o3 will stay in Azure for years. But the dynamic has shifted. Microsoft no longer needs OpenAI the way it did in 2023. And OpenAI, currently burning toward a $57 billion annual run rate while chasing a 2027 IPO, absolutely still needs Microsoft's distribution. That asymmetry is now visible. Expect Microsoft to use it.

Author: Yahor Kamarou (Mark) / www.humai.blog / 03 Apr 2026