In partnership with
Hey everyone,
There is a sentence Satya Nadella has never said in public. But this week, his company said it very loudly through actions.
"We don't need OpenAI anymore."
Officially, the partnership is safe until 2032. Microsoft still owns a chunk of OpenAI. GPT-5.4 still runs Copilot.
But the dynamic just shifted fundamentally.
The context: A company under pressure
Microsoft just had its worst quarter since 2008. The stock is down 24%.
The reason? They are spending nearly $150 billion a year on AI infrastructure, but their flagship product, Copilot, is struggling. Only 3% of commercial Office customers are actually paying for it.
Wall Street is asking hard questions. Microsoft’s answer was surprising. They didn't cut spending. They doubled down on independence.
Ad Break
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
Back to Main Point
The contract that changed
Until six months ago, Microsoft was legally blocked from building "superintelligence" on its own. The deal with OpenAI handcuffed them.
That changed in late 2025. They renegotiated. The new terms? Microsoft can now pursue AGI independently.
This week, we saw the first result.
Three new models
Microsoft launched three foundational models, built entirely in-house by teams of fewer than 10 engineers each.
MAI-Transcribe-1: It beats OpenAI’s Whisper on 25 languages. It is faster and cheaper. It is already running in Teams.
MAI-Voice-1: It generates 60 seconds of audio in one second. It can clone a voice from just 10 seconds of sample audio. This directly challenges startups like ElevenLabs.
MAI-Image-2: It debuted at #3 on the global leaderboard for image generation.
Why this matters
This is about "AI self-sufficiency."
Microsoft is proving they don't want to just distribute OpenAI's tools forever. They are building their own stack.
Mustafa Suleyman, the head of Microsoft AI, was clear. The goal is a frontier LLM by 2027 that replaces external models entirely.
The honest analysis
This is a smart long-term move. It gives Microsoft leverage. It scares competitors who rely on Azure for distribution.
But it doesn't solve their immediate problem.
The issue isn't that Microsoft lacked transcription or voice models. The issue is that users aren't clicking with Copilot.
Building better tech is one thing. Building a product people actually love is another. Microsoft has solved the tech problem this week. They still have a product problem to fix.
The Bottom Line
The "distribution era" of AI is ending. The "production era" has begun. Microsoft isn't just selling AI anymore. They are building it.
Catch you next time,
— Raja Tahoor Ahmad