Mustafa Suleyman says Microsoft is “not able to build models in the very largest scale yet” but its “computation ramp is coming to enable us to do” it in 2026
Tech giant's AI chief says it will have the resources to build frontier systems later this year
This update makes compute availability the disclosed constraint on that transition: Microsoft’s ambition is not only a research or product question, but a capacity-allocation question.
First-order effects
Microsoft cannot presently train at the largest frontier scale, leaving its near-term proprietary-model roadmap dependent on the promised 2026 compute ramp.
Suleyman’s team gains a clearer internal mandate to turn new capacity into frontier training capability, rather than relying solely on models from external partners.
Second-order effects
Microsoft’s cloud and model partners face a changing balance of leverage as the company pursues in-house capability; self-sufficiency would give Microsoft more options over which models it develops and deploys.
If Microsoft executes on the ramp, frontier-model development becomes more tightly coupled to access to capital-intensive infrastructure, favoring platforms able to finance and operate large compute fleets.
The larger competitive divide may increasingly be between firms that can secure frontier capacity and those that must differentiate through applications, specialized models, or external model providers; the timing and effectiveness of Microsoft’s ramp remain unproven.
The trend: This is one data point in AI industrialization: major platforms are treating frontier-model capability as an infrastructure-and-capacity program, not just a software research effort.
Mustafa Suleyman says renegotiating Microsoft's contract with OpenAI “unlocked [Microsoft's] ability to pursue superintelligence.” Though his new JD at Microsoft AI was only made public last month, he'd been preparing for the transition for 6-9 months. https://www.theverge.com/..…
Talk about moving goalposts. This one from MSFT's Suleyman may well take the cake. “Superintelligence” just went from intelligence beyond all humans to merely “delivering product value”. 🙄
Microsoft can pursue superintelligence independently after OpenAI declares AGI, using AGI's weights as the starting point. Fully automated AI research, combined with the sheer amount of compute available to Microsoft, could suddenly turn it into a frontier lab. [image]
Microsoft's new “superintelligence” game plan is all about business. It's launching a new transcription model today that is a step towards those goals, says Microsoft AI CEO Mustafa Suleyman https://www.theverge.com/...
A lot has changed in a year. April 2025: Suleyman told me there's no need for a frontier model at Microsoft because the reliance on OpenAI helped Microsoft develop AI products more efficiently [embedded post]
Mustafa Suleyman says renegotiating Microsoft's contract with OpenAI “unlocked [Microsoft's] ability to pursue superintelligence.” Though his new job description at Microsoft AI was only made public last month, he'd been preparing for the transition for 6-9 months. www.theverge.…
Three models. Three top-tier results. All shipped within just a few months by the @MicrosoftAI team. - MAI-Transcribe-1 dropped today, the most accurate transcription model in the world across 25 languages according to FLEURS WER benchmark. - MAI-Voice-1 sets a new standard f…
Microsoft just dropped MAI-Transcribe-1, a new SOTA speech-to-text model. The model is built to deliver high quality transcription in messy, real-world environments, while remaining incredibly fast and efficient. MAI-Transcribe-1 delivers SOTA speech-to-text transcription acros…
Developers, developers, developers! Three new models from @MicrosoftAI now in Foundry: speech→text, text→speech, and text→image. Less integration tax. Build agents with voice, captions, call analytics and automate support and creative workflows! @AIFoundryDevs @MSAzureDev
Artists and creators: You can now access a growing set of the most powerful speech tools in the world for the lowest price at @Azure. With the tools, it is up to you how to create or monetise your own ideas. @LarryJackson @1benm @tapmusic You could build something like the
The most accurate model across 25 languages, faster transcription speeds, and stronger performance in real-world noise. MAI-Transcribe-1 sets a new bar for speech recognition. Learn more + try it today: https://microsoft.ai/... [image]
Today, we announced the public preview of MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 on Microsoft Foundry, bringing our first-party AI models directly into the hands of developers. Read more: https://microsoft.ai/... [video]
BREAKING: Mustafa Suleyman just told me Microsoft will build a frontier large language model to compete directly with OpenAI's GPT — and revealed that until October 2025, Microsoft was contractually banned from even trying. 🔥🤖 Full story: https://venturebeat.com/... #AI #Microsof…
We're bringing our growing MAI model family to every developer in Foundry, including ... · MAI-Transcribe-1, most accurate transcription model in world across 25 languages · MAI-Voice-1, natural, expressive speech generation · MAI-Image-2, our most capable image model yet Start […
MSI is the most fun team I've ever worked with. This team ships. This team creates. This team innovates. This team believes in work-life balance, and none of that 70 hour or 996 bullsh*t We must build AI responsibly and sustainably, put users first, put our teams first, put
MAI-Transcribe-1 makes speech-to-text clearer, faster, and more reliable even in noisy audio. Ranked #1 on the industry-standard FLEURS word error rate benchmark. Now in public preview. Learn more: https://microsoft.ai/... [video]
One place MAI-Image-2 really knocks it out of the park is surrealist images. Try this one: Close-up zoomed in macro photo of a bright orange clownfish hiding among stark white peonies with bright yellow stamens. High contrast, shallow depth of field, vibrant wildlife [image]
Been awesome to have MAI-Image-2 out in the world and see people's creations. Wanted to start sharing some favorite prompts the team has come up with so you can test them out for yourself 👀 Will keep adding to this (and share yours too)