After the split with OpenAI, Microsoft AI is waiting for its visionary leader.

Who is the least influential key figure among the AI giants in the United States? The answer is likely to be Mustafa Suleyman, the CEO of Microsoft AI.

Take social media X as an example. Sam Altman of OpenAI undoubtedly has the highest number of followers, exceeding 5 million. He uses X as a free marketing platform for ChatGPT and OpenAI. A single post can get tens of millions of views. The post saying "Codex has over 1 million active users" has triggered more than 10,000 shares and responses.

Demis Hassabis of Google ranks second, with over 1 million users following him. However, he has always kept a low - key profile on social media and posts very infrequently. The last high - profile post was about the release of Gemini 1.0 in 2023. People pay attention to and expect more from him because of his authoritative status in AI academia and the development of DeepMind.

Dario Amodei of Anthropic is also very low - key. He doesn't use X as a social media platform at all and has only updated a few posts, directing readers to his official blog website. Although he has only 375,000 followers on social media, it doesn't prevent him from leading Anthropic to become the world's top - valued artificial intelligence startup.

In contrast, Mustafa Suleyman, ranked fourth, is very much like Microsoft's current position in the AI industry.

Suleyman is the most active on social media X, but the spread effect and scale of his content are obviously average. Two years ago, as one of the co - founders of DeepMind, Suleyman joined Microsoft as the CEO of Microsoft AI, taking full charge of Microsoft's consumer - end AI products and reporting directly to Microsoft CEO Satya Nadella.

In April this year, Microsoft and OpenAI ended their exclusive cooperation and officially adjusted to a non - exclusive partnership. Forty days before this happened, Suleyman stepped down from his position as the overall leader of Microsoft AI and focused on forming the MAI Super Intelligence Team internally, spending more energy on building independent and cutting - edge model development.

It can be said that Suleyman is an important promoter of Microsoft's "de - OpenAI" strategy and is also the soul figure who will carry Microsoft's AI strategy in the future under Nadella's halo.

"Our goal is to prove that we can rank among the top four in the world. There are three truly important labs: Google DeepMind, OpenAI, and Anthropic. We are not one of them yet, and this has always been my goal," Suleyman said after the recent Microsoft Build Conference.

To become the fourth in the industry, there must be a corresponding soul figure.

To become a soul figure in the AI industry, one is destined to have the first - principle as a strong belief, to be created by product and technology achievements, and to prove oneself through the tempering and polishing of time. For Suleyman, a more profound question is whether Microsoft's AI products are sufficient to make anyone a soul figure?

Humanistic Super Intelligence?

At the Microsoft Build 2026 Conference, Microsoft presented a series of products, including super apps, internal inference models, network security tools, and an AI assistant similar to OpenClaw. Especially in terms of models, it launched a flagship inference model, MAI - Thinking - 1, and six new models focusing on image, voice, transcription, and coding.

MAI - Thinking - 1 is Microsoft's first fully self - developed inference model, with 3.5 billion active parameters, an MoE architecture, a 256K context window, and is trained from scratch without any data distillation from OpenAI.

One month before, Microsoft and OpenAI ended their cooperation. Microsoft gave up the exclusive authorization of OpenAI's intellectual property rights and lifted the obligation to pay revenue shares to it. In return, it got an IP usage right until 2032 and OpenAI's continuous commitment to the Azure platform.

Now, Microsoft has to rely on itself. As the person in charge of Microsoft AI, Suleyman particularly emphasized when introducing Microsoft's self - developed inference model that "it is trained from scratch and designed for serious mathematical operations, coding, and actual enterprise deployment."

In its product introduction, MAI - Thinking - 1 is at the top level among models of the same level. In key software engineering benchmark tests, its performance is comparable to that of leading models. In blind - test comparative evaluations, its performance in terms of human preference has reached the same level as Sonnet 4.6.

Even though Microsoft emphasizes the advantages of this model in terms of performance and price, it still lags behind its peers OpenAI and Anthropic. Interestingly, Suleyman admits this gap. He publicly said, "We have closed a huge gap in six months."

The most obvious thing at the Microsoft Build Conference is the sprint towards the enterprise - level market occupied by Anthropic.

Scout, an intelligent agent for online office, is an intelligent assistant built on the open - source artificial intelligence platform OpenClaw. It can be used with Microsoft's entire office suite to help enterprises complete tasks such as managing schedules, expense reimbursement, and writing emails.

Suleyman publicly told the media that Microsoft is more concerned about the enterprise developer market that Anthropic focuses on and is less concerned about the consumer - oriented routes of Google, Meta, and OpenAI.

The label he created for the Microsoft AI brand is "humanistic super intelligence," which means putting people first. The intention is to dispel the outside world's resistance and fear of Microsoft's AI strategy in the current public opinion environment full of voices against AI. Under this label, the three application areas that Microsoft AI values more in the long run are AI assistants, medical super intelligence, and sufficient clean energy.

However, the current competition is more urgent. Microsoft's flagship model is still in the internal testing stage and is only open to some enterprises on Microsoft Foundry. There is still a long way to go before large - scale commercial deployment, which is also the fundamental reason why Microsoft AI is not yet able to generate a strong narrative.

In the past two years with Suleyman, Microsoft AI can hardly be called a success

During the two years when Suleyman was in charge of Microsoft's AI business, the weekly active users of its dialogue product Copilot have been hovering around 20 million, basically making no progress.

This is not entirely his responsibility. In 2024, Suleyman joined Microsoft. Microsoft spent $650 million to recruit him and the core team of Inflection AI. Suleyman was in charge of Microsoft AI and took over all consumer products such as Copilot, Bing, and Edge.

However, there has always been a constraint, which is the binding agreement between Microsoft and OpenAI. It clearly prohibits Microsoft from training models exceeding a specific computing power threshold, which blocks the possibility of Microsoft's self - developed large - scale models. So Suleyman later told Fortune magazine that this is a major limitation, especially for a company of Microsoft's size.

One year later, Suleyman released a major upgrade of Copilot: memory function, personalized companion positioning, real - time screen analysis, shopping, and other functions, enabling Copilot to complete tasks such as booking tickets on behalf of users.

However, in fact, these functions are technologies that ChatGPT and Google released a year ago. Microsoft Copilot is just filling in the functions that its competitors have already achieved. These homogeneous updates naturally cannot support the narrative of user growth, and it is difficult to impress users.

Half a year later, Google's Gemini 3.0 and ChatGPT were in a fierce competition, and OpenAI had to issue a red alert.

However, Suleyman triggered an online storm. His remarks on social media caused a lot of doubts and opposition. He said, "Oh, so many cynical people! Every time I hear someone say that artificial intelligence is not good, I find it funny."

"I grew up playing Snake on a Nokia phone! Now we can have smooth conversations with super - intelligent artificial intelligence and generate any images/videos. It's really beyond my comprehension that some people don't think much of it."

Many users oppose Suleyman's remarks not because they don't recognize AI technology, but because Microsoft has stuffed immature AI dialogue technology into computer systems and office suite products. People oppose issues such as the inability to opt - out, privacy, and data collection.

At that time, some media believed that this exposed a serious disconnect between Microsoft's product concept and users' actual demands, and it also exposed Suleyman's deep - seated dilemma at Microsoft: he believes in the product, but the product has not reached the level he believes in.

During this period, Suleyman had to be anxious about the business. Two weeks before, he became the head of Microsoft's MAI Super Intelligence Team and put forward the "vision of humanistic super intelligence," trying to find a solution to break the deadlock for the consumer product line that had stagnated in growth.

Looking back now, that was the beginning of Suleyman's role transformation. He was no longer an operator of consumer product business, and it also laid the groundwork for the split between Microsoft and OpenAI.

In March this year, Suleyman's authority was narrowed. He focused on super - intelligence and cutting - edge model research and development, aiming to deliver world - class models for Microsoft within the next five years. The consumer and enterprise Copilot teams were merged and handed over to Jacob Andreou, a former Snap executive. One month later, Microsoft and OpenAI officially announced their split, ending the exclusive authorization agreement.

Data released by Sensor Tower shows that in February this year, the Microsoft Copilot app had 6 million daily active users, which is Suleyman's report card. In the same period, OpenAI's ChatGPT had 440 million daily active users, and Google's Gemini had 82 million daily active users.

Waiting for Suleyman until 2031

Microsoft AI currently doesn't have a real soul figure, and this is not Suleyman's personal problem.

Microsoft's AI narrative has always been deeply bound to OpenAI. OpenAI provides the models, and Microsoft is the distribution channel. Now, their decoupling has just begun. Microsoft needs sufficient product and technology achievements to establish its own AI narrative rhythm.

A deeper question is, does Microsoft AI need a real soul figure? Or are Microsoft's AI products sufficient to make anyone a soul figure?

The answer may be cruel. In the current entire AI industry chain, Microsoft Cloud is unparalleled. It is the most profitable cloud provider. In 2025, the profit of Microsoft's Intelligent Cloud (including Azure) was $50.5 billion, with a profit margin of 42%. Both indicators are higher than those of its competitors Amazon AWS and Google Cloud. Its revenue increased by 34% year - on - year, and about 12% of the revenue increase was contributed by AI - related businesses.

Cloud providers connect upstream storage chip companies, downstream large - model manufacturers, and countless enterprise users. Microsoft has established a very mature commercialization mechanism: enterprise software licenses, Office 365 subscriptions, Azure infrastructure, and GitHub toolchains. This mature product network allows Microsoft AI to easily penetrate every corner of the Internet.

Do users choose to use Microsoft's AI products because Suleyman has done a good job?

No. The developer penetration rate of GitHub Copilot is due to Microsoft CTO Kevin Scott, who has long been responsible for the underlying technology strategy and developer ecosystem. Enterprises' paid use of large - scale AI model services is within the scope of responsibility of Scott Guthrie, who is in charge of Microsoft's cloud business, which is currently the most profitable and stable part of Microsoft AI. Jay Parikh, who is in charge of the developer platform and the intelligent agent factory, all report directly to Nadella.

Although Suleyman is nominally the person in charge of Microsoft AI, in fact, the most profitable and strategically significant areas are in the hands of other peer - level executives, which makes it difficult for him to become the soul figure of Microsoft AI at present.

What about the future? Suleyman has to fight a desperate battle.

Without relying on OpenAI, the long - term competitiveness of all Microsoft's AI products and Agents depends on its own underlying model construction. Whether the model can achieve high - performance, low - cost, and controllable implementation directly determines the long - term stability of Microsoft's upper - layer AI products and underlying cloud infrastructure. This is also Suleyman's statement after the adjustment of his authority: the model is the product.

Will Suleyman become the soul figure of Microsoft AI? The consumer business has already disproven this proposition. The progress of the model layer is more critical than ever. There is only one chance left, which is that the MAI model he leads can rank among the top four in the world within the next five years and stand on the same echelon as Google, Anthropic, and OpenAI.

This article is from the WeChat public account "Bluehole Business" (ID: value_creation), author: Zhao Weiwei, published by 36Kr with authorization.