Microsoft is very afraid of being marginalized by AI giants.

Once upon a time, OpenAI needed Microsoft.

Today, Microsoft needs to prove that it doesn't need OpenAI.

On June 2nd, Build 2026 kicked off. Microsoft CEO Nadella unveiled seven self-developed models, an AI workstation for developers, an enterprise Agent governance platform, and a quantum chip in one go. The information density was off the charts, and every move pointed to the same thing:

Microsoft is parting ways with its closest ally.

But if we look at a longer time frame.

You'll find that the situation Microsoft faces today is actually not unfamiliar.

Thirty years ago, Microsoft defined the PC era. Twenty years ago, with the rise of the Internet, Google defined search, Facebook defined social media, and Apple defined the mobile Internet, while Microsoft gradually retreated to the background, becoming the most profitable but least imaginative infrastructure company of the era.

Later, Nadella spent a decade bringing Microsoft back to the center of the stage.

Today, when AI becomes the new operating system, Microsoft suddenly finds itself in that familiar position again.

OpenAI and Anthropic are defining AI, Cursor is defining development, and Google is back at the AI table.

Microsoft seems to be selling cloud services again.

I. The Crack on April 27th

The turning point of the story is not the Build conference itself. It's the deliberately low-key protocol revision two months ago.

On April 27th, Microsoft and OpenAI simultaneously issued statements announcing that their cooperation was entering the next stage. The wording was mild, but the content was significant:

Microsoft's intellectual property license for OpenAI's models and products changed from exclusive to non-exclusive.

OpenAI can provide products through any cloud service provider and is no longer tied to Azure.

Microsoft will no longer pay revenue shares to OpenAI.

To put it bluntly, the breakup agreement has been signed, but they're still living under the same roof.

This means that the biggest moat Microsoft bought with $13 billion since 2019, which allowed only itself to run OpenAI models globally, has been broken overnight.

OpenAI can cooperate with AWS, Google Cloud, or anyone else.

Microsoft has changed from the exclusive partner in the AI era to one of the major cloud service providers.

This is why Build 2026 has become Nadella's most important press conference. He must answer a question: Without the exclusive rights to OpenAI, why should Microsoft still be the protagonist in the AI era?

II. The Real Questions Behind the Seven Models

The seven models released on the day of Build, MAI Thinking 1 (reasoning), MAI Code 1 Flash (code), MAI Image 2.5 (image), MAI Voice 2 (voice), MAI Transcribe 1.5 (transcription), almost cover all the core capabilities of the AI product chain. This is also Microsoft's largest-scale centralized release of self-developed models in history.

But what really matters about these models is not how large the parameters are or how high the scores are, but who the targets are.

Microsoft AI head Mustafa Suleyman said in an interview: We are more focused on the Anthropic-style direction, which is about enterprises, developers, and coding.

This is a direct indication that the competitor is definitely Anthropic.

MAI Code 1 Flash is directly compared with Claude Haiku 4.5 and scored 51.2% on the SWE Bench Pro, higher than Haiku's 35.2%. MAI Thinking 1 targets Claude Sonnet 4.6.

Why is the competitor Anthropic instead of OpenAI?

The answer is obvious.

According to Ramp AI Index data, in April 2026, Anthropic's enterprise paid adoption rate reached 34.4%, surpassing OpenAI's 32.3% for the first time. When enterprises purchase AI services for the first time, in about 70% of direct competitions, the final contract is signed for Claude rather than ChatGPT. Claude Code occupies 54% of the AI programming tool market, and GitHub Copilot has dropped to about 25%.

What's more embarrassing is that Microsoft's internal engineers are also using Claude.

According to media reports, Microsoft's internal evaluation found that core developers' satisfaction with its own Copilot is lower than that of external competitors. The development culture is being infiltrated by external tools.

This is the truth behind the seven models. It's not that Microsoft is also very powerful, but that Microsoft has to save itself.

In the past three years, the entire AI industry believed that models determine everything. But at Build 2026, Nadella repeatedly emphasized not a single model, but 11,000 models.

Behind this is actually a typical Microsoft logic.

In the future, there won't be just one super model. Models will become more and more like databases, servers, and cloud resources, becoming a standard capability.

Microsoft's release of seven models is not just to prove that it can make models. It's also an attempt to make the models themselves less important.

III. Making Money ≠ Being the Protagonist

If we only look at the financial reports, Microsoft's AI business seems to be doing well.

In the third quarter of fiscal year 2026, Azure grew by 40%, and the annualized revenue run rate of the AI business reached $37 billion, a year-on-year increase of 123%. These are real dollars.

But behind the numbers lies an embarrassing fact.

The vast majority of this $37 billion comes from running models for others. OpenAI runs on Azure, and part of Anthropic's computing power is also on Azure.

Microsoft makes money from infrastructure, not from applications.

What about Copilot, which directly faces users?

In the paid AI assistant market, according to Recon Analytics' statistics on paid subscription users in the United States, Copilot's market share dropped from 18.8% in July 2025 to 11.5% in January 2026, a shrinkage of 7.3 percentage points in half a year, a relative decline of 39%.

In the overall workplace working hours, the actual time spent on AI tools has been only 1% for years and has hardly increased for several consecutive years. It has neither seen explosive popularity nor a cliff-like collapse, and has always been stuck on the edge of the mainstream workflow.

The number of paid seats for M365 Copilot reached 15 million at the beginning of the year, but a Stackmatix survey showed that among employees with product access rights, the normal active conversion rate was only 35.8%. After many enterprises purchased authorizations in bulk, the daily usage rate of employees was far lower than the number of seats on the books. The number of people actually using it for work every day is far less than the numbers suggest.

This is the real dilemma Microsoft is facing at the moment. Making money and being the protagonist are two different things.

Azure makes the most money, but ChatGPT and Claude directly have users. Nadella knows that if Copilot's penetration rate remains in the single digits, Microsoft will be like AWS in the AI era, a large, money-making, but brandless pipeline.

No one remembers what AWS looks like. But everyone remembers ChatGPT, and this is exactly the problem Nadella has to solve.

IV. The Subtlety of Jensen Huang's Appearance

There is a detail worth noting at the Build event.

Jensen Huang appeared at the press conference via video link to promote the Surface RTX Spark Dev Box. He said that the PC is evolving from a personal computer to a personal AI.

Two days ago in Taipei, he just released the RTX Spark chip and announced that NVIDIA is officially entering the PC processor market. Microsoft is his most important partner on the PC side.

But if we look closely at this relationship, we'll find a more subtle conflict.

In the data center market, NVIDIA's largest customers are Azure, AWS, and Google Cloud. Whoever buys its chips becomes the winner in AI infrastructure. NVIDIA doesn't take sides and sells its products to everyone.

In the PC market, the RTX Spark chip is jointly developed with Microsoft and MediaTek. The first batch of OEMs are Dell, Lenovo, HP, and ASUS. It seems that Microsoft is the beneficiary, as every AI PC runs Windows.

But NVIDIA's real purpose of entering the PC market is not to help Microsoft sell Windows, but to sell its own local AI computing power. The core selling point of RTX Spark is its 1 petaflop of AI reasoning ability, which can run models and Agents locally. If the local computing power is strong enough, why do users need to go to the cloud? Why do they need Azure?

While Jensen Huang is promoting for Microsoft, he is actually loosening the foundation of Microsoft's cloud business.

This is the most magical part of the technology industry today. Everyone is a partner, and everyone is also a potential competitor.

V. Why Did Build Downplay Consumers?

Looking at the two-day agenda of Build 2026, ordinary consumers were almost downplayed from start to finish.

There was no release of new Copilot features. There was no demonstration of how AI can change your daily life. Even the concept of AI PC, which Jensen Huang shouted so loudly in Taipei two days ago, was deliberately downplayed by Microsoft at Build.

Instead, it was the promotion of products and capabilities for developers and enterprises.

Surface RTX Spark Dev Box: 1 petaflop of local AI computing power, 128GB of unified memory, a workstation built for developers.

Agent 365: An enterprise Agent governance platform that manages identities, permissions, access control, and compliance, and connects to all major clouds.

MXC (Microsoft Execution Containers): A bottom-level container that adds security guards for Agents.

OpenClaw on Windows: Allows personal Agents to safely enter the enterprise environment.

The logic behind this is very clear. In the consumer market, Microsoft can't win. Tools like ChatGPT, Claude Code, and Cursor directly face users, and Microsoft's Copilot can't get in.

But the enterprise market is different. Enterprises need compliance, identity management, auditing, data isolation, and multi-Agent permission control. These are things that ChatGPT and Claude can't do. They are inherently products for individuals.

Nadella's bet is that the AI era needs an enterprise operating system. This operating system is not the model itself, but the entire platform that allows models to run safely and compliantly in enterprises.

Whoever controls this platform controls the entrance to enterprise AI.

This also explains why the last highlight of Build was not the model release, but Agent 365, an AI management platform for managing AI.

VI. What Nadella Really Fears

What is Nadella really anxious about?

It's not that Azure doesn't make money, nor that he can't make models, and it's not that the competitors are too strong. What he fears is that after OpenAI and Anthropic go public, they no longer need Microsoft and he'll be sidelined.

Anthropic secretly submitted an S-1 prospectus to the U.S. Securities and Exchange Commission on June 1st, with a valuation of $965 billion. OpenAI is also preparing to secretly submit its application and is expected to do so in the second half of the year.

What will happen after they go public?

With their own money, they'll use it to buy computing power and build data centers. AWS has already deeply bound with Anthropic. SpaceX supplies $1.25 billion worth of GPU computing power to Anthropic every month, with an annual computing power bill approaching $15 billion.

When OpenAI and Anthropic no longer need to run models on Azure, how much of Microsoft's $37 billion in annualized AI revenue can be retained?

This is really Nadella's time window.

He must transform Microsoft from an infrastructure for running models for others into an enterprise platform that all AI must pass through before OpenAI and Anthropic become independent.

Models can be replaced, and clouds can be replaced.

But if your identity system, compliance framework, audit logs, and security containers all run on Microsoft's Agent 365, you can't replace Microsoft.

This is what Build 2026 is really doing. It's laying a layer of infrastructure that only Microsoft can do under all models.

[Beyond the Page] Words:

Another trend revealed by Microsoft Build 2026 is that models are gradually changing from the protagonist to infrastructure.

When more and more enterprises use OpenAI, Anthropic, Gemini, and various open-source models at the same time, what really matters is not which model to choose, but who is responsible for managing these models.

If in 2019, Microsoft got a first-class ticket to the AI era by leveraging OpenAI.

Then in 2026, Microsoft is doing something else. Nadella wants to prove that he is not just a passenger, but a driver.

He knows better than anyone that Microsoft once missed the mobile Internet era and once retreated from the protagonist of the era to the background.

This time, he doesn't want to experience it again.

This article is from the WeChat official account “Beyond the Page”, author: Huahua. Republished by 36Kr with authorization.