Editor's Note: Microsoft was one of the first giants to bet on OpenAI during the generative AI wave. With its investment in OpenAI and exclusive cloud partnership, Microsoft was once considered the most certain winner in the AI era: Azure reaped the model dividend, Office, Bing, GitHub, and its enterprise software line were fully integrated with Copilot, and Nadella was expected to complete another platform-level migration, just as he had led Microsoft to shift to cloud computing in the past.
Two years later, however, Microsoft's advantages began to become more complex. OpenAI was no longer just a technology supplier for Microsoft, but also a direct competitor vying for enterprise customers; models like Claude and Gemini were rapidly catching up, diminishing the lead gained from GPT's exclusivity; and the emergence of AI Agents further impacted Microsoft's long-standing SaaS business model. Stock price declines, lower-than-expected paid penetration of Copilot, and GitHub Copilot being overtaken by Cursor and Claude Code all forced Microsoft to re-evaluate its AI strategy.
The most noteworthy aspect of this article isn't whether Microsoft can catch up with OpenAI, Anthropic, or Google in terms of model capabilities, but rather that Microsoft is attempting to redefine its position: it's no longer betting everything on a single model, but rather shifting towards a "model-agnostic" enterprise AI platform strategy. In other words, Microsoft aims to become the foundational layer connecting models, data, security, workflows, cloud computing, and enterprise software. Models may come from OpenAI, Anthropic, or even Microsoft's own Superintelligence team in the future, but what truly remains within the Microsoft ecosystem are the enterprise customers' work platforms, data assets, development environments, and security frameworks.
This also explains why Nadella personally participated in the development of Copilot. For Microsoft, the AI competition is no longer just a model race between labs, but a systemic competition about organizational speed, product form, customer relationships, and capital expenditure. Claude Code and Claude Cowork demonstrate that AI agents can reshape software development and office processes; open-source projects like OpenClaw illustrate that an "always-on" AI assistant is moving from concept to reality. What Microsoft needs to do is package these more radical, AI-native experiences into a security, compliance, and governance framework that is acceptable to enterprise customers.
However, this path comes at a high price. To keep up with cutting-edge models and support agent-based products, Microsoft is pushing the AI competition towards gigawatt-level infrastructure investment: more data centers, larger chip clusters, and higher capital expenditures. By 2026, Microsoft expects its capital expenditures to reach approximately $190 billion. In other words, in the AI era, Microsoft must both rapidly experiment like a startup and continuously invest heavily like a cloud computing giant.
The real challenge Microsoft faces isn't whether it can remain the sole winner in the AI era, but whether it can maintain its core position in enterprise software amidst rapid model commoditization and the continued disruption of software business models by agents. For Nadella, this may not be a routine product adjustment, but rather a second entrepreneurial venture for Microsoft in its AI platform migration.
The following is the original text:
Mid-January 2026, Redmond, Washington. The weather was gloomy and overcast, the kind of morning perfect for setting an alarm to "sleep a little longer." But inside Building 92 of Microsoft's sprawling campus, a team of engineers had already arrived early.
They are fighting a tough battle, and they are already behind.
This team was developing a new AI product. It was more like a personal assistant, helping users book flights, reply to emails, and even find reliable local plumbers. The team members were well aware that other tech companies were developing similar products. Just then, Microsoft CEO Satya Nadella arrived. He wanted to show them something.
Nadella opened his laptop and launched an application. It was a system for directing and controlling multiple AI agents, which he called the "Chain of Debate." Nadella demonstrated it to the engineers. The team members exchanged knowing glances, like veteran basketball players suddenly realizing that a newcomer was actually quite skilled.
This application wasn't created by someone else at Nadella's request; he wrote it himself using the AI tool "Vibe Coding".
"This set the tone for how hard the entire team would push forward with their work," recalled Jacob Andreou, Microsoft's executive vice president in charge of Copilot's design. Nadella was in the same room with everyone, practically standing behind the engineers, and he himself was involved on his computer.
Seeing the CEO's excitement about building the new product himself, the team was also motivated. By late February, this sprint came to an end, and Microsoft launched Copilot Tasks—a personal assistant-like AI tool that can use computers. The prototype that Nadella had previously built also became a reference model for a feature called "model council" in Copilot and other components.
But Nadella's frequent involvement with AI product teams, even personally building prototypes, speaks volumes about Microsoft's current predicament. After all, this is a $3 trillion tech giant, not a grassroots startup where the CEO often works alongside developers to launch and write code.
Nadella's concerns about Microsoft's AI strategy are quite evident. Last October, he announced that he would step back from some business responsibilities to devote more energy to AI research, product innovation, and the construction of AI data centers.
This concern is not unfounded. Microsoft's stock price has recently experienced a difficult period. After hitting an all-time high last October, Microsoft's stock price fell by about 34% over the next five months. Meanwhile, revenue related to AI from Microsoft's Azure cloud computing platform has more than doubled in the past year.
Microsoft has also become a prime victim of the so-called "SaaS pocalypse." The emergence of AI programming agents triggered a sell-off in software stocks. Many investors began to believe that these products meant that companies would no longer buy AI products from Software as a Service (SaaS) vendors like Microsoft, and might even stop buying off-the-shelf software altogether.
Between October 28, 2025, and March 27, 2026, Microsoft's stock price fell by 34%. Sales of Microsoft's enterprise version of Copilot also fell short of the company's expectations. Of the 450 million users of the Microsoft 365 suite, less than 4.5% currently pay for Copilot features. Meanwhile, the usage of the consumer-facing Copilot chatbot lags far behind ChatGPT, Gemini, and Claude. The once-leading AI programming assistant GitHub Copilot has also been surpassed by AI startups Cursor and Anthropic's Claude Code.
Two years ago, Microsoft appeared to be one of the earliest winners in the AI era. Thanks to Nadella's forward-looking bet on OpenAI, Microsoft gained exclusive access to the high-growth AI startup's models and could integrate them into its own product ecosystem. If enterprises wanted to use OpenAI's technology, the only cloud service provider they could choose was Microsoft Azure. Microsoft even believed at one point that OpenAI gave it its best chance in years to challenge Google Search.
At that time, Nadella had been at the helm of Microsoft for ten years. He had led Microsoft through the platform migration from desktop software to cloud computing, and now it seemed poised to replicate that success in the AI era.
But AI is evolving too rapidly. Two years is enough to constitute a long cycle. The story that follows is how Microsoft missed its early AI lead and how it is trying to regain it.
Where is the problem?
Microsoft initially gained a leading position in the AI race thanks to its collaboration with OpenAI; however, this collaboration also put it in a passive position.
Microsoft recognized the young San Francisco company early on and initially invested $1 billion in 2019, bringing its total commitment to OpenAI to $13 billion. Leveraging OpenAI's technology, Microsoft has launched a range of AI products under the Copilot brand across its consumer and enterprise software product lines.
However, after the release of ChatGPT at the end of 2022, OpenAI's explosive growth and rapidly expanding ambitions quickly put pressure on the relationship between the two companies. The two companies clashed on several issues: OpenAI always wanted more computing resources; Microsoft believed that OpenAI had not fulfilled its contractual obligations and shared technological innovations in a timely manner; OpenAI began directly marketing its AI models to the same enterprise customers that Microsoft was also selling Copilot to; and when OpenAI sought restructuring, the two sides disagreed on how much equity Microsoft should receive in the new for-profit company.
Nadella knew that betting Microsoft's AI strategy on a startup that was not yet fully proven was inherently risky. In November 2023, this risk was laid bare: the nonprofit board that controlled OpenAI's for-profit business fired CEO Sam Altman for "failing to be consistently honest," and only informed Nadella minutes before announcing the decision to the public.
Nadella had to quickly reassure investors, emphasizing that Microsoft still retained access to OpenAI technology; meanwhile, he worked with Altman to pressure the board to reverse its decision. Nadella announced that Microsoft was prepared to hire Altman, as well as any OpenAI employees willing to follow him to Microsoft. The possibility of a mass exodus of employees ultimately forced the board to relent and reinstate Altman.
Within OpenAI, the five-day crisis was later referred to as "the blip." But according to people familiar with Nadella's thinking, the incident deeply affected him. He had to find a hedge for Microsoft's AI bet.
"When Nadella joined the company's AI engineering team for a sprint development, it set the tone for how hard the entire team would push forward with their work."
—Jacob Andreou, Executive Vice President of Microsoft Copilot
Microsoft's Plan B is Mustafa Suleiman.
Suleiman was a co-founder of Google DeepMind before leaving to found his own AI startup, Inflection. In March 2024, Microsoft hired Suleiman and Inflection's technical team in a $650 million deal and acquired their technology license. Subsequently, Suleiman was appointed CEO of Microsoft's new AI division, MAI, whose responsibilities include two parts: first, developing cutting-edge models within Microsoft as a hedge against the risks of OpenAI; and second, expanding the user base of Microsoft's Copilot chatbot.
However, this step did not go smoothly. Microsoft's cooperation agreement with OpenAI prohibited Microsoft from training models exceeding a certain size. Suleiman told Fortune, "At that time, we could basically only train Microsoft's own native models, and only at the size of SLM, which is a small language model."
The first public test of the general-purpose language model for MAI was called MAI-1 preview, which was launched in August 2025. However, it ranked quite low on various performance leaderboards and was never released on a large scale.
MAI also failed to turn the Copilot chatbot into a consumer-facing blockbuster product. According to media reports, a year after Suleiman took office, Copilot's usage stagnated at around 20 million weekly active users, while ChatGPT's user base skyrocketed, eventually reaching 900 million. In 2025, Microsoft gave Copilot a major upgrade, attempting to make it more like a task-oriented personal assistant, but this upgrade failed to revive growth. As for the new version of Bing search with AI features, it has barely shaken Google's share of the search market.
Meanwhile, Plan A also began to run into trouble.
In 2023, OpenAI's GPT model was far ahead of the industry. But by early 2025, Anthropic's Claude was frequently topping AI leaderboards, and many companies preferred to use it for complex tasks. Google's Gemini was also becoming increasingly competitive in vision tasks. Meanwhile, Microsoft's Copilot product was still entirely powered by GPT. The engine that once supported Microsoft's AI strategy began to feel like a heavy burden.
Microsoft Business CEO Judson Althoff admitted that the company did make a few mistakes. First, naming both consumer and enterprise products Copilot was inherently confusing. Althoff, who holds a private pilot's license, joked, "The worst thing is having more than one co-pilot."
Microsoft had also incentivized sales representatives to promote both the freemium and premium versions of the enterprise version of M365 Copilot, but only the premium version truly delivered value to enterprise customers. "We made a mistake," he said.
Microsoft is also striving to keep pace with the evolution of AI technology. A key turning point came in 2025 when Anthropic released Claude Code. Developers only needed to describe what they wanted, and it could autonomously write the complete program. This was no longer a "co-pilot," but "autopilot." In just six months, it reshaped the way software is developed.
Then, in January of this year, Anthropic launched Claude Cowork. This is an agent that can use software, including Microsoft productivity tools such as Excel and PowerPoint, and can complete tasks autonomously.
Claude Cowork poses a significant challenge to M365 Copilot and the AI Agent that Microsoft has been pushing its customers to adopt. In fact, it threatens not only Microsoft, but most business software. This perception triggered the so-called "SaaSpocalypse" sell-off in software stocks. Ultimately, the tech stock market lost over $2 trillion in market capitalization, including a single-day plunge of $357 billion for Microsoft.
How Microsoft fixes
By the fall of 2025, Nadella realized that Microsoft had to reboot its AI strategy. Since then, the company has struck a difficult balance: on the one hand, it needs to innovate rapidly like an AI startup; on the other hand, it must continue to reliably serve investors and enterprise customers like the stable Microsoft of the past.
Nadella delegated many business and day-to-day operational responsibilities to Microsoft veteran Althoff so that he could focus on AI product development. Althoff said he was in charge of Horizon Zero and Horizon One, while Nadella was in charge of Horizon Two and Horizon Three. At the same time, Nadella began to break down internal silos, making Microsoft faster, flatter, and more agile.
In March of this year, Nadella merged the consumer and enterprise Copilot teams. Suleiman no longer leads consumer AI products, instead leading a renamed model development project: the Superintelligence team. Suleiman stated that the name reflects the team's ambition and helps attract top researchers.
Jacob Andreou joined Microsoft in 2025, having previously worked at Snap and venture capital firm Greylock. He now leads the Copilot Experience for both consumer and enterprise clients, reporting directly to Satya Nadella. Joining Suleiman and Andreou on the Copilot leadership team are three senior executive vice presidents from Microsoft: Charles Lamanna, responsible for Copilot, AI Agents, and the platform; Ryan Roslansky, responsible for Microsoft Office and LinkedIn; and Perry Clarke, Chief Technology Officer for Applications.
Lamanna stated, "We envision it as a backend, a brain, driving both consumer and work scenarios." Nadella himself attends the weekly stand-up meetings of the Copilot leadership team and participates in a continuously running Teams channel dedicated to discussing Copilot's development progress.
Microsoft faces a delicate balance: it needs to innovate fast enough to keep up with AI rivals like Anthropic and Google, while continuing to be a reliable partner for large enterprise customers.
Andreou pointed out that the two new products demonstrate that the unified Copilot team is operating as Nadella envisioned: one is Copilot Tasks, a consumer-facing product that Nadella personally participated in prototyping in January of this year; the other is Copilot Cowork, a product for enterprise clients.
He said, "Both products offer a cutting-edge user experience, one geared towards consumers and the other towards enterprise users. Moreover, our team integrated resources and built them rapidly in just a few weeks."
Microsoft has also agreed to OpenAI's long-pending restructuring plan, with significantly reduced restrictions. The software giant acquired a 27% stake in OpenAI. If OpenAI goes public as widely expected, this will provide potential upside for Microsoft. However, the exclusivity provisions of the old agreement have been abandoned: OpenAI can now collaborate with other cloud service providers, and Microsoft can use models from other AI companies.
Suleiman stated that the new agreement finally allows Microsoft to build larger, more powerful, cutting-edge AI models and ultimately achieve self-sufficiency. However, he added that it will take Microsoft another two to three years to catch up with top AI labs.
The revamped partnership also allowed Microsoft to embrace Anthropic, OpenAI's main competitor. Last November, Microsoft pledged up to $5 billion to Anthropic and began offering its models on Azure. The ability to use Claude to drive Copilot has been popular among enterprise customers and has also helped Microsoft build Copilot Cowork.
"We must acknowledge that OpenAI and Anthropic are helping us run faster." — Judson Althoff, Microsoft Business CEO
But Microsoft isn't simply replacing its reliance on one loss-making AI startup with reliance on another. Behind its investment in Anthropic lies another assessment of the industry's trajectory: AI models will become increasingly commoditized. At least in the enterprise market, the real value won't be concentrated solely in the "brain" of AI, but will shift towards the tools, data, security, cloud computing, and workflow systems that operate around that brain.
This is precisely where Microsoft believes it can win.
It already possesses many key assets: software tools, security systems, data warehouses, and cloud computing capabilities. Microsoft has also created a suite of products under the IQ brand to help enterprises create customized workflows, aggregate their own data, and build, deploy, and monitor agents that run these workflows based on any AI model from any vendor.
Althoff stated, "We don't believe that companies will change their information work platforms, development environments, and security environments every time a new model is released."
This strategic shift has also brought about new business models.
In the past, Microsoft typically charged users based on licensing fees, such as $30 per user per month for Copilot. Customers liked this model because it made budget planning easier. However, if the AI agents in these products use models that Microsoft doesn't own, Microsoft must pay the AI vendors corresponding token consumption fees.
Therefore, Microsoft has begun to shift to a hybrid pricing model: the base portion is still charged based on user licenses and includes a limited token quota; any excess is charged per token. This is done to prevent a "model-agnostic" strategy from eroding profit margins.
Driven by cost control considerations, Microsoft has also begun streamlining its workforce. In April of this year, Microsoft announced its first-ever voluntary severance package, primarily targeting its longest-serving employees. The company stated that approximately 7% of its U.S. workforce, or about 8,750 employees, are eligible for the program, which is projected to cost $900 million.
There are signs that Microsoft's adjusted corporate strategy is working. As of the end of March, Azure revenue grew 40% year-over-year, and Microsoft's overall AI business achieved annualized sales of $37 billion, a 123% year-over-year increase. Currently, 20 million M365 users are paying for Copilot, a quarter of whom will be new users in the first four months of 2026. Althoff stated that adoption is accelerating.
UBS analyst Karl Keirstead said that a growing number of Microsoft customers are telling him they are beginning to see the value in Copilot. However, the overall user base remains unsatisfactory. He said, "I don't think they've reached a penetration rate that would satisfy Wall Street."
Microsoft's "model-agnostic" strategy may also have a flaw: what if those high-profile AI startups also start building Microsoft-style enterprise tools and connectivity systems?
This is no longer hypothetical. In February of this year, OpenAI launched its enterprise-oriented Frontier platform, offering many of the capabilities that Microsoft is building into its new tools. Anthropic is also moving in this direction, launching its Claude Managed Agents service.
Microsoft's argument is that decades of enterprise customer relationships, its reputation for reliability and security, and its deep integration with customers' existing software systems will give it an advantage. Althoff said he welcomes competition. "We have to admit that OpenAI and Anthropic are helping us run faster," he said.
However, some question whether a company the size of Microsoft can truly keep up with the agility of AI-native startups. UBS's Keirstead stated, "Microsoft, and frankly all software companies, are facing a situation they haven't encountered in over a decade: highly innovative new competitors. Expecting a large, established company like Microsoft to avert the rapid growth seen in OpenAI and Anthropic might be asking too much."
Bank of America analyst Tal Liani, however, sided with Nadella's camp. He believes that AI companies are unlikely to build the complete product suite offered by Microsoft. This means that Microsoft doesn't necessarily need to win the AI race; it only needs to avoid losing.
He said, "It may not be the best, but as long as it's good enough and offers high value through bundled sales, that's where Microsoft's value lies."
However, even just "not losing" doesn't come cheap.
Like other hyperscale cloud providers, Microsoft is investing heavily in data centers and dedicated chips. In fiscal year 2025, Microsoft's capital expenditures reached $88.2 billion, roughly equivalent to peers like Google Cloud and Amazon AWS. However, in hindsight, this was still overly conservative. The surge in demand left Microsoft with insufficient computing power and unable to recognize signed AI revenue as actual revenue at the expected pace.
"I thought we would catch up," CFO Amy Hood admitted during the earnings call last October, "but we didn't."
Now, Microsoft is ramping up its spending. The company projects capital expenditures to reach approximately $190 billion in 2026, more than three times the amount spent in 2024. While Wall Street has historically been wary of such large-scale spending, it now seems more tolerant of such massive investments. However, if investor sentiment reverses, Microsoft will be more vulnerable than ever before.
In November 2025, an independent developer named Peter Steinberger released OpenClaw. It is a free, open-source system that can transform any AI model into a long-term, autonomous, and always-online agent: it can develop software, act as a virtual administrative assistant, and even manage inventory for an online store.
OpenClaw has become very popular among developers and cutting-edge AI users. Nadella is reportedly one of them.
However, while OpenClaw is popular, it has a significant problem: to be truly effective, it requires access to systems, data, payment information, and passwords, making it extremely risky. Furthermore, it consumes tokens at an alarming rate.
"I can't launch OpenClaw at Microsoft," Nadella said at a tech conference in San Francisco this March. "I don't have the authority to do that because it would be seen as Microsoft releasing a virus. But at the same time, it is a truly remarkable innovation."
Nadella has tasked the unified Copilot team with creating a Microsoft version of OpenClaw: one that retains the fun and ease of use of a consumer product while also possessing the security and governance capabilities required by enterprises. Andreou sees this as a test for the new organization: "This is what we call a victory here."
Lamanna believes this could be key to igniting Copilot's growth. He said, "The hardest question has always been: How do you help people change the way they work?"
If a perpetually running AI assistant truly becomes feasible, it will make this change much easier. It also means that the basic unit of AI will shift from "models" to "always-on agents." This is precisely a paradigm shift, and it will test whether Microsoft's so-called "connected organization" strategy will remain valid as its core structure changes. Lamanna stated that an enterprise-level Microsoft version of OpenClaw is not far off.
"Gigawatt-scale"
During the week of March 30th, Suleiman convened a three-day remote meeting in Miami with the new Superintelligence team. This team comprised approximately 500 people from around the world. The purpose of the meeting was to develop a roadmap for achieving gigawatt-scale AI training and operation. This scale of training would allow Microsoft to directly compete with OpenAI, Anthropic, Google DeepMind, Meta, and xAI.
Suleiman stated that it is crucial for Microsoft to achieve self-sufficiency by 2030. Microsoft will lose access to OpenAI technology in 2032.
The entire team gathered in a huge banquet hall to listen to keynote speeches by Suleiman and Nadella, and to participate in an "Ask Me Anything" Q&A session. Suleiman recalled that Nadella described this moment as a "re-creation of the company" for Microsoft in response to the shift towards AI platforms.
This is a very meaningful statement.
After the keynote address, the meeting was broken down into different workflows. Each team gathered around 40 whiteboards placed around the banquet hall to brainstorm and plan for the upcoming eight-week sprint. Nadella didn't leave; he stayed.
For the next three hours, he moved between tables, talking to researchers, offering suggestions, and sharing ideas.
If this truly is a "reinvention," then Nadella is playing the role of a startup CEO. He's not taking any advantages for granted. He knows that Microsoft could lose everything, but it still has everything it can fight for.
