Key Points: After taking over as CEO from Ballmer in 2014, Satya Nadella revitalized Microsoft and transformed it into a cloud computing giant. But nearly four years into the AI boom, the software giant has yet to produce a blockbuster product at either the model layer or the application layer. One of Nadella's challenges is to once again restructure the business model, this time around the universal unit of account for the AI economy 鈥?the Token.

Microsoft CEO Satya Nadella speaks at a Microsoft AI tour event in Munich, Germany, on February 25, 2026.

On a clear day in late September in Seattle, about 50 executives from Microsoft's largest customers got a rare opportunity to ask CEO Nadella anything they wanted. One attending executive wanted to know how Microsoft could help enterprises adapt to the AI era. Nadella emphasized "diffusion and adoption within the enterprise" 鈥?meaning AI tools need to be deployed at scale inside companies. Dressed in a black fitted crew-neck top and dark gray trousers, Nadella paced the stage at the downtown Paca Pavilion and said: "All change is always bottom-up and top-down at the same time, and the middle layer also needs to push proactively."

Nadella, now 59, has spent half his life at Microsoft. This customer conference came at a critical juncture: the software giant is trying to amplify its influence in the AI wave. Recently Nadella has increasingly engaged in AI discussions, posting on X last month that model development needs "prudent speed control"; a few days later he responded to concerns about AI safety at the All-In Summit in Los Angeles. Although Microsoft's market cap is approaching $4 trillion, it has yet to firmly secure a position as an AI leader. NVIDIA dominates the AI chip market; Anthropic and OpenAI hold top-tier foundation models; Meta has just launched its first mass-market personal agent; Google's Gemini series is popular with both enterprises and ordinary consumers. Microsoft bet early on OpenAI for its AI positioning, but the alliance between the two has faded. Although Microsoft rents computing power to AI developers through its Azure cloud and the business volume is enormous, Copilot has yet to become widespread in the workplace, and its self-developed foundation models remain niche.

Microsoft will hold a Windows and Surface launch event in San Francisco on Wednesday, with the invitation describing the event as a conversation about how on-device AI can open a new chapter for PCs. Wall Street is cautious about Microsoft's AI story. Even after a strong rally in the third quarter, as of last Friday's close, Microsoft shares had risen only 7% this year, underperforming other large tech stocks and the Nasdaq overall. Since 2023, Microsoft has not beaten the S&P 500 Index for a full calendar year. Tiffany Wade, co-portfolio manager at Columbia Threadneedle and a Microsoft shareholder, said there is still a great deal of uncertainty around Copilot, the core product of Nadella's AI strategy. "I think the product needs to mature further before we can determine the optimal solution and pricing model," Wade said in an interview.

For long-term Microsoft investors, Nadella has a strong track record of navigating industry transitions. He succeeded Steve Ballmer in 2014. During Ballmer's 14-year tenure, Microsoft's stock fell about 33%, and the company was reduced to a struggling legacy software vendor. Nadella then transformed Microsoft into a cloud giant. He abandoned the perpetual Office software licensing model and pushed the company toward subscription-based SaaS software while simultaneously developing the Azure infrastructure cloud. He stopped forcing customers to use the full Microsoft product stack and allowed them to integrate products from rivals such as Oracle and Salesforce. Nadella is himself an electrical engineer and computer scientist, and also a seasoned dealmaker: acquisitions of Minecraft developer Mojang, professional social network LinkedIn, and code hosting platform GitHub have steadily expanded Microsoft's empire.

In recent months, Nadella has candidly told investors that he must once again remake the business model to adapt to the AI era's new trend of usage-based billing. This means Microsoft must sell Tokens at scale 鈥?Tokens are the unit of measurement for computing consumption and the circulating currency of the AI economy. Nadella said on the July earnings call that if the transition succeeds, it will "further expand the total addressable market (TAM)." A large part of the growth story Nadella describes revolves around software production: using top-tier AI tools to write more code and increase development speed. Nadella said at the customer conference late last month: "In just one year, we evolved from code assistance tools to agentic coding, and then to software factories." Now he wants Microsoft customers to build their own software factories, continuously increasing spending and producing code at scale, while also empowering knowledge workers in non-technical roles. At the event, Nadella's subordinates unveiled a new version of Copilot that can perform increasingly complex tasks, generate simple applications, and act as a virtual team member.

Nadella foresaw this transformation early. Microsoft invested in OpenAI in 2019 when it was still a startup. In 2021, one year before ChatGPT was released, Microsoft and OpenAI jointly launched GitHub Copilot, which automatically generates code snippets for applications and websites based on massive amounts of open-source code. This product set the market ablaze that year. But after the generative AI market exploded, Microsoft's share in the AI coding race was eroded by native AI startup products such as Cursor (now acquired by SpaceX) and Anthropic's Claude Code.

"Lessons Learned" 鈥?Microsoft's slower iteration pace caused GitHub to lose market share. Earlier this year, Microsoft adjusted the coding tool's pricing to align with computing consumption needs. Some customers initially resisted the price change, but by July the company said users had surpassed 50 million, up sharply from 2.6 million in October last year, with one in every five GitHub platform users using Copilot. Nadella said on the July earnings call that GitHub Copilot revenue accelerated 60% quarter over quarter, and usage-based revenue rose significantly after the new billing model was implemented. GitHub COO Kyle Daigle said in an interview: "GitHub Copilot has accumulated experience in real-world use." Daigle said the GitHub team shared its experience with other teams inside Microsoft, and other divisions are also studying usage-based billing models. Hayete Gallot, corporate vice president of Microsoft Security, said the company will charge customers based on their technology usage intensity. "I always tell the team that when customers generate usage, it means we have created value. All judgments should be customer-centric. If customers are unwilling to use it, it means we have not delivered value."

Microsoft's current enterprise business relies mainly on the subscription model. The core productivity suite Microsoft 365 (including Word, Excel, and Teams) has used annual subscriptions for years; Copilot charges an additional $30 per person per month. Copilot currently has 30 million seats, while Office 365 commercial users exceed 450 million. Mustafa Suleyman, head of Microsoft AI, speaks at a Copilot event in Redmond, Washington, on April 4, 2025, which was also Microsoft's 50th anniversary.

A major uncertainty in Microsoft's AI roadmap comes from Mustafa Suleyman. Suleyman is a co-founder of DeepMind, which Google acquired in 2014. In 2024, as Microsoft's relationship with OpenAI cooled, Nadella hired Suleyman to lead Microsoft's AI business. Last year, Microsoft was no longer OpenAI's exclusive cloud provider; this year, the intellectual property license Microsoft obtained from OpenAI also changed to a non-exclusive license. Although Microsoft has invested heavily in the AI lab, it no longer enjoys preferential privileges. Microsoft said in April that it remains OpenAI's "primary cloud partner." Suleyman's team has already launched coding, image generation, and reasoning models. This week he posted on X claiming that a new speech-to-text model is "the world's most accurate real-time transcription model, ranked number one!!!" But various independent benchmark leaderboards show that in basic reasoning capability, his team's models still cannot match Anthropic and OpenAI. Microsoft executives say there are synergistic advantages to having model research and application development within the same system. The team responsible for Teams and Office productivity apps uses Suleyman's transcription model to achieve better performance at lower cost; engineers also get direct feedback during product development. Executive Vice President Charles Lamanna said: "Collaboration with Mustafa's team is much tighter."

Suleyman has a high profile at Microsoft, but his responsibilities have been adjusted. In March, Microsoft announced the integration of engineering teams for AI assistants serving both enterprises and ordinary consumers 鈥?Copilot user growth had fallen short of expectations. Jacob Andreou was promoted to executive vice president of Copilot experiences; Suleyman's focus shifted to model research. At the same time, Nadella no longer separately emphasizes "building the world's strongest model." He wrote on his personal blog SN Scratchpad in June: "In my view, our primary goal is to build a frontier ecosystem, not just frontier models. Let value flow broadly to businesses in every industry and every country." At the September Copilot event, Nadella showed a slide: Copilot can connect to "all models," with the page displaying the logos of Anthropic, OpenAI, SpaceX AI, and Microsoft AI.

Reshaping the Commercial Profit Logic 鈥?Model makers are constantly crossing boundaries and entering Microsoft's traditional turf, launching products in areas such as cybersecurity and customer service. Jared Spataro, who leads workplace AI marketing at Microsoft, said that in January this year, Anthropic's Claude Code gained rapid traction, and Microsoft executives were amazed by its ability to call multiple tools to complete complex tasks. People familiar with the matter said some Microsoft orders were taken by Anthropic, prompting the company to deeply study this rising competitor. (Sources spoke on condition of anonymity due to confidentiality requirements.) A Microsoft spokesperson declined to comment. Spataro said the company realized it had to adjust its profit mechanism. "That was the moment we saw clearly: the fixed per-person, per-month pricing model could no longer consistently deliver optimal outcomes. We realized the business model had to be rethought." The company then pushed forward comprehensively with usage-based billing (UBB). Spataro said the first UBB product, Copilot Cowork, can take on large work tasks and show task progress in real time. Nadella said on the July earnings call: "At the beginning of this month, Cowork launched usage-based billing, and thousands of customers are already paying and actively using it." Cowen analyst Derrick Wood sees this as a strongly positive signal and is bullish on Microsoft stock. "It could potentially replicate the rally after Claude Code switched to usage-based billing, with Token bills rising markedly. Cowork has every chance to follow the same trajectory." Nadella himself uses the product. In an interview after the September event, he mentioned that he could have Cowork create a multi-scenario supply chain optimization spreadsheet. "It can produce extremely high-quality results." David Berry, vice president at consulting firm Directions on Microsoft, said budget-sensitive IT leaders are still wary of rolling out Cowork. "People are genuinely worried that once it is turned on, employee usage will wildly consume the budget, and there is no way to tell whether that spending actually generates returns."

As the ChatGPT era approaches its fourth anniversary, Microsoft is still searching for a blockbuster, whether at the model layer or the application layer. Nadella once won investor trust through transformation, but he cannot afford to become a second-tier player in the AI race. Berry still believes Nadella is the right person at the helm. "He is good at identifying crises and crafting plans for business transformation and evolution. The AI field is full of unknowns, and market dynamics are complex, but he seems to already have a complete plan and is consistently executing it." In Seattle, when asked about enterprise transformation, Nadella ultimately raised the role of executives. "Leaders who are personally engaged are the ones capable of driving change. Otherwise, it is just one launch event after another, right?" He added: "You have to be deeply involved."