MUSTAFA SULEYMAN arrived at Microsoft two years ago with a formidable résumé — co-founder of DeepMind, architect of Inflection AI — and a mandate to build the consumer AI product that would justify the company's $13.8 billion bet on OpenAI. On Tuesday, Satya Nadella quietly relieved him of that job. In a memo to staff, the CEO announced that Suleyman would step back from overseeing Copilot to focus on frontier model development, while Jacob Andreou, a former Snap executive who joined Microsoft only last year, would take charge of the entire Copilot division — consumer and commercial — as executive vice president, reporting directly to Nadella.
The reshuffle is the latest in a series of organizational convulsions at Microsoft's AI unit, and it arrives amid numbers that explain the urgency. Copilot for Microsoft 365 has sold roughly 15 million paid subscriptions — a figure Nadella has cited as "record" growth. But set against the company's 450 million-plus commercial Microsoft 365 customers, that represents a conversion rate of barely 3.3%. On the consumer side, the picture is bleaker still: Copilot's web app commands roughly 1.1% of global AI chatbot traffic, according to SimilarWeb, down from 1.5% a year ago. ChatGPT holds 64.5%. Google's Gemini has surged to 21.5%. Even Elon Musk's Grok, at 3.4%, now outpaces Microsoft's offering. The behemoth that owns 27% of OpenAI and spent $37.5 billion on AI infrastructure last quarter alone has, by the metrics that matter most, been losing ground to upstarts spending a fraction of that sum.
Copilot, co-problem
Yet the reshuffle is less a demotion than a diagnostic — an admission that Copilot's difficulties are structural, not merely managerial. Under Suleyman, Microsoft maintained separate engineering teams for consumer and commercial Copilot, producing what customers and staff alike described as a confusing proliferation of brands. Nadella's memo acknowledged as much, calling for Copilot to become "a truly integrated system" rather than "a collection of great products." The appointment of Andreou — whose Snap tenure was defined by product simplification and user-growth mechanics — signals that Microsoft now reckons its AI problem is less about research prowess than about product discipline.
Still, the deeper tension is one Andreou cannot solve alone. Microsoft's in-house models trail the frontier offerings from Anthropic, Google, and OpenAI itself. Suleyman's new mandate — to build "world-class models for Microsoft over the next five years" — is the company's most explicit commitment yet to reducing its dependence on its own portfolio company. The arrangement is exquisitely awkward: Microsoft holds a 27% stake in OpenAI, valued at roughly $135 billion, and receives 20% of OpenAI's revenue through 2032. It has contracted $250 billion in Azure cloud commitments from the ChatGPT maker. And yet it is simultaneously investing in a superintelligence team whose stated purpose is to make those very models unnecessary. As Suleyman put it in an interview, the future value "is going to accrue to the model layer" — a concession that renting someone else's models, even from a company you partly own, is an uncomfortable long-term posture for a $3-trillion-plus incumbent.
The competitive pressure is acute. Anthropic's Claude Cowork and Google's Gemini have both gained traction in the enterprise agentic-AI space that Microsoft covets. Copilot's paid subscriber share contracted from 18.8% to 11.5% in just six months through January 2026, according to Recon Analytics — a 39% decline even as the overall market expanded. Meanwhile, Microsoft's own employees reportedly favor competitors' tools, an awkward detail that has surfaced in industry discussions.
The new Copilot Leadership Team — Andreou, Suleyman, LinkedIn CEO Ryan Roslansky, Perry Clarke, and Charles Lamanna — is designed to force alignment between models, products, and platforms. It also fills the gap left by Rajesh Jha, the longtime executive vice president who announced his retirement earlier this month after 35 years, a departure that triggered much of this cascading reorganization. Whether a five-person committee can move faster than the nimble labs it competes against remains an open question.
What Nadella is really betting on is a two-track strategy: let Andreou fix the product while Suleyman builds the engine. If it works, Microsoft gets both a coherent AI assistant and proprietary models that slash its cost of goods sold. If it doesn't, the company will have spent years and hundreds of billions of dollars to discover that distribution — even distribution as vast as Windows, Office, and Azure — is no substitute for building something people actually choose to use. For now, the world's largest software company has done what large organizations do when a product underperforms: it has reorganized. Whether it has also rethought is a question the next earnings call cannot answer, but the one after that might. ■
For more, join 75,000 subscribers getting tech's favorite brief here
