Microsoft has spent the past two years turning its 450-million-seat productivity monopoly into the most aggressive AI distribution push in enterprise software history. $37.5bn in quarterly infrastructure spending. Copilot buttons threaded through every ribbon in Office. Bundling across Windows, Edge, and Teams, and a full leadership reorganization to concentrate the effort under Satya Nadella's personal control. The scale of the bet is unprecedented, the distribution is unrivaled, and the product now sits on the desks of virtually every knowledge worker in the Fortune 500. Which makes one sentence, tucked quietly into Microsoft's latest terms of use, rather harder to explain.
Copilot is for entertainment purposes only. The line, tucked in bold capitals under Microsoft's consumer terms of use and updated quietly last October, surfaced in April when someone finally read it. It landed at an awkward moment: after two years of marketing, $37.5bn of quarterly AI infrastructure spending, and a 450m-seat distribution monopoly pushing behind it, Microsoft's flagship AI product is the tool most professionals have actually tried — and most professionals are not impressed.
The numbers behind that impression are brutal. Microsoft has roughly 15m paid Microsoft 365 Copilot seats against 450m commercial M365 subscribers — a conversion rate of 3.3%, which The Register described as "an awkward figure that landed alongside Microsoft's AI splurge and its insistence that the payoff is coming.” Sensor Tower data reported by CNBC put Copilot at roughly six million daily active users in February 2026; ChatGPT had 440 million, and Gemini had 82 million. Gemini passed Copilot's paid-subscriber share back in late November, and Copilot's U.S. paid-subscriber share fell from 18.8% in July 2025 to 11.5% in January — a 39% contraction in six months. Recon Analytics, which tracks paying AI users, put Copilot's accuracy Net Promoter Score at -3.5 last July; by September it had cratered to -24.1, recovering only partially to -19.8 in January. Of users who had quit the product, 44.2% cited distrust of answers as the reason.
Pilot error
All of which has produced a strange and consequential misread. For most executives, partners, consultants, and senior operators — the professional class that sets budgets and narrates which technologies matter — Copilot is the AI they have used. It is embedded in their Outlook, volunteered in their Word doc, pitched by their CIO, and paid for out of their IT line. When they try it, they get a summary of an email they had already read, a PowerPoint that looks like a first draft, and a Teams recap that misattributes who said what. The reasonable conclusion, from inside that experience, is that AI has been oversold.
But that conclusion is built on the worst possible sample. The same executive who shrugs at Copilot will, in a different tab, ask ChatGPT to draft a board memo and get something close to publishable. A lawyer who finds Copilot's contract review useless runs the same clause through Claude and gets a structured analysis with counterarguments. Developers who ignore the Copilot button in Word are, at the same time, shipping features with Claude Code and Cursor at a pace that has already reshaped how engineering teams hire. GitHub Copilot — Microsoft's own, and a completely different product — reached roughly 4.7 million paid subscribers by January 2026 and is deployed at approximately 90% of Fortune 100 companies, a penetration rate reflecting genuine developer productivity rather than forced provisioning. The gap between the frontier and the side panel has become a chasm, and almost none of it is visible from inside the Microsoft 365 ribbon.
The structural reason is that M365 Copilot is not really one product but a thin orchestration layer over models it does not own, drawing on SharePoint and Outlook tenants that most enterprises had not bothered to clean. The models themselves have advanced at a pace that has startled their own builders; the wrapper has been held back by permission audits, compliance reviews, and the awkward fact that the easiest Copilot features to adopt are also the least useful. Satya Nadella has apparently concluded this is existential. He reportedly began delegating other responsibilities from September 2025 onward to focus personally on the AI product roadmap. Last month Microsoft reorganized its Copilot teams under a new EVP, Jacob Andreou, reporting directly to Nadella, while Mustafa Suleyman — brought in to run Microsoft AI after the Inflection acquisition — was moved to frontier model development.
What happens next depends on whether the professional class ever runs the comparison. For now, the firewall between "the AI my company bought" and "the AI I use in a browser tab at home" has held remarkably well — good news for anyone selling narratives about an AI bubble, rather less good for anyone trying to forecast white-collar productivity. The shift, when it comes, will not arrive as a product launch. It will arrive as an executive noticing, somewhere between a mediocre Teams summary and a ChatGPT draft that saved her weekend, that the problem was never the technology. Microsoft, in fairness, has been honest about one thing. It really is for entertainment purposes only. ■
For more, join 75,000 subscribers getting tech's favorite brief
