Microsoft's Copilot: 3.3% conversion rate and the harsh reality of paid AI

Satya Nadella is trying to put a brave face on a rather ambiguous situation. During the latest quarterly report, the Microsoft CEO proudly announced that Copilot's user base had grown nearly threefold compared to the previous year. It sounds like a triumph, if you don't read too much into the details.

Analysts at The Register quickly dampened the optimists' enthusiasm: it turns out that only a measly 3.3% of those who interact with the AI assistant within Microsoft 365 actually pay for it. The rest either use free versions or are stuck in endless testing mode.

This is a classic situation for a market overheated by expectations but not yet capable of proving its actual economic value in daily routine.

To understand the scale of the problem, you need to remember how aggressively Microsoft pushed this tool. After the company invested tens of billions in OpenAI and its own Azure infrastructure, it desperately needed to demonstrate a return on investment. Copilot became the flagship product meant to transform the hype around neural networks into a stable cash flow.

However, the $30 per user per month price tag turned out to be a serious barrier. For a large corporation with ten thousand employees, that's an additional $3.6 million per year.

And before signing such a check, the CFO naturally asks whether the accountant will work twice as fast or if it's just an expensive toy for composing polite rejections in Outlook.

The conversion problem doesn't lie in the fact that the technology is bad, but in the fact that it hasn't yet become indispensable. We've all seen beautiful demo videos where AI transforms a text sketch into a perfect PowerPoint presentation in seconds. In reality, users often encounter neural networks hallucinating, making errors in numbers, or simply producing mediocre text that still needs to be rewritten. When you pay for an Office subscription, you know for sure that Excel will calculate the formula correctly. When you pay for Copilot, you're buying a lottery ticket: whether it will help you today or add more work checking facts is an open question.

Moreover, the market is now in a state of "AI fatigue." Almost every software vendor—from Notion to Salesforce—has added its own neural network and is asking for separate money for it. Companies are starting to calculate the total cost of ownership of all these "smart" add-ons and conclude that it's easier to hire one more human employee than to pay for a fleet of digital assistants that are still learning to be useful. Microsoft traditionally wins the market through persistence and deep integration into its ecosystem, but the current 3.3% shows that even their loyal audience has a limit to patience and rationality.

Investors are beginning to ask uncomfortable questions. If Azure growth slows down and AI service revenue doesn't start growing exponentially, tech giants' stock prices could go down. Microsoft has to balance: on one hand, it needs to continue pouring billions into Nvidia chips to not lose the arms race against Google and Anthropic. On the other hand, it needs to convince conservative businesses that Copilot isn't just a chatbot but a new operating system for work. So far, the arguments are working poorly, and the thirty-dollar barrier remains the highest wall on the path to mass adoption.

The bottom line: Microsoft has faced a value crisis. Growth in the number of free users is not a success but a deferred problem. Will the company be able to prove that their AI is worth its money before investors run out of patience?