Microsoft Admits Copilot Has a Problem

Microsoft has poured billions into AI. It co-owns the world's most talked-about AI lab. It embedded AI into the software that 1.4 billion people use for work. And yet, on any given day, only 6 million people open Copilot. ChatGPT gets 440 million.

Something had to change.

The Reorg

On Tuesday, CEO Satya Nadella sent an internal memo announcing a restructuring of Microsoft's Copilot leadership. The move has two distinct parts—and reading them together tells a more complete story than either does alone.

Jacob Andreou, a former Snap executive who joined Microsoft's AI unit, is being elevated to Executive Vice President. He'll own the entire Copilot experience—both consumer and commercial—and report directly to Nadella. Separately, Ryan Roslansky, Perry Clarke, and Charles Lamanna will lead Microsoft 365 applications and the Copilot platform infrastructure, also reporting to Nadella.

That frees up Mustafa Suleiman—former co-founder of DeepMind, acquired by Google in 2014, and the architect of Inflection AI before Microsoft absorbed him in 2024—to do what he says he's always wanted to do: build frontier models.

"The model is the product," Suleiman said flatly in an interview. He's aiming to deliver "world-class models for Microsoft over the next 5 years"—covering code generation, image and audio synthesis, and reasoning systems. Microsoft holds IP rights to OpenAI's models through 2032, but the message is clear: that's a runway, not a destination.

The Numbers That Explain Everything

The usage gap is stark. Copilot: 6 million daily active users. ChatGPT: 440 million. Google Gemini: 82 million. Even Anthropic's Claude, which has been embroiled in a public standoff with the U.S. Department of Defense, crossed 9 million daily users so far in March—still ahead of Copilot.

Sponsored

Advertise with Us

[email protected]

Sponsored

On the enterprise side, the picture isn't better. Only roughly 3% of commercial users with Microsoft 365 subscriptions have activated the Copilot add-on. Companies are paying for the platform but not using the AI layer. That's a monetization problem dressed up as an adoption problem.

In search, Google commands 90% of the market. Bing sits at 5%—virtually unchanged despite years of AI investment.

This context makes the reorg less of a routine shuffle and more of an acknowledgment: the current approach to Copilot isn't working.

Two Bets, One Company

Microsoft is essentially placing two parallel bets. The first is operational: put a product-focused executive in charge of Copilot and fix the user experience, the distribution, the messaging. Andreou's background at Snap—a company that lived and died by consumer engagement metrics—signals that Microsoft wants someone who thinks about retention and daily habits, not just enterprise contracts.

The second bet is existential: build proprietary models so that Microsoft isn't permanently dependent on OpenAI for its core technology. Suleiman's framing—"COGS-optimized, enterprise-specific model lineages"—is deliberately unsexy. He's not promising AGI. He's promising cheaper, faster, more reliable AI that makes Microsoft's products better and its margins healthier.

These two bets aren't in conflict. But they do reflect a company that has spent two years trying to win on borrowed technology and is now reckoning with what it actually owns.

What Investors and Enterprise Buyers Should Watch

Microsoft shares are down 17% year-to-date. The iShares Expanded Tech-Software Sector ETF has dropped 19% over the same period. The broader market anxiety is real: if AI models commoditize software, the incumbents—Microsoft included—could be disrupted by the very tools they're selling.

For enterprise buyers, the 3% activation rate is a useful benchmark. If your organization has Microsoft 365 Copilot licenses sitting unused, you're in the majority—not the exception. The question worth asking internally isn't whether to adopt AI, but whether the specific tool you're paying for is the right one.

For investors, the Suleiman pivot is a long-term signal that deserves attention. A company with Microsoft's compute resources and enterprise relationships building its own model stack is a different asset than one that's essentially a distribution layer for OpenAI. Whether that model bet pays off in three years or ten is the open question.