Everyone’s asking this question right now — and the data makes it uncomfortable to ignore. As someone working inside Microsoft’s AI ecosystem every day, here’s my honest read.
Let’s start with the numbers, because they’re stark.
Copilot currently holds around 1.1% of web AI market share. ChatGPT sits at 64.5%. Gemini has exploded to 21.5%. Even Grok and Claude have pulled ahead of Copilot in paid subscriber share. Of Microsoft’s 450+ million commercial 365 seats, only 15 million are paid Copilot seats — that’s a 3.3% conversion rate on the largest enterprise software install base in the world.
And when enterprise users have access to both Copilot and ChatGPT simultaneously? Only 18% choose Copilot. When all three platforms are available, Copilot’s preference share drops to 8%.
Those are not good numbers for a product that was supposed to redefine enterprise productivity.
So is Copilot actually failing?
Not exactly — but it’s also not winning the way Microsoft hoped.
Here’s the distinction that matters: Copilot the consumer product is struggling. Microsoft the AI infrastructure play is not.
Azure AI revenue is on a $13 billion run rate. Azure grew 39% last quarter — faster than AWS (19%) and Google Cloud (30%). Microsoft is spending $150 billion annually on AI infrastructure. The enterprise demand for Azure AI services is real and accelerating.
The problem is that Copilot — the flagship product that was supposed to monetize all that infrastructure investment for the average knowledge worker — isn’t converting. And that gap between infrastructure success and product adoption is exactly what has Wall Street spooked. Microsoft stock is down over 20% year-to-date, its worst performance since 2008.
Why is Copilot struggling?
Having worked with enterprise customers on Copilot deployments, I can tell you the issues aren’t mysterious. They’re predictable — and they go back to the readiness gap I’ve been writing about in this series.
1. Most organizations deployed Copilot before they were ready for it. No data classification. No sensitivity labels. No access hygiene. Copilot landed in environments where it could surface confidential files, mix up document contexts, and produce outputs nobody trusted. The accuracy NPS for Copilot hit -24.1 in September 2025. When users don’t trust the answers, they stop using the tool.
2. The consumer Copilot product got confused with the enterprise one. Microsoft’s Copilot branding became a mess — Copilot for Windows, Copilot for 365, Copilot Studio, Security Copilot, Copilot+ PCs. Users didn’t know which one they were using or why. Microsoft has now restructured the entire Copilot division under a single leader reporting directly to Satya Nadella. That’s an acknowledgment that the fragmentation was a real problem.
3. The Windows AI push created backlash. Forcing AI features into the OS — Recall, Copilot buttons, AI-generated summaries in File Explorer — without clear opt-in triggered a privacy and control backlash. Users who didn’t choose AI resented having it pushed on them. That sentiment bleeds into enterprise perception.
4. The model itself fell behind. Microsoft’s own AI models are currently considered less capable than offerings from Anthropic, Google, and OpenAI. For enterprise users who can compare, the quality gap is noticeable. Microsoft has acknowledged this publicly — Mustafa Suleyman has stated the goal is to reach frontier-level models by 2027.
The Azure vs Copilot paradox
This is the most interesting part of the story from an architecture perspective.
Microsoft is winning the AI infrastructure war. Enterprise AI workloads are running on Azure at record scale. OpenAI runs on Azure. Anthropic partners with Azure. The hyperscaler bet is paying off.
But the product layer — the thing that sits on top of all that infrastructure and is supposed to be the face of Microsoft AI for the average employee — is underperforming.
One analyst framed it well: “The enterprise AI playbook may need to shift from Copilot-as-a-product to AI-as-infrastructure.” That’s a significant pivot from where Microsoft positioned Copilot just two years ago.
What this means for enterprise customers
If you’re an IT leader or architect making decisions right now, here’s my take:
Don’t write off Copilot. The 15 million paid seats, while below expectations, represent real enterprise commitment. Copilot Studio’s agent-building tools are seeing strong adoption. The M365 integration depth — in Word, Excel, Outlook, Teams — is genuinely valuable when the underlying data foundation is solid.
Do build the foundation first. The organizations seeing real Copilot value are the ones who did the unglamorous work first — data classification, sensitivity labels, identity hygiene, access controls. The ones who skipped that work are the ones generating the bad NPS scores.
Watch the model story closely. Microsoft is building its own frontier models, targeting state-of-the-art capability by 2027. If they close the quality gap, the distribution advantage of 450 million 365 seats becomes a massive accelerant.
Don’t over-index on consumer market share numbers. Copilot’s 1.1% web share is a consumer metric. Enterprise Copilot lives inside 365 apps, not on copilot.microsoft.com. The measurement frameworks don’t fully capture where enterprise usage is actually happening.
Bottom line
Is Copilot falling behind? It depends on what you’re measuring.
Behind — yes — in raw model performance compared to frontier models, in user preference when people have a choice, and in product polish where the experience has been inconsistent. Those are real gaps that Microsoft is actively working to close.
Not behind — at all — in enterprise distribution, security and compliance depth, workflow integration, or monetization channel. No competitor has 450 million commercial seats, an identity layer in Entra, a data governance layer in Purview, and a security stack in Defender all feeding into the same AI platform. That’s not a small advantage — that’s a structural moat.
The real question isn’t whether Copilot is failing. It’s whether Microsoft can close the model and experience gaps fast enough to fully leverage the distribution and ecosystem advantages it already has. The signs point toward yes — and when that happens, the platform play will look like exactly the right call.
That answer will define Microsoft’s next five years. And we’ll have a much clearer picture by the end of 2026.
The views expressed in this post are my own and do not represent the views of Microsoft.
— Jean-Paul Abi Atme