Microsoft and the Intracompany AI Rollout Problem

The Q3 FY2026 numbers are out, and on the surface they look about as clean as Microsoft has looked in years. $82.9 billion in revenue, up 18% year-over-year. Azure growing 40% in constant currency, ahead of the 39% Wall Street had penciled in. The AI business crossed a $37 billion annual revenue run rate, up 123% year-over-year. CFO Amy Hood said results exceeded expectations across revenue, operating income, and earnings per share.

Hard to argue with that read.

What is interesting is what happens when you stop looking at the top-line and start asking where that growth is actually coming from, and more importantly, where it is not. Azure is doing what Azure does. The infrastructure side is holding. The part that deserves more scrutiny is Copilot, and specifically the distance between the seat count headline and what seat adoption actually looks like across a typical enterprise deployment.

Microsoft reported 20 million paid M365 Copilot seats as of Q3 FY2026, up 5 million in a single quarter and the fastest seat growth since launch. Satya Nadella confirmed seat adds were up 250% year-over-year. That is a real acceleration. And yet, 20 million seats against a 450 million seat commercial base is still roughly 4.4% conversion. The majority of the installed base has not moved.

Here is where it gets interesting: among paid AI subscribers tracked across the competitive landscape, Copilot’s market share has actually contracted, falling from 18.8% in mid-2025 to 11.5% in early 2026 as Gemini has pulled users away. When employees have simultaneous access to Copilot, ChatGPT, and Gemini, Copilot’s active usage share drops to around 8%. That is a product positioning problem, not a distribution problem. Microsoft has the distribution advantage no competitor can replicate. The issue is that access and usage are different things.

Slight tangent, but it matters: most enterprise Copilot deployments are not stalling because of skepticism about AI. They are stalling because deploying Copilot inside a large organization first requires a data remediation project. Permissions, access controls, compliance frameworks. Over 40% of organizations start with pilots before any broader rollout. The UK government ran a 20,000-user pilot and found employees saved an average of 26 minutes per day. That is meaningful productivity data. But pilot to full deployment is a longer road than Microsoft’s seat count velocity implies.

The capex picture complicates all of this further. Microsoft guided roughly $190 billion in capital expenditures through the end of calendar year 2026, a figure that came in well ahead of analyst consensus. Q3 alone saw $31.9 billion in capex and finance leases, up 49% year-over-year. Gross margin at 67.6% was the lowest since 2022, with depreciation from the data center build-out squeezing it. Management was direct about remaining capacity-constrained through 2026. That level of infrastructure commitment is a statement of conviction. It also creates a revenue obligation on the other side.

Institutional flows are positioned constructively heading into tomorrow. Volume on MSFT hit roughly 79.65 million shares on June 1, more than double the 30-day average of 38.4 million, tied in part to the Microsoft Build Conference and confirmation of the Pershing Square stake. MSFT’s dividend date falls June 11, which adds a near-term institutional bid. The buy-side framing here is Azure momentum and AI infrastructure exposure. Copilot’s adoption trajectory is being watched, but it is not the primary reason funds are accumulating the position.

Tomorrow morning’s May CPI release from the Bureau of Labor Statistics lands at 8:30 a.m. ET on June 10. April came in at 3.8% year-over-year, the highest reading since May 2023, driven heavily by energy costs tied to the Iran conflict. Wells Fargo economists are estimating May headline CPI rose 0.52% month-over-month, which would push the annual rate to around 4.2%, a three-year high. BofA is at a similar estimate. CME Group FedWatch futures are not pricing any rate cuts in 2026 at this point. A hotter May read puts more pressure on rate expectations and, by extension, on how long-duration AI infrastructure names get valued through year-end.

The Azure story is structurally intact. Whether Copilot’s adoption curve bends fast enough to absorb $190 billion in infrastructure spending is a question that does not resolve on any single earnings call. That one is worth staying close to.

Take a closer look.

The Editors