MSFT Q3 Earnings Prep: Running the Engine Hot
Microsoft earnings are Wednesday and there’s so much more to listen for than Azure growth and OpenAI. Here’s a quick analysis of their recent data and some things to listen for.
They’ve been running the engine hot using 3 levers:
They’ve amped margins by doubling server depreciation to 6 years - FY2020: 2-3 yrs → FY2023: 2-6 yrs. Every GPU dollar now hits the P&L at half the speed it did in 2020, meaning margins are higher by billions. All the major tech firms making big infrastructure investments did this, but also are GPUs suddenly relevant for 6 years when they used to go stale quickly? Listen for any further change to useful life. The charge coming due would be an impairment charge, but it’s way too early for this.
Brutal headcount discipline is subsidizing capex - The Wall Street Journal picked up on this this morning, but headcount has been frozen at FY24 levels as revenue grew $36B. Rev/employee shot from $959K → $1.24M. S&M as % of revenue dropped 460bps. This is what is funding $65B/yr capex. It’s not because of AI efficiency today, it’s because they’re feeding the beast.
MSFT is stretching contract durations as insurance against long-term data center build out - % of RPO recognized in next 12mo went from 45% in 2023 to 25% in 2Q26. Customers are pre-committing 2.5+ yrs, while MSFT investments in AI are still in early stages (land leases etc.). They’re starting down a long expensive road and can’t have customers bail half way. Listen for this ticking back up, that will signal the capex math getting shaky.
Microsoft can’t not make this bet -- AI is sure to have a huge role in the knowledge work they make their money on going forward -- and they can and must lead. But they need the cracks to hold and they need to show AI wins their customers will pay for to get through the transition.
Look for cracks at the segment level. Intelligent Cloud gross margin went from 67% in FY23 to 62% in FY25. COGS grew 36% last yr vs. 22% revenue. The upshot is that costs have been growing 65% faster than revenue inside the AI engine. Listen for IC GM below 60%, which would mean AI infra economics are tougher than consensus bets.
Also, the legacy engine that pays for everything has to speed back up. M365 Commercial seat growth has been decelerating: 20% (2020) → 6% (Q2 FY26). Future growth = Copilot ARPU at $30/seat. Listen for Copilot attach rates.
Lastly, the OpenAI deal has two interesting implications. First, Copilot is a major source of dollars for the capex build out. It’s an untenable drag to have OpenAI capture these dollars, so this agreement releases funding pressure on MSFT. Second, it positions their relationship more as a hedge than a profit sharing agreement. MSFT’s upside revenue share is capped, but the 27% equity loss share stays uncapped. Meanwhile, Copilot revenue sharing flows one way now — MSFT stopped paying OpenAI for Copilot, keeping 100% of every $30 seat. OpenAI’s fortunes and Copilot are decoupled.
In the end, MSFT operating margin went from 37% (2020) to 46% (2025) during $15B to $65B/yr capex. That doesn’t happen without working these 3 levers. If one cracks Wednesday, it’s a wobble. If two crack, the margin expansion thesis needs new insights and a clear adjustment plan. And if we see Copilot show strong attachment numbers and M365 growth is back up, then they look well positioned for this transition.
All numbers are from Revelata’s deepKPI, which gets you 1-click auditing on every data point and hooks into Excel, Claude, ChatGPT, OpenClaw, and your API. Check it out, it’ll change your workflow like crazy.
 | A guest post by
|