Microsoft: the capacity build is peaking and the margin story is next
Microsoft has spent heavily on data centers to serve artificial intelligence demand, and the market worries the spending will not earn a return. We think the spending rate is close to its peak and that the shift from training to inference lifts cloud margins from here.
We rate the stock a Buy with a $585 target. The central test is whether Azure capacity reaches 85% utilization by fiscal 2028.
Company overview
Microsoft sells cloud infrastructure, productivity software, business applications, gaming and the Windows operating system. The Intelligent Cloud segment, led by Azure, is the main growth driver and now the largest source of operating profit.
Subscription revenue from Microsoft 365 and Dynamics adds a stable base under the cloud growth.
- Headquarters
- Redmond, WA
- Founded
- 1975
- Employees
- ~228,000
- Market cap
- ~$3.8T
- Exchange
- NASDAQ: MSFT
Management overview
One of the most stable senior teams among large technology companies. Capital allocation has been consistent: invest ahead of demand, return the surplus.
CEO since 2014. Drove the shift to cloud subscription and the partnership with OpenAI.
Long-serving finance chief; sets the capital spending envelope and the margin guidance.
Leads legal, regulatory and public policy work, relevant to antitrust and AI regulation.
Runs Azure and the data center build program at the center of this thesis.
Other considerations
Investment thesis and considerations
Our argument is about timing: the investment cost arrives first and the revenue follows. We think the market is at the point where the cost is visible and the revenue is not yet.
1. Capex intensity peaks within the year
Capital spending as a share of revenue levels off as new capacity fills.
Management has guided to a slower rate of growth in spending after this year, and lead times on new buildings give a view of the next eighteen months. As capacity fills, depreciation grows more slowly than revenue and the cloud operating margin recovers. We model capital spending flattening near 30% of revenue and falling slowly after that.
2. Inference lifts the revenue mix
Serving trained models to customers is steady, recurring usage, and carries better margins than building them.
Training demand is lumpy and tied to a few customers. Inference demand comes from millions of daily users of Copilot features and customer applications built on Azure. It grows with adoption and uses capacity continuously. We estimate that inference reaches a majority of AI workloads by fiscal 2028, which supports gross margin expansion in the cloud segment.
3. Software franchise protects the floor
Productivity subscriptions add steady cash flow while AI features raise the price per user.
Microsoft 365 has hundreds of millions of paid seats and price increases tied to AI features have so far been accepted. Even if AI revenue disappoints, the base business earns enough to cover the build-out. That asymmetry is why we are comfortable owning the stock before utilization data confirm the thesis. We would revisit on two consecutive quarters of utilization below 75%.
Bull case
Bear case
Financial profile
USD billions except per-share data; fiscal years end in June.
| Metric | FY21 | FY22 | FY23 | FY24 | FY25 |
|---|---|---|---|---|---|
| Revenue | 168.1 | 198.3 | 211.9 | 245.1 | 281.7 |
| Cloud revenue | 69.1 | 91.2 | 111.6 | 137.4 | 168.9 |
| Operating income | 70.0 | 83.4 | 88.5 | 109.4 | 128.5 |
| Net income | 61.3 | 72.7 | 72.4 | 88.1 | 101.8 |
| Diluted EPS | $8.05 | $9.65 | $9.68 | $11.80 | $13.64 |
Financial projections
| Metric | FY26E | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|
| Revenue | 325.0 | 372.0 | 420.0 | 468.0 | 515.0 |
| Cloud revenue | 208.0 | 257.0 | 308.0 | 358.0 | 406.0 |
| Operating margin | 46.0% | 46.8% | 47.9% | 48.6% | 49.2% |
| Diluted EPS | $15.70 | $18.20 | $21.00 | $23.70 | $26.40 |
Key assumptions
- Azure grows 30% in FY27 and slows to 22% by FY30.
- Capital spending peaks near 30% of revenue in FY27.
- Productivity revenue grows 10% a year including AI feature pricing.
Valuation profile
| Method | Weight | Value / share |
|---|---|---|
| Discounted cash flow | 55% | $595 |
| Sum of the parts | 35% | $575 |
| Trading comps | 10% | $555 |
| Blended target | 100% | $585 |
Discounted cash flow carries the most weight because the thesis is about the path of cash flows. The sum of the parts values Azure, productivity and gaming separately and acts as a check against the market's blended multiple.
File repository
SEC filings
7 filesModels
6 filesResearch reports
6 filesEducational research produced by graduate students. Not investment advice. Figures are illustrative samples.