The question I get about Microsoft has changed over the past year. It used to be “should I own it,” and the answer was almost boringly obvious. Now it is “how much is the AI build-out going to cost, and does the cloud business actually earn it back.” That is a harder question, and it is the right one to be asking. Microsoft is spending money at a rate that would have looked reckless for this company three years ago, and the market has decided to trust that the spending pays off. I think it probably does, but the timing of when that shows up in the numbers matters more than most buyers seem to think.
The Spending Is the Story Now
A few years ago Microsoft’s capital expenditure ran in the high twenty billions a year. In the most recent fiscal year it was up around eighty billion once you include the finance leases the company uses to secure data center capacity, and management has signaled that the next year steps up again. Put simply, Microsoft is now spending more on property and equipment annually than most companies in the S&P 500 earn in profit. Almost all of it goes toward data centers, power agreements, and the Nvidia and custom silicon that fills them, and the Microsoft financials page tracks that line alongside margins and cash flow each quarter.
The bull case is straightforward. Azure has been growing in the mid to high thirties in percentage terms, a meaningful chunk of that from AI services, and demand is still running ahead of the capacity Microsoft can bring online. If you believe that demand curve holds, then front-loading the infrastructure is the correct decision and the spending is a sign of strength. The bear case is equally straightforward. Data centers depreciate. Every dollar of capex today becomes a drag on operating margin over the following four to six years as it runs through the income statement, whether or not the revenue shows up on schedule. I have made the broader version of this argument in a piece on whether AI capital spending has run ahead of the revenue meant to support it.
Where the Margin Math Gets Tense
Microsoft’s operating margin has sat near the mid forties, which is remarkable for a company this size. Cloud gross margin has already ticked down as AI infrastructure costs entered the mix, and management has been candid that this pressure continues near term. The company is betting that two things happen fast enough to offset it: Azure AI consumption keeps compounding, and Copilot, the AI assistant bolted onto Office and Windows, converts from a product people trial into a line item companies renew without thinking about it.
Copilot adoption has been real but uneven. Some enterprises have rolled it out broadly. Others ran pilots, did not see the productivity lift they expected, and did not expand. The per-seat pricing is high enough that it is a real budget decision, not an impulse add-on. I would not call the monetization a failure, but I would not call it settled either, and the current stock price treats it as closer to settled than the evidence supports.
The Next Two Years, Written Down
Here is how I am framing the next two years. Fiscal 2026 is a spending year, with revenue growth in the low to mid teens and margins roughly holding as long as Azure stays strong. Fiscal 2027 is the test. That is when the depreciation from the current capex surge is fully loaded into the cost base, and it is also when Microsoft needs Copilot renewal rates and Azure AI revenue to be large enough that the incremental margin turns back up. If that happens, the stock grows into its valuation and the current multiple looks reasonable in hindsight. If Copilot monetization stalls while depreciation keeps climbing, earnings growth slows to single digits for a year or two and a premium multiple does not survive that. The wild card underneath all of it is electricity, and I have written separately on why power, not chips, is turning into the real constraint on AI, because it is the input most likely to make this build-out slower and costlier than the plan assumes.
| Scenario | Azure & Copilot outcome | Likely earnings path | What the stock does |
|---|---|---|---|
| Base case | Azure stays in the 30s, Copilot renewals hold | Low to mid teens EPS growth | Compounds, multiple roughly steady |
| Upside | AI consumption accelerates, Copilot re-rates as a standard spend | High teens EPS growth | Multiple expands, stock outruns the index |
| Downside | Copilot stalls, depreciation outpaces AI revenue | Single digit EPS growth for a year or two | Multiple compresses, flat to down |
What Would Get Me Buying Again
I still own Microsoft and I am not selling it. The business quality is not in question, the balance sheet is a fortress, and the company has earned the benefit of the doubt on big bets before. What I have done is stop adding. Buying a great company at a full price, where everything then has to go right, is the setup I try hardest to avoid, and it is roughly where Microsoft sits after the AI re-rating. The Microsoft quote page has the live multiple, and the quant rating tool makes the tension plain: strong Growth, stretched Valuation. I would get more aggressive on a pullback toward a high-twenties forward multiple, with Azure still growing and Copilot renewals holding. Short of that, I hold what I have and let fiscal 2027 settle the argument.
Gavin Thorne writes on equity strategy and company-level research. This article reflects his personal research process and is intended for informational purposes only. It does not constitute investment advice.