Microsoft's 2026 Capex Forecast Fell $15B on an Accounting Change, Not a Spending Cut
Microsoft reported fourth-quarter capital expenditure of $41 billion, up 70% year over year, and simultaneously lowered its 2026 capex forecast to $175 billion from $190 billion. The company was explicit that spending plans have not changed. An accounting change did. That combination deserves more attention than it will get, because a $15 billion reduction in a guidance figure that everyone on the street uses as a proxy for compute added is not a reduction in compute added. It is a reclassification.
The rest of the print was strong enough to bury the detail. Revenue of $90.1 billion, up 18% against an $87.62 billion consensus. Azure and other cloud services up 43% versus roughly 40% expected, with fiscal 2026 Azure revenue crossing $100 billion for the first time. Microsoft 365 Copilot at more than 30 million paid seats against more than 20 million a quarter earlier. The stock rose more than 7% after hours. Against that, the offsetting weakness was in the consumer hardware tail: Windows OEM and Devices revenue down 7%, Xbox hardware down 13%, Xbox content and services down 10%. Nobody owns this stock for Xbox.
The capex line is where the analytical work is. Spending that moves off the capital account does not evaporate. It reappears as lease obligations, as prepaid capacity, as cost of revenue, or as some hybrid structure that puts the asset on someone else’s balance sheet and the payment in Microsoft’s operating expense. Each of those routes has a different consequence for the income statement than a purchased server does. A purchased asset depreciates over a stated life and hits below the gross margin line. A leased or contracted equivalent tends to land closer to cost of revenue and compresses the cloud gross margin directly, quarter by quarter, without ever appearing as a depreciation surprise. Investors who spent the last year arguing about useful-life assumptions were watching the wrong line if the spend is migrating out of the account those assumptions apply to.
This also breaks cross-company comparison, which is the practical damage. The hyperscaler capex table that gets rebuilt every quarter now contains at least one number governed by a different policy than the others. Meta, in the same session, reported free cash flow of $784 million, down 91% year over year, and raised the bottom of its 2026 capex range to $130 billion from $125 billion while leaving the top at $145 billion. Meta is putting the spend through the capital account and eating the free cash flow consequence in plain view. Microsoft is guiding to a smaller number while spending the same money. Those two disclosures are not comparable, and a reader who ranks them by headline capex will conclude the opposite of what is happening.
The demand evidence sitting underneath all of it is not in dispute. Brookfield and NextEra announced a $100 billion campus with more than 1.2 gigawatts at a former Department of Energy uranium-enrichment site in Kentucky, targeted to open in 2032. That is a seven-year construction horizon being committed today, which is the clearest possible statement about who thinks the compute demand is durable and who is willing to finance it. Megawatts contracted and delivered is the honest metric. Capex guidance no longer is.
The moat question resolves in Microsoft’s favor and not through infrastructure. Azure at 43% growth on a base above $100 billion is impressive, but the durable asset is the 30 million paid Copilot seats attached to enterprise agreements that renew on multi-year cycles with switching costs measured in identity systems and document estates. That is a distribution moat, and it is the reason the company can afford to spend $175 billion or $190 billion without the market questioning the return. It is also, incidentally, the thing the UK Competition and Markets Authority has now opened an investigation into, on the theory that customers paid more to renew Microsoft 365 after Copilot was added. Regulators found the moat before the sell side finished modeling it.
The line to reconcile in the 10-K is the finance and operating lease disclosure against the $175 billion figure. If the delta shows up there, the guidance cut was cosmetic and the cloud gross margin is the number at risk.