MSFT · AI infrastructure profile
Microsoft AI Capex in 2026: $190B Spending Guide
Calendar 2026 guidance
About $190B
Companywide capital expenditure. It is not a disclosed pure-AI total.
- Research desk
- Yield Theory Research
- Reviewed
- Evidence
- 2 external references · Method
- 2026 guidance
- About $190B
- Calendar 2026
- FY2026 Q3 capex
- $31.9B
- Roughly two-thirds short-lived assets, primarily GPUs and CPUs
- Sources
- 2
- Primary company materials
- Reviewed
- 2026-08-10
- Refresh after material guidance changes
Microsoft 2026 capex guidance
Microsoft expects roughly $190 billion of calendar-year 2026 capital expenditure. Management says approximately $25 billion of that envelope reflects higher component pricing, while demand for Azure capacity remains above available supply. The number is a company capital-spending plan, not a disclosed pure-AI total.
What the number includes
Microsoft's reported capex includes additions obtained through finance leases as well as purchased property and equipment. That makes total capex different from cash paid for property and equipment in a quarter. The company also reports on a June fiscal year while the $190 billion guidance is explicitly framed on a calendar-year basis, so fiscal-year comparisons require normalization.
- Short-lived assets such as GPUs and CPUs that can enter service and support revenue relatively quickly.
- Long-lived data-center sites and infrastructure intended to support monetization for 15 years or more.
- Storage, networking and replacement equipment needed to keep Azure and first-party products operating.
How investors should read the spending
Microsoft provides the clearest split between short-lived compute and long-lived infrastructure among the four companies in this tracker. That split matters because servers generally begin affecting depreciation and capacity sooner than a large campus. The central question is how quickly contracted cloud demand and product usage convert the new capacity into revenue relative to the depreciation and operating costs arriving behind it.
What to monitor next
- Quarterly capex and the proportion allocated to short-lived compute versus long-lived sites.
- Finance-lease additions versus cash paid for property and equipment.
- Azure capacity constraints, backlog conversion and management's timing for bringing capacity online.
- Depreciation growth and cloud gross-margin pressure as recent investments enter service.
Comparison limits
- The company does not disclose a standalone audited AI-capex figure.
- Calendar-year guidance does not align directly with Microsoft's June fiscal year.
- Finance-lease timing can make total capex volatile and different from cash investment in the same quarter.
Primary sources
- Microsoft FY2026 Q3 earnings call ↗2026-04-29
- Microsoft FY2026 Q2 earnings call ↗2026-01-28
Guidance can change each quarter. Yield Theory preserves the company's accounting definition and does not relabel the entire amount as AI spending. Read the research methodology for our sourcing and correction standards.
Compare hyperscaler capex
Continue the analysis
The capex number is the start of the thesis.
Members get the source-linked investment case, catalysts, risks, and evidence that would change the view.
$15/mo or $150/yr · cancel future renewals anytime · sources included