AI

Microsoft cut $15 billion from its AI budget. It didn’t cut anything.

Starting in fiscal 2027, Microsoft will depreciate data centers over 25 years instead of 15 — and more future leases move off the capex line entirely. Calendar-2026 capital spending falls from roughly $190 billion to about $175 billion. The company says the underlying investment plan is unchanged. The stock rose 15%.

N Noah · The Sharp Brief · July 30, 2026 · 4 min read
Rows of server racks in a data center hall with an accountant's ledger and calculator in the foreground

Buried near the end of Microsoft’s fiscal fourth-quarter call Wednesday, before the outlook section, CFO Amy Hood made an accounting disclosure: effective at the start of fiscal 2027, the company is extending the estimated useful life of its data centers and office buildings from 15 years to 25 years, reflecting, in her words, its operating history and expected use of those assets. She added that the change affects only the timing of future depreciation.

That is true, and it is also not the part that matters. The larger consequence is downstream: because the assumed life of the asset is longer, more of Microsoft’s future data center leases get classified as operating leases rather than finance leases. Finance leases count toward reported capital expenditures. Operating leases do not. The company promptly lowered its calendar-2026 capex outlook from roughly $190 billion to about $175 billion — and said explicitly that, excluding the effect of the useful-life change, its calendar-2026 investment forecast is unchanged.

So: $15 billion came off the number without a dollar coming out of the buildout. Investors, who spent this earnings season punishing AI capital spending, sent the stock up about 15% Thursday. Revenue of $90.01 billion against a street near $87.62 billion helped. So did Azure growth of 43%. But the capex line not blowing past expectations was the thing the market said it wanted, and Microsoft delivered it partly through the depreciation schedule.

Our take: None of this is improper — it was disclosed on the call, the reasoning is defensible, and Hood flagged the FY27 operating-income benefit as minimal. But it does mean the single metric the market has been using to judge AI discipline just became less comparable, at the exact moment the market started using it. Microsoft and Meta reported the same night and only one got paid; Alphabet fell 7% on a capex raise to $205 billion. Headline capex is now the scoreboard, and one player just changed how his points are counted. Watch whether the rest of the field follows. Depreciation assumptions are the most quietly powerful lever in hyperscaler earnings: stretch the life of a GPU hall, and yesterday’s expense becomes tomorrow’s problem.

The numbers that didn’t move

Strip the accounting out and the buildout is accelerating, not slowing. Capital expenditures and finance leases for the quarter came in at $41 billion, up 69% year over year. Management guided quarterly capex to top $50 billion in fiscal Q1 2027. Azure is guided to 45% growth in constant currency.

The physical footprint tells the same story. Satya Nadella said Microsoft brought 88 data centers online across fiscal 2026, including 31 in the June quarter alone spanning five continents, adding roughly a gigawatt of capacity in the quarter and remaining on track to roughly double total capacity within two years. Time from loading dock to live GPU in its largest regions fell by nearly half over the year.

That is a company spending harder and installing faster while reporting a smaller capex number. Both things are real. Only one of them is on the slide investors screenshot.

What to watch

The AI trade spent this week deciding it wanted evidence of spending discipline. Microsoft gave it a $15 billion reduction and a 69% increase in the same report. Read the second one.

Advertisement

Get the day, decoded — at 7 PM ET

The Sharp Brief: AI, money, business & performance in five sharp minutes. Free.

Free bonus: subscribe today and The 2026 AI Playbook (PDF) lands with your welcome email.

Recommended by 5+ newsletters across AI, markets & business.