Microsoft is spending more on data centers than many countries spend on infrastructure — $80 billion in nine months. The whole bet rests on AI demand staying this hot.
↓ the brief below
The size of the bet · capex, nine months
$80.1
B
Spent on property and equipment in nine months — up 69% from a year earlier. Almost all of it is AI data centers and the chips inside them.
Source · 8-K · Item 2.02 — Cash Flows Statements (additions to property and equipment) · 9M FY26 (nine months ended Mar 31, 2026) · Filed Apr 29, 2026
From the filing · Microsoft's own warning
Results could differ materially because of factors such as significant investments in products and services that may not achieve expected returns.
↳ The company itself flags it: $80B of spending only pays off if the AI revenue keeps coming. If demand cools, that's a mountain of expensive, half-used data centers.
Software used to cost Microsoft almost nothing to deliver. AI is different: the data centers cost a fortune, and that cost becomes a charge against profits for years (depreciation) — whether or not the AI revenue shows up.
Wall Street calls this
Capital intensity
It's why a wrong-sized bet wouldn't just waste cash once — it would weigh on profits for a decade.
From the filing · and they're not alone
Actual results could differ materially because of factors such as intense competition in all of our markets, and our focus on cloud-based and AI services presenting execution and competitive risks.
↳ Amazon and Google are pouring in just as much. If everyone builds at once, the AI capacity could outrun the demand — and prices fall for all of them.