Microsoft pledges to double AI infrastructure capacity in two years, signals higher AI-related capex

Microsoft is spending like a company that thinks AI demand is real, durable, and still under-served. That is the story here. Not a vague “we’re bullish on AI” story — a concrete one about datacenters, GPUs, leases, power, and a balance sheet that now has to carry all of it. In Microsoft’s April 29, 2026 fiscal Q3 results, Satya Nadella and Amy Hood made clear that the company is still in build mode because customer demand is outrunning available capacity. (microsoft.com) ### What did Microsoft actually say? The key line did not come out of nowhere this week — it comes from the company’s own FY26 setup. In Microsoft’s October 29, 2025 fiscal Q1 call, Nadella said Microsoft would increase total AI capacity by more than 80% in fiscal 2026 and “roughly double” its total datacenter footprint over the next two years. That is the pledge people are reacting to now because the Q3 numbers show Microsoft is still following through, not backing off. (microsoft.com) ### Why is this bigger than normal cloud capex? Because this is not just routine server refresh. Microsoft is building the physical layer for AI inference, training, and its own first-party products at the same time. In Q1, the company said roughly half of capex went to short-lived assets like GPUs and CPUs, while the rest supported longer-lived cloud infrastructure. That mix matters — it me(microsoft.com)around it. (microsoft.com) ### What do the latest numbers show? The latest quarter was huge. Microsoft posted $82.9 billion in revenue for the quarter ended March 31, 2026, up 18% year over year, and said its AI business had surpassed a $37 billion annual revenue run rate, up 123%. Capital expenditures were $31.9 billion in Q3, with Hood saying the quarter-to-quarter dip was just timing noise from infrastructure buildouts and finance lease delivery. (microsoft.com) ### So is demand really there? Basically, yes. Microsoft Cloud revenue reached $54.5 billion in Q3, up 29%, and commercial remaining performance obligation almost doubled to $627 billion. That backlog is the tell. Customers are committing ahead of delivery, which usually means they expect scarce capacity and want to lock in supply. Microsoft has been saying for multiple quarters that AI demand is stronger than what it can currently serve. (microsoft.com) ### Where does Australia fit in? Australia is one example of how broad this build-out has become. Microsoft said last week it is making a A$25 billion commitment in Australia covering AI and cloud capacity, cybersecurity, and skills. Separate Microsoft materials say the company is building datacentres in the ACT, New South Wales, and Victoria, and that its Australian datacentre footpr(microsoft.com)5 billion push announced in 2023. (news.microsoft.com) ### Why are people suddenly talking about efficiency? Because once capex gets this large, every layer above the hardware starts to matter more. Model routing, token use, inference efficiency, storage layout, utilization rates — all of that turns into margin protection. The catch is that Microsoft can keep winning demand and still feel pressure if the cost of serving that demand s(news.microsoft.com)re-efficiency story. (microsoft.com) ### Does this change what Microsoft is? Not completely — but it does stretch the old picture. Microsoft still sells high-margin software. But the AI era is forcing it to behave more like a hybrid of software vendor, cloud landlord, and industrial builder. When a company is guiding to another heavy capex quarter while saying margins can still rise, it is signaling confidence that the revenue engine is catching up to the spending engine. (microsoft.com) ### Bottom line? Microsoft is not hinting at an AI infrastructure boom. It is already financing one. The real question now is not whether the company will spend — it is whether it can turn all that steel, silicon, and power into durable profit before the next capacity crunch hits.