Microsoft targets 38 GW data‑center capacity to triple cloud power by 2032

SentiSense · Published · Updated

Bloomberg reported, and Reuters relayed, that Microsoft plans to expand data-center capacity to roughly 38 gigawatts by 2032, more than triple its current 12 gigawatt footprint. Only about 2 of those 12 gigawatts are AI-specific today, a share expected to reach about a third of the 38 gigawatt target. Microsoft guides to $50 billion of capital expenditure in its fiscal first quarter of 2027 and roughly $175 billion across calendar 2026, and is extending data-center leases to over 25 years from 15.

MSFT is planning to build its data-center footprint out to roughly 38 gigawatts by 2032, more than triple the about 12 gigawatts it runs today, according to a Bloomberg report relayed on the Reuters wire. For scale, the wire noted that 38 gigawatts would exceed the peak electricity consumption of New York state.

The AI portion is where the shift is sharpest. Only about 2 of the current 12 gigawatts is centered on AI-specific chips, a share expected to grow to roughly a third of the 38 gigawatts Microsoft plans to have online. The spending behind it is already guided: about $50 billion of capital expenditure in the fiscal first quarter of 2027 and roughly $175 billion across calendar 2026. Microsoft is also extending long-term data-center leases to over 25 years rather than 15, which pushes the commitment further out and signals confidence that the demand is structural rather than a cycle.

This is as much a constraint story as a growth story, and the two are causally linked. Reporting around the plan cites real business lost to capacity: Temu moved a major cloud deal to Oracle because Microsoft could not supply capacity in preferred regions, GitHub suffered an eight-hour outage in August tied to server constraints and rerouted to AWS, and Microsoft restricted Xbox cloud-gaming usage for paying subscribers. The Information framed the same plan around that shortage rather than around growth. Executives had temporarily paused select data-center developments in early 2025 over overbuilding concerns, and this buildout reads as the correction to that pause.

The read-through runs well past Redmond. A commitment of this size is demand visibility for the power, cooling, networking and accelerator supply chains, and it raises the bar for what the hyperscaler capex line looks like through the rest of the decade. Two caveats belong on it: the underlying Bloomberg feature sits behind a paywall, so the figures here come through the Reuters wire summary rather than the original, and a 2032 target is a plan rather than a booking. What to watch is whether the quarterly capex guide tracks toward the implied run rate, and whether the capacity shortfalls that cost Microsoft deals this year narrow or persist.

Powered by SentiSense - Intelligent Market Analysis