Microsoft Plans to Triple Data Center Capacity to Over 38 GW by 2032

The scale of the buildout matters less than the exclusion of neocloud compute from the figure: Microsoft is signaling that owned and leased capacity, not rented GPU time, is the strategic base of its AI infrastructure. That distinction shapes how the broader cloud and AI compute supply chain reads demand signals, since third-party capacity becomes a flex layer rather than a foundation. Whether the roadmap holds depends on construction timelines and whether customer preferences or new technology shift the assumptions baked into it.