Microsoft is planning to more than triple its data center capacity, to more than 38 gigawatts in 2032 from about 12 gigawatts now, according to a Bloomberg report citing people familiar with the plans. Bloomberg reports that the road map covers company-owned and leased facilities, excludes computing power rented from so-called neoclouds such as CoreWeave, and could change because server farms take years to develop and customer preferences and technology shift in the meantime.
The company does not endorse the figures. Bloomberg reports that after the story was published, Microsoft spokesperson Frank Shaw said the planning numbers were not accurate, while declining to provide other figures beyond what the company disclosed in its most recent earnings report. That leaves a buyer with a range rather than a number: the company's own disclosures at one end and Bloomberg's sourced projection at the other.
Behind the number is a shortage. Bloomberg reports that a lack of available servers has been Microsoft's biggest impediment of late, made worse by a decision to pause some data center development around the start of 2025, and that the company restricted new cloud subscriptions at a number of key server farm hubs in the United States and Europe, according to documents Bloomberg reviewed. Some customers went elsewhere: the report says Chinese retailer Temu signed a large contract with Oracle's cloud last year after it could not get additional capacity from Microsoft in the regions it wanted.
The relief valve Bloomberg identifies is a new cluster of data centers under construction in the Atlanta area, called East US 3, which Microsoft cloud infrastructure executive Alistair Speirs said should help ease a capacity shortage in Virginia, the company's largest hub. It "essentially makes more capacity available everywhere when we open up a new data center," Speirs said in an interview with Bloomberg. The operational detail behind that sentence: about 300 megawatts is due to come online in the region this year, expanding to more than a gigawatt in the coming years, according to people familiar with the project, and the site is built around CPU-based servers from companies such as Intel rather than the AI chips typically supplied by Nvidia.
That mix is the part buyers should note. Bloomberg reports that only about 2 gigawatts of Microsoft's current 12-gigawatt footprint is centered on AI-specific chips, expected to grow to about a third of the 38 gigawatts, and that most facilities are filled with central processing units running databases and cloud applications rather than training models. Newer AI tools consume growing amounts of CPU capacity alongside GPUs, Speirs told Bloomberg, so all types of computing power need to expand. "GPUs by themselves don’t make great AI infrastructure," Speirs said.
Strain is already visible in services buyers use. Bloomberg reports that GitHub crashed for almost eight hours in August in large part because of a data center capacity shortage, that the coding platform has relied on other providers including Amazon Web Services, and that the Xbox unit told subscribers it would start capping cloud game streaming time and charging those who exceed the new limits.
For an infrastructure buyer, the useful reading is that Microsoft's own executive describes capacity relief in terms of one new region and a multi-year ramp, not an immediate change to what can be provisioned in Virginia or Europe. Anyone sizing an Azure commitment now has a reason to treat regional availability as a dated contract term rather than an assumption.
Leave a Reply