Microsoft plans to triple its data centre capacity by 2032
A compute crunch has cost the company business and closed deals.
Microsoft is short on compute, and it's costing the company deals. The solution? A plan to more than triple its data centre capacity from 12GW to 38GW by 2032.
The news came from a Bloomberg report yesterday and refers to Microsoft's global computing network. Here's what it means, and the implications.
Compute shortage
Citing people in the know, Bloomberg says Microsoft plans to more than triple its data centre capacity, taking it to 38GW in 2032, up from 12GW today.
Microsoft has disputed the planning numbers with Bloomberg, but declined to provide a figure of its own.
What is undisputed, though, is that a shortage of computing has become a significant impediment for Microsoft. It has been unable to close some deals, and has lost business as a result.
Only a third is for AI
According to the report, just 2GW of data centre capacity currently goes towards powering AI-specific chips. The plan is to grow that significantly, to a third of the 38GW, or 12GW.
Which means the pressure is clearly on traditional workloads such as cloud servers and databases. This is already evident at Microsoft-owned units such as GitHub and Xbox. GitHub has faced crashes attributed to data centre capacity, while Xbox has started capping its cloud-based, all-you-can-play game streaming service.
Microsoft's push to build will no doubt be complicated by others looking to snap up AI and cloud capacity, as well as by growing pushback against data centres globally.
The quiet build-out never stopped
Unlike data centre operators and tech startups, the cloud and tech giants, which the industry calls hyperscalers, rarely announce their data centre plans publicly.
But despite the secrecy and the NDAs, it is impossible to keep a lid on everything, especially when you consider the massive ecosystem of consultants, professionals, and vendors needed to build these facilities.
This report explains a phenomenon I've observed for a while now: hyperscalers have never stopped growing in this region, whether in Malaysia or in other markets across Southeast Asia and APAC. I've chatted with suppliers shipping specialised systems and components to hyperscalers in Japan, Thailand, and Australia. And it makes sense, because digitalisation continues to be a solid demand driver.
Then there is AI, which I believe will take off with the rise of "useful AI." But another area that has seen scant coverage is how AI contributes massively to traditional compute demand. That's a story I'll return to in my commentary this weekend.
