The Zhitong Finance App learned that Microsoft Corporation (MSFT.US) plans to more than double the data center capacity. This move is expected to help the company overcome the shortage of computing power. It is this problem that has forced Microsoft to shut out some artificial intelligence and cloud businesses. According to people familiar with the matter, Microsoft's global data center network capacity will exceed 38 gigawatts by 2032, compared to about 12 gigawatts currently. This would exceed New York State's electricity consumption during peak periods.
These numbers rarely reveal how much computing power Microsoft can expect to get after investing heavily in data centers. Microsoft's capital expenditure reached 145 billion US dollars in the most recent fiscal year, and analysts expect this expenditure to continue to grow in the next few years.
According to the person familiar with the matter, this roadmap includes the company's own and leased facilities, but does not include computing power leased from so-called new cloud service providers such as Coreweave Inc. (CRWV.US). These sources cautioned that the plan could change as server farms take years to develop, and changes in customer preferences and new technology could disrupt current assumptions.
Microsoft and its peers must also deal with strong opposition to data centers. Polls show that most Americans oppose building data centers in their own communities. A number of elected officials, including the governors of Texas and New York, have ordered the suspension of the new server farm project.
Shortage of computing power is Microsoft's biggest obstacle recently. This problem was exacerbated by the company's decision to suspend development of some data centers around the beginning of 2025. The lack of usable servers has prompted some customers to move their new business elsewhere, frustrating Microsoft sales people who are anxious to promote cloud and artificial intelligence services, and also displeased investors who want to see the return of huge investments in the AI field.
Competitors such as Google (GOOGL.US), a subsidiary of Alphabet Inc. are also trying to keep up with demand and are investing huge sums of money to try to change this situation. In this race for computing power, the four major players — which also include Amazon (AMZN.US) and Meta Platforms Inc. (META.US) — have committed a total of nearly $2.4 trillion over the next few years, mainly for data center equipment and leasing.
More and more leaders within Microsoft expressed the hope that the company will finally begin to overcome the problem of limited computing power. CEO Satya Nadella (Satya Nadella) said in July that the company is “bringing computing power online faster than ever before” and is trying to improve the efficiency of existing hardware, thereby increasing profitability.
The most concrete sign of Microsoft's optimism is a new data center cluster being built in the Atlanta area. Cloud infrastructure executive Alistair Speirs said the project, called East US 3, should help mitigate the shortage of computing power in Virginia, Microsoft's largest data center hub.
In an interview, Speirs said, “When we launch a new data center, we can basically free up more computing power everywhere.” He said the company is also moving the development of some of its own products and other in-house work to less crowded data centers to make room for customers in their favorite hubs.
According to people familiar with the matter, about 300 megawatts of data center electricity will be brought online in the region this year, which will expand to more than 1 gigawatt in the next few years. One of the sources said that the total development cost of the project was tens of billions of dollars. The project will focus on CPU-based multi-purpose computing, using servers from companies such as Intel (INTC.US) instead of AI chips commonly provided by Nvidia (NVDA.US).
More than three years after the rise of the artificial intelligence boom, the vast majority of Microsoft's data centers are still dominated by general computing. Most facilities are equipped with central processing units (CPUs) that drive databases and cloud applications, rather than chips to train AI models.
According to the above sources, out of Microsoft's current 12 gigawatts of capacity, only about 2 gigawatts are concentrated on AI-specific chips. They said that this ratio is expected to grow to about one-third of the 38 gigawatts that Microsoft plans to launch by 2032.
According to Speirs, in addition to using GPUs, newer AI tools are also increasingly using CPU servers, which means Microsoft needs to expand all types of computing power. “GPUs alone can't build a great AI infrastructure,” he said.
Microsoft is still struggling to overcome the suspension of data center construction initiated by Chief Financial Officer Amy Hood (Amy Hood) almost two years ago. Hood was worried at the time that the company was at risk of overconstruction. Many senior leaders later regretted the decision and blamed it in part on bottlenecks the company is still dealing with.
Hood said on Wednesday that Microsoft is putting data centers into operation faster than before. She told investors at an event hosted by Goldman Sachs that it's about shortening the time it takes “from the time we go online to the hardware ready to generate revenue.”
Microsoft has limited new cloud subscriptions at key server farm hubs in the US and Europe due to shortages, according to documents.
In some cases, customers choose to sign contracts with Microsoft's competitors. For example, according to people familiar with the matter, the Chinese retailer Temu is an important cloud customer, but last year it signed a major contract with Oracle Corporation (ORCL.US) for cloud services after not being able to get the additional capacity it wanted from Microsoft. Temu did not respond to requests for comment.
In addition to curbing sales, server shortages have also been known to cause system failures. In August, Microsoft's programming platform GitHub crashed for nearly eight hours, largely due to lack of data center capacity.
According to a source familiar with the matter, GitHub has to rely on other data center providers, including Amazon Web Services (Amazon Web Services), in the face of huge usage growth brought about by people turning to AI to generate more code. The source said that GitHub sometimes has to support US products from overseas data center regions because nearby sites lack service capabilities, and this practice may make the software run slower.
GitHub Chief Operating Officer Kyle Daigle said in an interview that in order to cope with exponential traffic growth, GitHub is adding more capacity and redesigning the system to improve efficiency.
Microsoft's gaming business is also clearly showing signs of pressure. The Xbox division previously provided cloud game streaming services for an unlimited period of time. Subscribers were told last week that they would begin limiting game time and charging users who exceed the new limit.
The big question facing Microsoft is whether it can build enough data centers in a timely manner to meet the rapidly growing demand for cloud and AI services. Notably, some of the company's online capacity today was initially planned in the early days of the pandemic, when demand for working from home drove Microsoft to expand its server farm network.
“We're always thinking long term when it comes to building this infrastructure and how to responsibly meet growing demand,” Speirs said.