TL;DR — Key Takeaways
- Microsoft reportedly plans to expand its data center capacity from about 12 gigawatts today to more than 38 gigawatts by 2032.
- Capacity shortages have already constrained Microsoft’s cloud business and contributed to infrastructure issues affecting services such as GitHub.
- Only a portion of the expansion will be dedicated to AI accelerators, with CPUs and general-purpose systems remaining a major part of Microsoft’s infrastructure needs.
Microsoft is planning a major expansion of its data center network, with a goal of more than tripling computing capacity by 2032 as it works to overcome shortages that have constrained its cloud and AI businesses.
The company’s data center capacity is projected to grow from about 12 gigawatts today to more than 38 gigawatts in 2032. The plans include facilities that Microsoft owns or leases, but not capacity purchased from specialized cloud providers like CoreWeave.
The scale of the buildout highlights the vast compute requirements facing Microsoft. The company spent $145 billion on capital expenditures in its most recent fiscal year and expects $50 billion in capital spending for the first quarter of fiscal 2027. Its projected capital expenditures for calendar 2026 are $175 billion.
Still, Microsoft has struggled to put enough computing resources in place to serve demand. Concerns about building too much capacity prompted the company to some data center development around the beginning of 2025. That decision contributed to shortages that have limited Microsoft’s ability to accept new cloud business in some parts of the U.S. and Europe.
The costs of those limits are already apparent. Temu, a major Microsoft cloud customer, signed a large cloud contract with Oracle last year after Microsoft lacked sufficient capacity in locations the retailer wanted.
GitHub has also encountered infrastructure problems. The Microsoft-owned development platform suffered an outage lasting almost eight hours in August, with limited data center capacity playing a role. As AI drives greater volumes of software development, GitHub has used AWS infrastructure and, in some instances, served U.S. products from overseas data centers.
The exact scope of Microsoft’s expansion remains uncertain. Data centers take years to develop, and forecasts can change based on any number of variables. The growth figures were reported by Bloomberg, citing people familiar with Microsoft’s plans. Microsoft disputed the numbers after the report was published but did not provide alternative figures.
Faster Than Ever
Microsoft CEO Satya Nadella said in July that the company is “bringing capacity online faster than ever,” while also working to extract greater efficiency from hardware already deployed. The company is shifting some internal workloads to less congested regions to reserve capacity in high-demand locations for customers.
A major data center cluster known as East US 3 is under construction in the Atlanta area, partially to relieve pressure on the company’s heavily used Virginia infrastructure. About 300 megawatts of capacity is expected to become available in the region this year, with the total eventually exceeding one gigawatt.
Microsoft’s expansion is about more than GPUs. Only about 2 gigawatts of its current 12-gigawatt footprint is devoted to AI-specific chips. By 2032, AI-focused hardware is projected to account for roughly one-third of the planned 38 gigawatts.
The rest will support the CPUs and general-purpose systems needed for databases, cloud applications and other workloads. AI itself also requires CPU infrastructure to operate alongside GPUs. The Atlanta expansion, for example, will emphasize multipurpose CPU-based systems using processors from vendors including Intel.
Microsoft has plenty of competition in the infrastructure race. Amazon, Alphabet, Meta and Microsoft have collectively committed nearly $2.4 trillion to future spending, much of it directed toward data center equipment and leases.
That expansion faces a physical constraint that cannot be solved by faster processors: building data centers requires enormous amounts of power, land and time. Public resistance to new facilities has prompted political pushback in cities and states across the U.S.