Microsoft Plans 38GW Data Center Expansion to Overcome AI Compute Shortages

Microsoft’s contracted but undelivered cloud-services backlog reached $678 billion, an 84% increase in one year; roughly 45% of that booked pipeline is tied to OpenAI.
In Microsoft’s most constrained regions, demand is reportedly about three times available capacity, while data-center construction timelines have stretched from roughly six months to more than 18 months.
Capacity shortages have already driven customers and Microsoft services to seek alternatives: Temu reportedly signed a major cloud contract with Oracle after failing to secure sufficient Microsoft capacity, GitHub rented server space from Amazon Web Services during an outage, and Xbox cloud gaming imposed usage limits to manage server loads.
Microsoft’s planned 38-gigawatt network would consume more electricity than New York state uses during peak-demand periods, and the plan includes both company-owned and leased facilities but excludes capacity rented from specialized neocloud providers such as CoreWeave.
Microsoft’s expansion follows a temporary construction slowdown: Chief Financial Officer Amy Hood halted some data-center projects over overcapacity concerns early last year, a decision that analysts say contributed to later infrastructure shortages.
Microsoft is racing to triple its global data-center capacity to about 38 gigawatts by 2032, up from roughly 12 gigawatts today, as demand for artificial intelligence tools and cloud services vastly outpaces available infrastructure Yahoo Finance. The company has already restricted new subscriptions in some regions and lost business due to shortages of electricity, transformers, and cooling systems, according to Eastern Herald.
Microsoft's contracted but undelivered backlog has surged to $678 billion — an 84% increase in just one year — with roughly 45% tied to OpenAI commitments Yahoo Finance. In the company's most constrained regions, demand is about three times the available capacity, forcing customers like Temu to seek alternatives and major construction timelines to stretch from six months to over 18 months Eastern Herald.
Today, only about 2 gigawatts of Microsoft's 12-gigawatt capacity runs AI chips. By 2032, roughly one-third of the expanded 38-gigawatt network — about 12-13 gigawatts — will be dedicated to AI Yahoo Finance. This shift reflects the explosive growth in demand for OpenAI's models and other artificial intelligence tools, which require far more computing power than traditional cloud services Eastern Herald.
Microsoft's infrastructure gaps have forced major customers to seek alternatives. Temu signed a major cloud contract with Oracle after failing to secure sufficient Microsoft capacity Eastern Herald. During a recent outage, GitHub rented server space from Amazon Web Services to keep services running. Xbox cloud gaming also imposed usage limits to manage server loads Eastern Herald.
Microsoft's planned 38-gigawatt network would consume more electricity than New York state uses during peak demand, according to Eastern Herald. The expansion includes both company-owned and leased facilities, though it excludes capacity rented from specialized providers like CoreWeave Eastern Herald. The company plans to spend $50 billion on capital expenditure to support this growth Yahoo Finance.
Several factors could slow Microsoft's expansion. Permitting delays, grid constraints, and community opposition may block projects in some regions Yahoo Finance. Shifts in technology or customer demand could also reshape priorities. Notably, Microsoft's CFO Amy Hood halted some data-center projects early last year due to overcapacity concerns — a decision analysts say contributed to later infrastructure shortages Eastern Herald.
Publishers
14
Articles
28
Reach
42