Microsoft is significantly accelerating its data center expansion to tackle the intensifying shortage of computing power for AI and cloud services.

According to a Bloomberg report on Thursday, Microsoft plans to boost its global data center capacity to over 38 gigawatts (GW) by 2032, more than tripling the current level of approximately 12 GW. Sources indicate this projection encompasses both company-owned and leased facilities, excluding capacity sourced from "neocloud" providers like CoreWeave. The 38 GW target surpasses the peak electricity demand of New York State, underscoring the massive scale of the undertaking.

This ambitious capacity roadmap highlights the core bottleneck in Microsoft's AI infrastructure push: the constraint is not a lack of demand but a deficit of compute resources. Due to insufficient data center capacity, Microsoft has previously had to restrict new cloud service subscriptions in key regions across the US and Europe, inadvertently directing some potential business toward competitors.

Compute shortages are already impacting operations, prompting Microsoft to accelerate its catch-up strategy. The company's capital expenditure reached $145 billion in its latest fiscal year, with analysts forecasting continued growth in the coming years. Despite this heavy investment, supply pressures remain unresolved.

Bloomberg noted that Microsoft paused some data center development projects in early 2025, as then-CFO Amy Hood worried about potential overbuilding. However, as AI demand surged rapidly, many internal executives later regretted that decision, viewing it as a significant contributor to the current infrastructure bottleneck.

Microsoft is now moving swiftly to reverse course. CEO Satya Nadella stated in July that the company is "bringing capacity online faster than ever before," while also optimizing utilization of existing hardware to enhance compute output and profitability.

By 2032, AI-specific compute is projected to account for roughly one-third of capacity. Currently, only about 2 GW of Microsoft's 12 GW data center capacity is dedicated to AI-specific chips. By 2032, that allocation is expected to grow to approximately one-third of the total 38 GW.

Alistair Speirs, a Microsoft cloud infrastructure executive, emphasized that next-generation AI tools rely increasingly on CPU compute in addition to GPUs, stating that "you can't build great AI infrastructure with GPUs alone." A notable example is the East US 3 data center under construction near Atlanta. This project is expected to add around 300 megawatts (MW) of capacity this year, expanding to over 1 GW in the coming years, with total development costs reaching tens of billions of dollars. It will primarily support general-purpose computing with CPUs from manufacturers like Intel, rather than AI-specialized computing dominated by Nvidia chips.

Microsoft's compute challenges are not unique. Google, Amazon, and Meta are also making massive investments to secure data centers, power, and AI chips. Bloomberg reports that these four tech giants, the most aggressive in the race for computing resources, have collectively committed nearly $2.4 trillion in spending over the next several years, with the bulk allocated to data center equipment and leases.

However, data center expansion faces increasingly strong real-world constraints. Polls across the US show widespread local opposition to large data center projects, and governors in states like Texas and New York have even called for moratoriums on new server farm developments.

For Microsoft, the 38 GW figure represents a current roadmap, not an immovable final target. Data centers often take years to move from planning to operation, and rapid shifts in AI technology and customer demand could quickly alter compute requirements. The real test lies in whether Microsoft can convert this "paper capacity" into revenue-generating infrastructure in time to keep pace with the relentless growth of AI demand.