The shortage of computing power drives AI infrastructure expansion, Microsoft plans to double its data center capacity to 38GW

Microsoft is significantly ramping up data center construction to address the ongoing shortage of AI and cloud computing power.

According to a Bloomberg report on Thursday the 10th (ET), Microsoft plans to increase its global data center capacity to over 38 gigawatts (GW) by 2032, more than tripling from the current approximately 12 GW. Sources indicate this plan includes both Microsoft-owned and leased data centers, but does not count compute rented from "new cloud" providers like CoreWeave. A capacity of 38 GW would even exceed peak electricity demand in New York State.

This rare scale of planning also highlights the core bottleneck in Microsoft's expanding AI infrastructure: the challenge is not a lack of demand, but a shortage of computing power. Due to insufficient data center capacity, Microsoft previously had to restrict new cloud service subscriptions in certain key regions in the US and Europe, causing some potential business to flow to competitors.

Computing Power Shortage Is Already Dragging Down Business; Microsoft Accelerates the Catch-Up

Microsoft’s capital expenditure in the most recent fiscal year has already reached $145 billion, and analysts anticipate further growth in the coming years. However, this massive investment has not fully relieved the pressure on compute supply.

Bloomberg reports that Microsoft previously paused some data center development projects at the start of 2025, as then-CFO Amy Hood worried the company might overbuild. As AI demand exploded, several executives later expressed regret over this decision, seeing it as one of the main reasons for the current data center bottleneck.

Microsoft is now rapidly reversing this situation. In July, Microsoft CEO Satya Nadella stated that the company is "bringing capacity online at the fastest pace in our history," while simultaneously boosting output and profitability by improving the efficiency of existing hardware utilization.

By 2032, AI-Dedicated Compute Capacity Will Make Up About One Third

Microsoft’s new plan is not simply betting on GPUs but expanding both general-purpose and AI computing power.

Currently, only about 2 GW of Microsoft’s roughly 12 GW data center capacity is dedicated to AI chips; by 2032, AI compute is expected to make up about one-third of the 38 GW total capacity.

Alistair Speirs, an executive with Microsoft Cloud Infrastructure, said that, aside from GPUs, the next generation of AI tools is increasingly reliant on CPU computing power, so "GPUs alone cannot build a great AI infrastructure."

A typical example is the East US 3 data center currently being built in the Atlanta, US area. This project is expected to add around 300 megawatts of capacity this year, expanding to over 1 GW in the coming years, with total development costs reaching tens of billions of dollars. Its main focus is on general-purpose computing using CPUs from vendors like Intel, rather than primarily using NVIDIA AI chips for dedicated AI workloads.

The Four Tech Giants to Invest Nearly $2.4 Trillion in the Coming Years

The pressure Microsoft faces is not unique. Google, Amazon, and Meta are also spending heavily to compete for data centers, electricity, and AI chips.

Bloomberg reports that these four tech giants, which are most aggressively competing for computing resources, have already committed to invest a combined total of nearly $2.4 trillion in the coming years, with the vast majority earmarked for data center equipment and leasing.

At the same time, data center expansion is also facing increasingly tough real-world constraints: polls in many US regions show widespread public opposition to the construction of large data centers locally, and governors in states such as Texas and New York have even called for a halt to new server park projects.

For Microsoft, 38 GW is only the current plan—not the final, unchangeable target. It typically takes years for a data center to go from planning to operation, while fast-evolving AI technology and shifting customer demand can quickly alter the compute landscape. The real test is: as AI demand continues its rapid growth, can Microsoft turn this “paper computing power” into real capacity that generates revenue in time?