Global Data Center Capacity Forecast to Triple to 213 GW by 2031 as AI Demand Surges

The digital economy is colliding with the physical limits of the electrical grid. For years, cloud computing expanded without triggering national power crises because data centers became more efficient as they grew, keeping total energy consumption relatively flat. But artificial intelligence has broken that decoupling. The mechanism is straightforward: an AI training cluster requires dense, continuous computation that draws exponentially more power per square foot than traditional web hosting. Now, the physical infrastructure required to support this shift is coming into focus, and the numbers are staggering.[1][5]

The global colocation data center market—the physical buildings that lease space, power, and cooling to tech companies—is projected to more than triple its power capacity over the next five years. According to a new forecast from digital infrastructure analyst firm Structure Research, global operational capacity will surge from 63.8 gigawatts in 2026 to 213.1 gigawatts by 2031. This represents a compound annual growth rate of 27.3 percent, a pace of physical infrastructure buildout rarely seen outside of wartime or rapid national industrialization.[2]

To understand the scale of 213 gigawatts, it helps to look at national power grids. That figure is roughly equivalent to the entire electricity generating capacity of a major industrialized nation like the United Kingdom or Spain. It means the data center industry plans to build the equivalent of a G20 nation's entire power grid in just five years, driven almost entirely by the race to train and deploy frontier AI models.[1][6]

The mechanism driving this is a fundamental shift in how data centers are designed. Traditional enterprise cloud facilities spread servers out, drawing roughly 10 to 15 kilowatts of power per server rack. AI workloads, however, require thousands of specialized graphics processing units to be networked tightly together to minimize latency during model training. This density pushes power requirements to extremes. Structure Research notes that capacity exceeding 150 kilowatts per rack has entered the installed base for the first time in 2026.[2][4]

This density changes the physical requirements of the building itself. Air cooling, the standard thermal management system for decades, cannot dissipate the heat generated by a 150-kilowatt rack. Facilities must now pipe liquid coolant directly to the chips or submerge servers entirely. Consequently, liquid-capable data center capacity more than doubled in 2026, reaching 12.1 gigawatts globally, as operators retrofit existing buildings and design new ones around fluid dynamics rather than airflow.[2]

The capital flowing into this transition is immense. The combined revenue for colocation and interconnection services is forecast to jump from $151.9 billion in 2026 to $458.7 billion by 2031. This capital is reshaping global real estate and power markets, as hyperscale operators—the massive cloud providers like Microsoft, Google, and Amazon—secure land and grid connections years in advance to ensure they have the physical footprint necessary to support future AI generations.[2]

The geography of this buildout is also shifting. While Northern Virginia remains the undisputed global capital of data centers, ranking first in Structure Research's market tiering, the sheer volume of power required is forcing developers to look elsewhere. The firm assessed 101 global markets, classifying 21 as top-tier hubs. Asia-Pacific now holds the largest cohort of Tier 1 markets, including Tokyo and Singapore, as AI deployment creates new centers of network gravity far from traditional Silicon Valley hubs.[3]

However, the evidence supporting the 213-gigawatt forecast relies on a critical assumption: that local power grids can actually generate and deliver the electricity. This is where the forecast meets physical friction. In emerging European hubs like Madrid and Barcelona, which are projected to see massive growth, grid operators are struggling to distinguish credible data center projects from speculative power requests, prompting new legislation to manage grid queues.[2]

The International Energy Agency corroborates the demand trajectory, projecting that global data center electricity consumption could reach 945 terawatt-hours by 2030. In the United States alone, data centers are expected to account for half of all electricity demand growth over the next five years. This sharp increase changes the nature of the challenge for grid operators, who must balance the tech sector's demands with broader electrification efforts like electric vehicles and heat pumps.[5][6]

This creates a transparent uncertainty in the market. While the capital and the silicon exist to build 213 gigawatts of capacity, the transmission lines and generation facilities do not yet exist in many target markets. Grid constraints are already causing delays, prompting hyperscale operators to explore alternative power sources. Some are co-locating facilities directly at nuclear power plants, while others are investing in advanced geothermal energy to bypass congested regional grids.[4][6]

Ultimately, the data indicates that artificial intelligence is no longer just a software trend; it is the primary determinant of global heavy infrastructure. The transition from 64 gigawatts to 213 gigawatts will test the limits of supply chains, utility planning, and local zoning boards. As the decade progresses, the bottleneck of AI advancement is shifting away from semiconductor manufacturing and squarely onto raw electrical power.[1][2]