An Unprecedented Spending Spree
The numbers behind the AI infrastructure build-out are difficult to comprehend. Projections from PricewaterhouseCoopers suggest that global spending on data centres could reach a monumental $31.6 trillion by 2050. To put that in perspective, this figure
dwarfs the historical investment cycles of railways, electrification, and even the initial build-out of the internet. On an annual basis, capital expenditure is forecast to climb from around $800 billion in 2026 to $1.1 trillion by 2030. The primary drivers are the tech industry's biggest names—Microsoft, Amazon, Alphabet (Google), and Meta—which collectively spent over $400 billion in 2025 and are on track to spend a combined $745 billion to $760 billion in 2026 alone. This isn't a short-term trend; unlike past infrastructure projects that tapered off, the AI cycle is expected to accelerate, as the specialised chips and servers at the heart of these facilities need to be replaced every four to six years.
Why AI Needs Its Own Fortresses
Not all data centres are created equal. An AI-optimised facility is a different beast entirely from a traditional one used for cloud storage or web hosting. The key difference lies in the hardware and the immense power density required. AI models, particularly during the training phase, rely on thousands of interconnected Graphics Processing Units (GPUs) that work in unison. These high-performance chips, primarily supplied by companies like Nvidia, consume far more energy and generate significantly more heat than the Central Processing Units (CPUs) of the past. A single server rack in a conventional data centre might draw 10-15 kilowatts of power; an AI rack can demand anywhere from 50 to over 100 kilowatts. This has a direct impact on construction costs, with AI-ready facilities costing $20 million or more per megawatt to build, compared to the $7-12 million per megawatt for older designs.
The Power Problem
This insatiable demand for computation translates directly into an enormous appetite for electricity. In 2024, data centres already accounted for about 1.5% of global electricity consumption. With the AI boom, that figure is set to surge. The International Energy Agency projects that electricity demand from data centres will more than double by 2030. Some forecasts suggest AI alone could drive a 160% increase in data centre power demand by the same year. An AI-focused hyperscale facility can consume as much electricity as 100,000 homes or more. This immense strain is not just an environmental concern; it's a primary logistical bottleneck. The availability of affordable, reliable, and increasingly low-carbon power is now the single most decisive factor in determining where these multi-billion dollar facilities are built.
A New Global Map of Infrastructure
The race to build AI infrastructure is redrawing the global technology map. Currently, the United States is expected to capture the lion's share of the investment, projected to attract nearly half of all spending through 2050—a sum of $15.1 trillion. However, a significant portion of capital is also flowing to the Asia-Pacific region, with a forecast of $8.2 trillion, led by growth in China and India. Europe is also a major player, with a projected $5.6 trillion in investment. This geographic distribution is increasingly influenced by what is known as 'sovereign AI'—a push by governments to ensure their nations have domestic control over their own AI infrastructure and data. This trend is accelerating investment in Europe, the Middle East, and parts of Asia as countries seek to reduce their reliance on US-based hyperscalers. However, local opposition citing environmental and resource concerns has also emerged as a significant factor, delaying or blocking dozens of projects globally.
