Between $3.7 and $5.2 trillion needed to finance AI infrastructure by 2030
The required investments to build data centers capable of supporting artificial intelligence are reaching unprecedented levels. According to McKinsey projections, between $3.7 trillion and $5.2 trillion will need to be deployed by 2030, depending on how demand evolves. These amounts reflect a profound transformation of the digital economy, driven by American and Chinese hyperscalers, as well as by energy constraints that are redefining the conditions for setting up these infrastructures.
Financing Needs That Surpass Previous Tech Bubbles
Spending Concentrated in the Hands of a Few Tech Giants
American hyperscalers are leading the bulk of current investments. Microsoft, Google, Amazon, and Meta have announced a combined budget of $364 billion for 2025, up from $224 billion in 2024. Between 2015 and 2025, the capital expenditures of these four companies have skyrocketed from $23.8 billion to $315 billion, a thirteen-fold increase. Amazon Web Services shows the most significant growth, with its annual budget rising from $12 billion in 2015 to an anticipated $100 billion by 2025.
Currently, these companies control over 50% of the global data center capacity ready for AI, according to McKinsey estimates. By 2030, they are projected to account for 60 to 65% of AI workload hosting in Europe and the United States. The year 2024 set a record with $57 billion in global data center investments, with hyperscalers contributing to 80% of this. In North America, colocation providers' capacity increased by more than 40% in one year, from 12.4 gigawatts in 2023 to 18 gigawatts in 2024, with an additional 30 gigawatts planned.
The Stargate project, announced in January 2025 by Donald Trump with the backing of OpenAI, Oracle, and Japan's Softbank, exemplifies this trend. With a $500 billion fund over five years, it plans the construction of fifty campuses in the US, the first of which in Abilene, Texas, includes ten buildings of 50,000 square meters each. In August 2025, OpenAI and Oracle expanded this partnership with five new sites representing over $400 billion in investment over three years, bringing the total planned capacity to nearly 7 gigawatts. Beyond the US, China announced $138 billion in investments, while France is mobilizing €109 billion by 2030. Notably, Canadian firm Brookfield committed to investing €20 billion in AI infrastructure, and a partnership with the United Arab Emirates is planning between €30 and €50 billion for a one-gigawatt campus.
Access to electricity becomes the main bottleneck
Electricity availability has become the major limiting factor for the development of data centers. The International Energy Agency forecasts that the power consumption of these facilities will more than double by 2030, increasing from 415 terawatt-hours in 2024 to 945 terawatt-hours, which is equivalent to Japan's current consumption. This growth will account for nearly 3% of global electricity demand, compared to 1.5% today. Goldman Sachs estimates that data center power demand will increase by 165% between 2023 and 2030, necessitating $720 billion in investments in power grids.
In the United States, fifteen states concentrate 80% of the demand. In Virginia, where data centers already account for 26% of total electricity consumption, Dominion Energy estimates that needs will quadruple by 2035, equivalent to the consumption of an additional 8.8 million households. In Maricopa County, Arizona, electricity demand is set to increase by 40% by 2030. Operators are responding by negotiating several hundred megawatt contracts directly with energy suppliers, often years before construction starts. Amazon Web Services has signed three agreements with Energy Northwest, X-Energy, and Dominion Virginia to deploy over 600 megawatts of nuclear capacity in Washington and Virginia.
AI-dedicated data centers have an energy density four to five times higher than traditional centers. A 100-megawatt data center consumes as much electricity as 100,000 households annually, but the largest projects under construction will consume twenty times more, equivalent to two million homes. This is causing local tensions. In several mature markets like Northern Virginia, Dallas, and California, vacancy rates have fallen below 1%, with moratoriums imposed by grid operators. McKinsey estimates that, in the US alone, the supply deficit could exceed 15 gigawatts by 2030, even if all currently known projects are delivered on time. Building permit acquisition times have lengthened, from the previous six to twelve months to two or three years today in markets like Northern Virginia, Dublin, and Singapore, due to local opposition and stricter environmental regulations.
This content has been automatically translated using artificial intelligence. While we strive for accuracy, some nuances may differ from the original French version.
