Economy

Data centers and AI: why electricity grids are at their limit

According to the IEA’s World Energy Outlook, data center consumption will double by 2030. Between shortages of transformers and saturated networks, here’s what’s happening

The expansion of data centers And of artificial intelligence is putting a strain on the world’s energy infrastructure. This is what emerged from the latest report of theInternational Energy Agencythe World Energy Outlook.

The Agency’s forecasts

The numbers contained in the Agency’s report leave little room for interpretation. Data center electricity consumption is expected to double by 2030, pushing national networks to the limit of capacity.

Let’s talk about 1,000 TWh per year by 2030, a figure equivalent to approximately 3% of global electricity demand. Suffice it to say that, according to the agency’s satellite measurements, just 93 large “AI factories” currently under construction will require around 50 GW of power.

For comparison, this is an energy requirement equivalent to the entire peak demand of Great Britain.

Of the 120 GW of new capacity of data centers expected by the end of the decade, more than three-quarters will be intended to power AI-optimized servers.

Because the networks are under pressure

The point, however, is not the generation of electricity itself, but its physical delivery. In several major markets, the power of data centers already announced it would exceed the peak demand of the entire local electricity system.

The problem is first and foremost one of timing, because while a data center is being built two or three yearsa high-voltage line or a new substation requires roughly twice as much, and grid expansion plans can require 10-15 years.

To this we add one global shortage of critical components. Power transformers, once delivered in 6-12 months, today require 24-48 monthswith tips up to five years for higher class models.

Order books for transformers, cables and switchgear are saturated, copper is in short supply and production capacity has not been sized for a simultaneous jump in demand. The result, therefore, is that the data center is ready, but is waiting for connection.

The impacts on the energy mix

Since networks cannot keep up, additional demand is increasingly covered with “behind the meter” solutions, with consequences on the emission profile.

The International Energy Agency predicts that the Data center-related carbon dioxide emissions grow by 70% by 2030despite the green energy contracts of the large technology groups, because part of the growth is satisfied by energy plants natural gas and coal.

In the United States, natural gas is expected to provide about 60% of data center electricity by the end of the decade. To bypass connection queues, Microsoft, Meta, Amazon, Google and others have signed agreements for natural gas turbines on a large scale, often installed directly on sites or in behind-the-meter configurations.

In parallel, the agreements for the revival of nuclear energy and for i small modular reactors (SMRs). Amazon is targeting 12 Xe-100 reactors in Indiana for 960 MW, Google has contracted 500 MW of SMR with Kairos by 2035, and Meta has signed deals for up to 1.2 GW of nuclear capacity.

However, they are solutions that look very far ahead in time, given that these reactors will not arrive before the next decade. In the meantime, finished data centers remain in the dark, awaiting connection to the electricity grid.