Data Centers: Electricity Consumption to Quadruple by 2035

What happened?

A TechCrunch report, based on data from the International Energy Agency (IEA), projects that global electricity consumption by data centers will quadruple by 2035. According to the report, new facilities built by 2033 could require as much energy as India currently consumes, a country with over 1.4 billion people and an annual consumption of approximately 1,300 TWh. This unprecedented growth is driven by generative artificial intelligence, training models like GPT-4 or Gemini, and the expansion of cloud services. In 2023, data centers consumed around 460 TWh, equivalent to 1-2% of global electricity. By 2035, they could reach between 1,200 and 2,000 TWh, representing up to 8% of the total, according to the IEA.

Why is it important?

This exponential increase is historically unprecedented. During the 2010s, data center consumption grew at a moderate pace thanks to efficiency improvements, but generative AI has broken that trend. For example, a query on ChatGPT consumes about 10 times more energy than a Google search. Training a model like GPT-3 required about 1,300 MWh, equivalent to the annual consumption of 130 U.S. households. Additionally, the rise of cloud computing and edge computing adds pressure. While chip energy efficiency has improved by 60% in the last decade, processing volume grows at a rate of 20-30% annually, outpacing any gains. By 2035, data centers are expected to consume more electricity than the entire global steel or aluminum industry.

Consequences for businesses and users

For tech companies, operating costs will skyrocket. Electricity already accounts for an estimated 30-40% of a data center's total cost. Giants like Google, Microsoft, and Amazon have reported significant increases in their energy bills: Microsoft, for example, spent $3.7 billion on electricity in 2023, 25% more than the previous year. This could translate into higher prices for cloud services and AI subscriptions. End users could see increases of up to 10-15% on SaaS platforms. Additionally, regions with weak power grids, such as parts of Africa or Latin America, could suffer outages or restrictions. In Virginia, USA, which handles 70% of global internet traffic, data center demand has already caused delays in grid connection. The environmental impact is also critical: if electricity comes from fossil fuels, carbon emissions could increase by 200-300 million tons annually by 2035, equivalent to the emissions of Germany's entire vehicle fleet.

What should readers know?

Despite the challenges, there are positive developments. Companies like Google have made their data centers 1.5 times more efficient than the industry average through liquid cooling and custom chips (TPUs). Microsoft has invested in modular nuclear power for its data centers, while Amazon plans to reach 100% renewable energy by 2025. However, the speed of growth could outpace these improvements. The IEA warns that without strict regulations, consumption could double every four years. Governments like Singapore's have imposed moratoriums on new data centers, and the European Union is developing efficiency standards. For investors, renewable energy companies (NextEra, Enphase) and efficiency technology (Nvidia with its GPUs, Vertiv in cooling) could benefit. In contrast to the dot-com bubble, where energy consumption grew 10% annually, the current AI cycle could be more intensive: it is estimated that by 2030, AI will consume 10-15% of global electricity. The key will be innovation: from underwater data centers (Microsoft) to fusion energy (Helion). The digital future depends on how we manage this demand; the transition to clean sources and efficiency are more urgent than ever.