AI data centers have sparked a new fever: the so-called “bring your own power.” The demand and consumption The pressure these plants impose is so enormous that they do not want to depend on external sources. The solution is theoretically simple, and we are already seeing how when a new data center is built, it is normal for some type of power plant to be built next to it.
We are seeing it now. The data centers that OpenAI and Oracle are building in West Texas are accompanied by the creation of a natural gas-based power plant. Both xAI’s Colossus 1 and Colossus 2 in Memphis take advantage of gas turbines. And as they also indicate in The Wall Street Journalmore than a dozen Equinix data centers across the US are powered by stand-alone fuel cells. If the conventional electrical grid cannot be used, nothing happens: you create a power plant and that’s it.
The US has an electrical problem. The technology giants would prefer to connect to the conventional grid, but bottlenecks in the supply chain, bureaucracy – permits, licenses – and the slowness in building the necessary transmission infrastructure prevent this. According to the ICV firmThe United States would need to add about 80 GW of new generation capacity per year to keep pace with AI, but right now less than 65 GW per year are being built. There is another direct consequence of this problem: the rise in the electricity bill.
Data centers that look like cities. The needs and ambition of AI companies has made data centers become calculation and resource consumption monsters. One can only consume as much electricity as 10,000 stores in the Walmart electronics chain, WSJ estimates. Before 2020, data centers represented less than 2% of US energy consumption. By 2028 they are expected to represent up to 12%. A 1.5 GW data center, for example, would have consumption similar to that of the city of San Francisco, with about 800,000 inhabitants.
China has a lot of advantage over the US in this. While the US deal with that lack of powerChina does not stop investing in new energy generation. According to data According to the National Energy Administration, the Asian country added 429 GW of new energy generation in 2024, while the US only added 50 GW. It is true that China has four times the population, but its centralized planning is helping to avoid problems that affect the US electrical grid.
The white knight to the rescue. Faced with this shortage, natural gas has become the preferred resource for on-site energy generation. Although large turbines have long delivery times, smaller turbines or fuel cells that use natural gas are being used because of their rapid availability and installation.
Renewables lose steam. Meanwhile, things are not promising for renewable energies (solar and wind, especially). There are about 214 GW of new generation theoretically in projectbut spending on such technologies could decline due to the potential loss of tax credits: the Trump administration criticizes that those clean energies do not provide a constant flow necessary for AI.
The nuclear alternative. Faced with this apparent decline of nuclear energy, there is a growing interest in compact nuclear reactors (SMR), which allow us to provide the advantages of this type of center and a flexibility that can be very interesting for AI data centers. amazon, Google, Goal either Microsoft They are betting part of their future on nuclear powerbut that It doesn’t mean there aren’t challenges to overcome..
Image | Wolfgang Weiser
In Xataka | World record in nuclear fusion: the German Wendelstein 7-X reactor has broken all records
GIPHY App Key not set. Please check settings