New reactors: Google relies on nuclear power for AI
Google wants to accelerate the development and construction of SM reactors. The data company needs gigawatts for AI.
(Image: Pavel Ignatov/Shutterstock.com)
"The grid needs new sources of electricity to support AI," Michael Terrell, Google's Senior Manager for Energy and Climate, strains the chain of causality. Data centers for artificial intelligence need an astonishing amount of electricity. After Microsoft, Google is now also relying on nuclear power.
But while Microsoft wants to revive a decommissioned nuclear power plant (built in 1969-1978), Google is planning new reactors. Specifically, the so-called Small Modular Reactors (SMR) with salt cooling, which are currently not ready for the market. To accelerate their innovation, Google has signed a contract with Kairos Power. In this agreement, Google undertakes to buy the entire electricity production of several Kairos SMRs for an undisclosed period of time, up to 500 megawatts of capacity.
Videos by heise
This prospect of guaranteed income should help Kairos to accelerate SMR development so that it can supply Google with electricity as early as 2030. Kairos' first reactor will then have an output of 50 megawatts. By 2035, six more reactors of 75 megawatts each are to be added at other US locations, which would give Google a total of 500 MW. In the summer, Kairos began construction of a demonstration reactor in Tennessee, which will only produce heat and not yet electricity.
More locations for nuclear power plants
"The smaller size and modular design can reduce construction times and enable construction at more sites," expects Googler Terrell, which should make the completion date more predictable. He does not reveal how much Google is paying. It won't be cheap: "Nuclear power has the highest economic impact of any power source, according to the US Department of Energy," he admits. But Google has a goal: AI should develop its "full potential for everyone", and that requires electricity.
So much electricity that equipment such as transformers, switches and generators for new data centers have become scarce. The real estate specialist CBRE currently expects electricity connection lead times of up to four years. OpenAI would like to distribute several 5-gigawatt data centers across the USA. By comparison, the Berlin distribution grid has a total capacity of 2.1 gigawatts.
(ds)
