Artificial intelligence (AI) requires substantial energy, and Amazon believes that nuclear could be the answer. Amazon Web Services (AWS) is reportedly investing up to US$650M in a data centre campus located next to a nuclear power plant in Pennsylvania, with plans to construct several data centres on the site, as reported by The Information.
AWS is a prominent cloud services provider, supporting millions of customers worldwide, including startups, major corporations, and public sector organizations. It reported sales of US$24B in the fourth quarter of the fiscal year 2023. Along with Microsoft Azure and Google Cloud, the three major cloud providers account for about two-thirds of the overall cloud computing market, as per Statista. Cloud services enable companies to reduce their physical tech infrastructure, access the latest technological advancements, and save costs.
Previously, the site had garnered attention from other cloud providers who were in search of sufficient power to support AI computing. Although AWS is not the first to consider nuclear power for AI, it is the first to have a campus with direct access to nuclear power. This proximity had formerly caused some cloud providers to hesitate due to safety concerns.
Amazon has positioned this decision as part of a larger initiative to invest in carbon-free energy. “In addition to our wind and solar energy projects, which rely on weather conditions for energy generation, we are also investigating new innovations and technologies and investing in other forms of clean, carbon-free energy. Our agreement with Talen Energy for carbon-free energy is one such project.” stated an Amazon spokesperson.
Amazon has made an initial payment of US$350M to Talen Energy Corporation for the site, with the potential to pay up to an additional US$300M, depending on certain milestones. Talen has indicated that Amazon will develop a data centre campus with a capacity of up to 960MW.
This deal highlights the energy consumption dilemma posed by AI. Since around 2010, data centres have been estimated to consume about 1% of global electricity production, according to the Wall Street Journal.
However, the power requirements of AI, especially Nvidia’s graphics processing units, which need significantly more than regular chips, could considerably increase the proportion of power used. A researcher from Vrije Universiteit Amsterdam’s School of Business and Economics has projected that due to AI, the energy required to power global data centres could increase by 50% by 2027.
