"By 2027, 40% of the world's artificial intelligence (AI) data centers will face difficulties due to power shortages. Expansion of self-supplied power sources such as small modular reactors (SMRs) is necessary."
Bob Johnson, vice president and analyst at U.S. IT research advisory firm Gartner, said in a recent written interview with ChosunBiz, "With increased use of AI, data center power consumption will surge over the next two years," and made the remark. Johnson has been an expert researching data centers and the power sector for more than 40 years.
According to Gartner, the power needed in 2027 for global data centers to operate AI-optimized servers is expected to reach 500 terawatt-hours (TWh) per year. That is 2.6 times the power used by data centers in 2023 and, according to the U.S. Energy Information Administration (EIA), is similar in scale to Korea's annual power consumption (586 TWh·2022). Korea ranks sixth in the world in power consumption.
◇ big tech moving quickly to prepare for power shortages
Johnson said, "Large language models (LLMs) that support generative AI applications are scaling rapidly, but it can take years to build new infrastructure such as transmission and distribution and secure generation capacity, so the power shortage problem will persist." He added, "Soon, the establishment of new data centers and the growth of generative AI will be determined by the availability of power to operate them," and said, "Corporations need to identify the potential risks that power shortages could pose to their products and services."
Johnson introduced that global big tech firms are actively pursuing power procurement strategies. He explained, "Amazon Web Services (AWS) has signed a contract with Talen Energy to receive up to 1 gigawatt (GW) of power directly from a nuclear power plant operating in Pennsylvania," and added, "Microsoft plans to restart modules of the Three Mile Island nuclear plant that were shut down several years ago, and Google recently signed power supply agreements with two hydroelectric dams located in Pennsylvania."
◇ rising costs due to power shortages… fossil fuel use expected to increase
Johnson predicted that if the power shortage problem becomes imminent, power prices will inevitably rise and the operating costs for LLMs will also increase. He analyzed, "As competition for power among corporations intensifies, electricity rates needed to run data centers will rise sharply," and said, "These expenses will be passed on to suppliers of generative AI products and services."
Johnson predicted that to cope with surging power demand, corporations will use every means available, which could lead to keeping fossil fuel power plants that were scheduled to close in operation. For that reason, he argued that an increase in carbon dioxide emissions is unavoidable. Johnson said, "If data center power use increases, carbon dioxide emissions will also rise in the short term," and added, "It will become more difficult for data center operators and customers to achieve strict sustainability targets for carbon dioxide emissions."
◇ "Need to expand intelligent grids and self-supplied power sources… SMR data centers will appear by 2030"
Johnson said technologies like HVDC (high-voltage direct current transmission) will not fundamentally solve the power shortage and emphasized the need to enhance intelligent power grids or expand self-supplied power sources.
He advised, "Governments should identify inefficiencies in the distribution networks currently in operation and, where possible, integrate 'smart grid (an intelligent power grid that integrates ICT to monitor and control the entire process of power generation, transmission, distribution and consumption in real time)' technologies to maximize efficiency." He added, "Expansion of data centers in a microgrid form—with self-supplied power sources such as SMRs that can independently generate, store and consume power within a limited area—is needed."
Johnson said increasing the supply of green renewable energy alone will not be sufficient to meet surging power demand. He said, "Wind and solar power have the problem of not being able to supply power continuously," and added, "SMRs, which can supply power 24 hours a day, year-round, without being affected by weather and sunlight changes, are emerging as an alternative." According to the industry, capacity factors for wind power facilities are only 35–40%, and for solar power about 25%.
Johnson said, "Many corporations are developing small modular reactors (SMRs) that do not carry the meltdown (the phenomenon where nuclear fuel rods melt from excessive heat) concern," and said, "By around 2030, the first SMR-based data centers will emerge." He added, "I understand Microsoft is also considering introducing new SMR technology into its future data centers."