The growing demand for artificial intelligence (AI), including generative AI (GenAI), is putting immense pressure on electricity resources, according to a report by Gartner Inc. The technology research firm forecasts a 160% increase in data center energy consumption over the next two years, with 40% of existing AI data centers potentially constrained by power shortages by 2027.
“The explosive growth of hyperscale data centers to support GenAI is driving an insatiable need for electricity, outpacing the ability of utility providers to expand their capacity,” Bob Johnson, VP Analyst at Gartner, said in a media release. “This will disrupt energy availability and hinder the development of new data centers.”
By 2027, AI-optimized servers in data centers will require 500 terawatt-hours (TWh) of electricity annually, more than double the 2023 levels. Johnson explained that the large-scale facilities being built to process the massive data needed for GenAI applications face long-term challenges.
“New power generation and distribution capabilities could take years to become operational,” he said.
Gartner also warned that power shortages will drive up electricity prices, making it more expensive to run AI systems.
“Operators will pass increased energy costs to AI product and service providers,” Johnson noted. He advised organizations to negotiate long-term power contracts and consider alternative, less energy-intensive solutions.
Impact on sustainability goals
Efforts to meet zero-carbon goals are likely to be negatively affected. Some energy suppliers are resorting to keeping older fossil fuel plants operational to meet surging power demands, which could increase carbon dioxide emissions.
“The reality is that higher data center use will lead to more CO2 emissions in the short term, challenging sustainability targets,” Johnson said.
Gartner suggests businesses re-evaluate their sustainability objectives while preparing for increased energy costs. Companies developing GenAI applications are encouraged to use efficient computing power, explore edge computing, and consider smaller models that consume less energy.
Long-term solutions could help meet power needs sustainably. However, these technologies will take years to mature.