Currently, data centers consume 4.4% of U.S. energy, and this number is likely to triple to 12% by 2028.
That’s according to data from Berkeley Lab, which analyzes power usage in data centers, including servers, storage, networking equipment and other infrastructure, on behalf of the U.S. government.
this Report As artificial intelligence booms driving power demand Data centerand Gartner Forecast 40% of operators will face power restrictions. This has led to big tech companies looking into alternative energy sources, including nuclear.
The report shows that data center energy usage remained stable at around 60TWh from 2014 to 2016, with almost no growth since 2010. GPU Power on AI Energy consumption began to increase, increasing to 76TWh in 2018 and 176TWh in 2023, accounting for 4.4% of the total energy consumption in the United States.
Given this growth, the report says data center power demand could increase to 325TWh to 580TWh in the next four years alone, accounting for 6.7% to 12% of U.S. electricity consumption.
However, this still lags behind other countries. The data center has been depleted 18% of Ireland’s electricity.
The scope is so broad, the report says, because it captures different scenarios based on future device shipments, changes in cooling, and more.
“Device changes are based on assumptions about the number of GPUs shipped annually, which is dependent on future GPU demand and the manufacturer’s ability to meet those demands,” the report explains.
“Average operating practices for GPU-accelerated servers represent the computing power required to meet the demands of AI workloads and the frequency of usage of installed AI hardware,” the report adds.
“Changes in cooling energy use depend on the type of cooling system selection and the efficiency of those cooling systems, such as moving toward liquid-based cooling or away from evaporative cooling.”
Cloud history
The report points out that this is not the first time innovation in the technology industry has led to an increase in data center usage, but not necessarily an increase in energy use.
Between 2000 and 2005, data center electricity consumption approximately doubled. But starting in the early to mid-2010s, the shift from on-premises infrastructure to the cloud did not result in a significant increase in energy use. This is primarily because colocation and cloud facilities tend to be more efficient than those operating at corporate headquarters.
“Industry efficiency strategies to avoid increased energy demand during this period include improving cooling and power management, increasing server utilization, improving computing efficiency, and reducing server idle power consumption,” the report states.
While many claim that artificial intelligence will be able to solve challenges related to the climate crisis, that may not be possible this time. This is because the prosperity of artificial intelligence requires new types of hardware.
“Most notably, the rapid growth of accelerated servers has caused total current data center energy demand to more than double between 2017 and 2023, and the continued growth in the use of accelerated servers for artificial intelligence services may lead to a decline in the use of accelerated servers by this decade. There was further substantial growth at the end of the year,” the report stated.
But the industry should strive to do better at improving data center efficiency, and transparency in data center energy use and benchmarking will help do that.
“The surge in energy demand currently and likely in the near future highlights the need for future research to understand the early, rapidly changing field of AI in the data center industry and identify new efficiency strategies to minimize this ongoing Growth and increasingly important resource impacts.