With the rapid growth of data centers to support the increasing adoption of artificial intelligence (AI) models, the demand for electricity to power GPU-filled servers is soaring.
According to a study by Epoch AI, the compute capacity required for large language models (LLMs) has been growing exponentially since 2010, driven by major releases from OpenAI, Meta, and Google DeepMind.
Epoch AI
Major AI service providers like Amazon Web Services, Microsoft, and Google are turning to nuclear power plants to meet the escalating electricity needs of their data centers. The White House has also announced plans to support the development of new nuclear power plants to increase carbon-free electricity sources.
The World Economic Forum (WEF) reports that the computational power required for AI is doubling every 100 days, emphasizing the need to balance AI’s growth with sustainability goals.
According to the WEF, the energy demand for AI tasks is increasing rapidly, with projections indicating that by 2028, AI could consume more power than Iceland did in 2021.
Jack Gold, a principal analyst at J. Gold Associates, highlights the significant power consumption of AI models, particularly during training processes, which require substantial computational power.
While AI models like LLMs are not continuously training, the data centers housing them must maintain peak power availability. Gold emphasizes the energy-intensive nature of AI deployments, especially with the increasing use of GPUs.
Tech companies are exploring various power sources, including nuclear energy, to meet the surging power demands of AI applications, rather than solely relying on green energy solutions.
For more information, visit www.computerworld.com