If this is true, server farms could consume up to five percent of the continent’s electricity supply.
“In the last two decades, no technology has driven the need for energy infrastructure development in Europe as much as artificial intelligence” – was written the analysts in the McKinsey report just published. According to the consulting firm’s estimates, the electricity consumption of European data centers will reach 150 terawatt hours (TWh) by 2030 from today’s 62 TWh at the current rate of expansion. McKinsey estimates that European utilities will need to build about 25 gigawatts of generation capacity to run all the power-hungry GPUs that will be used to train and run generative AI models and services, and much of that will be “green.” must come from a breed.
McKinsey analysts note that this last detail is somewhat problematic. “The data center industry faces a major challenge to decarbonize its footprint and reach net zero goals by the 2030-40s. Some – including former Google CEO Eric Schmidt – argue that we shouldn’t worry about AI’s energy hunger , because technology itself is our best chance to deal with the increasing prevalence of greenhouse gases. “We’re not going to meet the climate goals anyway because we’re organized to do it,” Schmidt admitted at a recent AI summit in Washington.
Unless the climate promise of artificial intelligence is realized, McKinsey says renewables will remain the most popular solution for cloud and data center operators to offset emissions from their facilities. However, the consultancy believes that such systems have “minimal impact on the long-term emissions of energy systems and rarely encourage the development of new projects or the production of clean energy”.
One way to reduce the impact of AI is to place data centers near places that generate clean energy. This approach may work best if these facilities have the hardware used to train the models—a process that requires enormous amounts of computing power, but does not need to be physically close to population centers. However, as the composition of the AI workload shifts towards inference – actually running the models – this will change, and latency requirements will dictate that facilities be built near cities.
In the long term, McKinsey looked at many possible solutions, from carbon capture to on-site energy production using small modular reactors. Such reactors have received a lot of attention from cloud providers in recent months, despite the fact that none have yet been deployed for commercial use. But nuclear energy is not the only option for on-site energy production, it is just one of the cleaner and more energy-saving options. Diesel generators, fuel cells, and utility-scale batteries typically provide supplemental power and smooth out fluctuations in grid power.
Although data center energy consumption is expected to increase, McKinsey is another recent according to its meaning overall European energy demand may fall short of previous forecasts. The company believes that even 40 percent of the predicted 460 TWh increase may not be needed. According to McKinsey, the growing population and gross domestic profit are expected to push energy consumption up by seven percent across the continent by 2030. Despite these trends, analysts observe that demand has actually slowed — largely due to improvements in energy efficiency, the shift to a service-oriented economy, milder winters and high energy costs that have led to a trend toward deindustrialization.
Source: sg.hu
