The rapid increase in demand for computational power has led artificial ielligence to the ceer of energy debate in the United States. Data ceers that support cloud services, online broadcasting and storage platforms are currely consuming large quaities of electricity, but the emergence of artificial ielligence tools has greatly increased these needs. According to federal forecasts, the share of national electricity consumption by data ceers can increase from 2 % a year to 2 % by year 2.
Since the performance of an artificial ielligence writer or hosting a large language model is more energetic than usual web activities, the sloping power consumption curve has a sharp slope. This expansion is not only changing the relationship between technology companies and electricity installations, but also changing how to distribute electricity costs in society. The price of electricity in the United States has risen by more than 5 perce since year 2, and a study of Carnegie Melon and North Carolina warns that by year 2, another 5 perce will be added to prices across the coury.
In states such as Virginia, this increase can reach 5 %. Electricity companies argue that infrastructure upgrades are esseial, but the main concern is who will pay for it. This is just the beginning of the story, because when a Frenchman asks the ChatPTT about the next strike time, Americans pay more electricity. How is this possible? When every person everywhere in the world asks a daily chat, the extra energy consumed by that request is absorbed io the US power grid.
This is because the chat system is implemeed on the US -based servers hosted by US data ceers and fed by the coury’s electricity grid. If technology companies guaraee large capacity allocation and delay projects, small households and small businesses may have to pay for unused infrastructure. The case of Unicorn Ierest in Virginia, where delayed facilities made local customers pay millions of dollars in upgrades, shows the danger well.
To deal with such problems, the American Electric Power company in Ohio proposed a rating plan that calls on data ceers to pay 2 % of their demand capacity regardless of actual consumption. Despite opposition from cloud service providers who proposed at least 2 %, state regulators confirmed the move. Some companies are looking to bypass traditional electricity companies with their electricity generation. Amazon, Microsoft, Google and Meta are currely running renewable energy facilities, gas turbines and diesel backup generators, and some are planning to build nuclear facilities.
These companies not only generate electricity for their operations, but also sell surplus energy in wholesalers and compete with traditional suppliers. In rece years, such sales have generated billions of dollars in revenue and have infiltrated the large cloud services in some areas. Another challenge is the unstable use of artificial ielligence training that can volatile between peak and low. Even a 5 % change in demand can destabilize networks and force electricity companies to ierfere with fabricated work loads. Given that in some states, households pay more each moh, there is concern that consumers will eveually pay for the cost of keeping large language models and artificial ielligence writers.
(tagstotranslate) Artificial Ielligence
