Rene Haas, CEO of ARM, stated that by 2030, artificial intelligence could consume 25% of the electricity in the United States. Currently, AI data processing centers consume 4% of the electricity. He believes that without a significant increase in efficiency and/or significant strengthening of government regulation, the current trend of AI development will not be sustainable.
According to the January report of the International Energy Agency, which states that ChatGPT consumes approximately 2.9 kWh of electricity per request, which is 10 times more than standard Google search. Therefore, if Google fully switched its search to AI, the company would consume at least 11 TWh of electricity per year, instead of the current 1 TWh.
The statement was made on the eve of the expected partnership agreement between the US and Japan in the field of artificial intelligence, as well as against the backdrop of recent developments, such as Sora by OpenAI, the current version of which, according to Factorial Funds estimates, requires at least one NVIDIA H100 graphics processor per hour to generate five minutes of video. One 700 W NVIDIA H100 consumes approximately 3740 kWh per year.
The US Energy Information Administration (EIA) stated that in 2022, the United States produced 4240 TWh, of which only 22% came from renewable sources. Consumption amounted to 3900 TWh.
Sustainable development should also take into account the likely increase in demand from other industries and the scales of consumption of renewable and non-renewable resources. Considering that the cost of electricity has nearly doubled since 1990, greater regulation seems justified.
Rene Haas expressed hope that the international partnership between Japan and the US will help reduce these extremely high power consumption estimates. However, corporate greed and demand for computing power are also international phenomena.
Source: Tom's Hardware
Comments (0)
There are no comments for now