AI training and operating data centers may soon have millions of chips, spend several hundred billion dollars, and if the current trends are captured, the equivalent of a large city electricity grid requires electricity.
Accordingly A new study From the researchers of Georgettown, Epot AI and Rand, from 2019 to this year, AI data centers around the world have seen a way to grow. Co-authors compiled and analyzed a data set of more than 500 AI data centers projects and discovered that when the data centers’ calculated performance is more than twice the annual, there is also electricity requirements and capital expenditures.
Searchs portray the challenge of creating the necessary infrastructure to support the development of AI technologies over the next decade.
Openi, which was recently said About 10% of the world’s population Its ChatzP is using the platform, the US (and probably elsewhere) has partnerships to raise $ 500 billion with softbank and others to set up a network of AI data centers. Other technology giants, including Microsoft, GoogleAnd AWS, collectively promised to spend a few million dollars this year that extends their data center footprints.
According to Georgettown, Epoche and Rand Study, hardware costs for AI data centers like Jai’s Calls, which cost about $ Billion, $ 1.9x per year between 2019 and 2025, when the need for power in the same period has risen to 2X. (Calsus draws approximately 300 MW of energy, about 250,000 families.)
The survey also found that data centers have become more energy efficient in the last five years, a basic metric – wat -per -calculated performance – from 2019 to 2025 per year, 1.34x. Nevertheless, these improvements will not be enough for growing electricity. By June 2030, the top AI Data Center can have 2 million AI chips, can spend 200 billion dollars and need 9 gigawatt strength – about 9 nuclear reactors output.
This is not a new manifestation that the AI Data Center has electricity claims At the speed of straining the power gridThe The Data Center Energy Energy is predicted to increase by 20% by 2030, According to the recent Wales Fergo analysisThe It may push the sources of renewable energy, which depends on the changing weather, in their limits – Sparring up a ramp Renewable, like fossil fuels, at environmentally harmful electricity sources.
AI Data Centers also give other pose Environmental threatSuch as the use of high water, and taking valuable real estate, as well as state tax bases also erodes. A study by Good Jobs First, a Washington, DC-based non-profit, Assumption That that loses more than $ 100 million per year in tax revenue in at least 10 state data centers, which results in excessive liberal enthusiasm.
It is possible that these assumptions may not be effective, or the scalters of the time are off-kilters. Some hypersscale, like AWS and MicrosoftData centers have returned to projects in the last several weeks. In a note of investors in the middle of April, Cow’s analysts It is noticed that the data center has been a “cool” in the market in the early 2021, which indicates the fear of expanding the instability of the industry.