The increasing demand for AI has caused data center power consumption to skyrocket, creating a bottleneck in AI development

In recent years, major technology companies have been investing heavily in the development of generative AI tools, and construction projects for AI data centers are underway around the world. However, the AI bubble has caused a surge in data center power consumption, and power supply issues are becoming a bottleneck that threatens the growth of AI, the Financial Times reports.

Booming AI demand threatens global electricity supply
https://www.ft.com/content/b7570359-f809-49ce-8cd5-9166d36a057b

In an interview on X (formerly Twitter) Space in April 2024, businessman Elon Musk expressed the view that while AI development in 2023 was limited by chips, the limiting factor for AI in the next one to two years will be 'power supply.' Amazon CEO Andy Jassy also said in January that 'large-scale language models consume a lot of power, and right now we don't have enough energy,' pointing out that power supply is a challenge in AI development.

According to research group Dgtl Infra , global investment in data centers is estimated to exceed $225 billion (about 34.72 trillion yen) in 2024. In addition, NVIDIA CEO Jensen Huang said that $1 trillion (about 154 trillion yen) worth of data center infrastructure needs to be built in the next few years to support generative AI, which consumes large amounts of power and processes huge amounts of data.

According to calculations by the International Energy Agency (IEA) , the electricity consumed by data centers worldwide is expected to increase to 1,000 TWh by 2026, equivalent to the total electricity consumption of Japan. Therefore, the IEA stated that 'updated regulations and technological improvements such as efficiency are essential to mitigate the rapid increase in energy consumption by data centers.'

The International Energy Agency predicts that the electricity consumed by data centers around the world will rival that of Japan by 2026, and that energy demand doubled due to AI and virtual currencies will be covered by 'clean' electricity such as nuclear power - GIGAZINE

Major technology companies such as Amazon, Microsoft, and Google are investing billions of dollars in building computing infrastructure such as data centers to improve the performance of generative AI. However, data centers cannot be built anywhere. In addition to vast tracts of land to house large numbers of servers, chips, and other physical components, they also require water to cool the computers and a stable power supply. Already, popular locations for building data centers, such as Northern Virginia in the United States, are facing capacity limits.

Dominion Energy, a Northern Virginia power company, temporarily halted new data center connections in 2022 while it analyzed how to handle a surge in electricity demand. In a regulatory filing in October, Dominion Energy said the increased demand for electricity was a challenge as it experienced 'significant increases in load from data center development.'

'There's always been a demand for data centers, but never before has there been this much,' said Pankaj Sharma, vice president of data centers at industrial equipment maker Schneider Electric , which is working with Nvidia to build data centers optimized for AI workloads. 'Right now, we don't have enough capacity to run all the data centers we'll need by 2030.'

Daniel Golding, a former executive at Google's data center division and CTO of manufacturing consulting firm Appleby Strategy Group , said, 'One of the things that will limit the deployment of the new AI economy is where to build data centers and how to get power. At some point, the reality of the power grid will get in the way of AI.' Of the 50 potential data center sites that Golding looked at, only about two could actually be developed, so finding a suitable location for building a data center is also a challenge.

These power demand concerns have led data center developers to turn to on-site generation and even nuclear energy. In 2023, it was reported that Microsoft was considering deploying small modular nuclear reactors to cool its data centers.

Microsoft criticized for sucking up water during droughts and using it to train AI, while Microsoft begins considering small nuclear reactors as a power source for data centers - GIGAZINE