‘AI is useless without electricity.’ Are there any measures to supply power?

Source: Shutterstock

Artificial intelligence (AI) is changing our daily lives. Beyond processing complex calculations in an instant or quickly finding necessary data, the area of ​​use has recently expanded to areas that require creativity, such as video production, composition, and writing. As technology is still developing rapidly, what AI can do will increase in the future. But at this point, you need to think about it. If AI continues to develop like this, will only good things happen?

.

AI that eats electricity… How much power will we need in the future?

Among the many problems that are emerging due to the development of AI technology, the one that has recently become the most troublesome is ‘electricity shortage’. In order for AI to do more things, it must collect and continuously learn from the enormous amount of data being generated in real time from all over the Earth. This requires a data center that can retrieve and store that data. However, compared to the rapid increase in data centers over the past few years, the number of facilities producing the electricity needed for data centers is not keeping up with the pace.

While each Google search uses 0.3Wh of power, Chat GPT uses 2.9Wh of power, nearly 10 times more, per search (Source: Yonhap News)

The bigger problem is that the full-scale AI era has not yet begun. In the future, as AI plays a larger role in our daily lives, the power required for data centers will also increase exponentially. Let’s take the search function as an example. According to the Electric Power Research Institute (EPRI), the power used per Google search is only about 0.3 Wh, but Chat GPT uses 2.9 Wh per search. Therefore, EPRI predicted that if AI search functions such as Chat GPT are integrated into Google search in the future, the amount of power required for one search will increase to 6.9 to 8.9 Wh. This means that power consumption can increase by up to 30 times just by adding an AI search function to an existing search portal. (EPRI, ‘Powering Intelligence’, 2024)

So how much more will the amount of power needed for AI technology increase in the future? According to the International Energy Agency (IEA), the annual power consumption of data centers around the world is expected to increase from 460 TWh in 2022 to 620-1,050 TWh in 2026. Considering that Korea’s total electricity consumption was 568TWh as of 2022, more than 80% of that electricity was already consumed two years ago to operate the data center. In addition, it is expected that at least 1.3 times and up to 2.2 times more electricity will be needed in the next two years. (IEA, ‘Electricity 2024 – Analysis and forecast to 2026’)

Microsoft data center located in Middenmeer, Netherlands (Source: Shutterstock)

To secure a power production facility of this size, a huge amount of money, time, and effort must be invested. Goldman Sachs, a global investment bank, estimates that electricity demand in the United States will increase by about 2.4% between 2022 and 2030, of which 40% is for data centers. Goldman Sachs estimated that the amount that needs to be invested for this 2.4% is about $50 billion (about 69 trillion won), and a simple calculation shows that it will cost about $20 billion (about 27.5 trillion won) to secure power to supply data centers in the United States alone. (Goldman Sachs, ‘Generational growth – AI, data centers and the coming US power demand surge’, 2024)

.

AI technology development vs. overcoming climate crisis, Big Tech at a crossroads

There is one more thing to think about here. Currently, countries around the world are changing the electricity production method that relied on existing fossil fuels to an environmentally friendly method in order to overcome the climate crisis. However, the cost of producing the same amount of electricity is still much higher compared to fossil fuels for all new and renewable energies such as solar power, wind power, and green hydrogen. The pace of increase in power production is already not keeping up with the power demand demanded by AI, and there is an added constraint that power must be produced in a less efficient manner.

Google’s first in-house data center opened in Hanau near Frankfurt, Germany last year (Source: Yonhap News)

Accordingly, recently, there have even been cases of people siding with the side of AI technology development among the two conflicting values, ‘AI technology development’ and ‘overcoming the climate crisis.’ Google, which gave up the title of ‘the first company to achieve carbon neutrality’ last July, is a representative example. In its annual environmental report released in the same month, Google revealed that it emitted 14.3 million tons of greenhouse gases last year alone, which is a 13% increase from the previous year. Compared to 2019, four years ago, the increase rate is close to 50%. Google cited increased data center energy consumption as one of the reasons. Microsoft also announced in this year’s sustainability report that carbon emissions have increased by 29.1% since 2020, pointing to semiconductors, fuel, and building materials used in data center construction as the cause.

In order to quickly secure insufficient power, some countries are reconsidering their policies of shutting down or reducing dependence on nuclear power plants. French President Emmanuel Macron declared at the World Economic Forum (WEF) earlier this year, “We will expand nuclear power generation to respond to the AI ​​era.” France is already building six nuclear power plants and is considering building eight more nuclear power plants.

The era of ‘immense power’ AI… Attention again to ‘small nuclear power plants’ (Source: SBS News official YouTube channel)

.

Demand for data center electricity is also rapidly increasing in Korea… The problem of concentration in the metropolitan area is also serious.

In Korea, data center electricity demand is also rapidly increasing. According to the Ministry of Trade, Industry and Energy, the number of domestic data centers is expected to increase from 147 as of September 2022 to 784 in 2029. Accordingly, data center power demand during the same period is also expected to increase from 1.8GW to 41.5GW in 2029.

The bigger problem is that 550 of the 637 new data centers, or about 86%, are planned to be built in the metropolitan area. The reason this is a problem is because power generation sources, which are sources of power, are spread all over the country. It is difficult to quickly increase power production in response to rapidly increasing power demand, and the power system must be expanded at the same pace in order to bring the produced power to the metropolitan area. It is difficult to do just one thing well, but there is one more task to solve.

*Power system: A power grid that connects power plants, substations, transmission and distribution lines, etc. to supply power from power generation suppliers to power consumers.

It is also necessary to consider that the proportion of renewable energy connected to the power system is increasing. In the case of renewable energy that uses natural sources such as sunlight or wind to generate electricity, operation time varies depending on the weather, and the operation rate is lower than that of fossil fuel-based power sources. In addition, when considering location conditions such as sunlight, wind volume, and environmental impact, power generation facilities are often built in areas relatively far from the metropolitan area. Building a power grid that takes into account the characteristics of renewable energy is another task.

.

Increased power demand: Is the solution to building a flexible power grid?

Taewon Choi, Chairman of SK Group and Chairman of the Korea Chamber of Commerce and Industry, is giving a welcoming speech at the ‘2024 World Climate Industry Expo (WCE)’ held last month. (Source: Korea Chamber of Commerce and Industry)

Accordingly, the need to build a new messenger network for smooth power supply and demand in the AI ​​era has recently emerged. For example, Taewon Choi, Chairman of SK Group and Chairman of the Korea Chamber of Commerce and Industry, said in relation to this at the ‘2024 World Climate Industry Expo (WCE)’ held last month, “We need to build a flexible power grid that takes into account the expansion of distributed power sources.” He also emphasized that “the power grid-related legislation must be reformed from a regulation-centered system to an incentive system so that many companies can participate.”

Various solutions are being proposed as a methodology for building a flexible power grid, but the most discussed solution is ‘regional distribution’. It can be broadly viewed from the demand side and the supply side. Regional dispersion on the demand side means dispersing the demand concentrated in the metropolitan area to areas outside the metropolitan area. To this end, methods to strengthen incentives when relocating data center construction areas, such as subsidy support, tax benefits, financial support, and site rent discounts, have been mainly used, and data center operation efficiency has been improved by local governments, such as mountainous areas with high cooling efficiency or around dams. Methods are also being promoted to actively utilize the location of each region to increase .

Regional dispersion on the supply side means securing small-scale power generation sources, or distributed power sources, near demand sites to meet this dispersed demand. To this end, plans to actively utilize distributed power sources, focusing on renewable energy, are being promoted, and recently, plans to install fuel cells or small modular reactors (SMRs) near metropolitan areas where power demand is concentrated are also being discussed. The advantage of dispersing demand and supply sources to each region and connecting them is that the burden of expanding the power system can be reduced and supply and demand adjustment can be more efficient.

In addition, efforts are needed to resolve the intermittency and uncertainty of renewable energy, which makes it difficult to produce as much electricity as desired at the desired time. To this end, a plan is being promoted to predict power production and demand by linking a virtual power plant (VPP), which controls various distributed power sources in an integrated manner, with the power system, and to secure and supply the necessary amount of power accordingly.

In order to respond to the rapidly increasing demand for electricity and the need to build a new power grid, SK Eco Plant is conducting a solid oxide fuel cell (SOFC) business, which is attracting attention as a representative distributed power source, along with renewable energy businesses such as solar and wind power. It is being actively developed. In addition, a lot of effort is being put into building an eco-friendly data center business model. In particular, the ‘Bupyeong Data Center’, which SK Eco Plant is building with Singapore’s ‘Digital Edge’, is the first data center in Korea with a capacity of 330kW. Solid oxide fuel cells (SOFC) will be installed and will be used as an auxiliary power source, raising even greater expectations.

A bird’s eye view of Korea’s largest Bupyeong data center being built by SK Eco Plant.

In addition, the company has entered the renewable energy power brokerage business in earnest and is also participating in a pilot project to improve the power market system and power trading of 50 MW renewable energy power generation resources by linking 91 renewable energy power plants in Jeju Island based on VPP. In the future, the plan is to expand renewable energy recruitment resources nationwide to stabilize the power grid while also contributing to the expansion of the virtual power plant market. In this way, we hope that SK Eco Plant’s actions in rapidly responding to the rapidly increasing demand for electricity in the AI ​​era will lead to meaningful results.