“Data center plans likely have half as phantom demand… Limit power for AI to renewable energy” - 경향신문

This article was translated by an AI tool. Feedback Here.

Lee Heon-Seok, Policy Committee Member at Energy Justice Action, explains the problems with pushing AI data centers during an interview held on the 15th at the Kyunghyang Shinmun office in Jeong-dong, Jung-gu, Seoul. He said, “The large-scale promotion of AI data center construction will lead to excessive carbon emissions,” adding, “Electricity should be supplied from renewable energy.” Kang Yoon-Joong, Senior Reporter

Government-promoted 18.4 GW of data centers could spark conflicts over electricity rates and moreProperly reviewing power demand can prevent building more power plants than neededPoliticians and others call them ‘a goose that lays golden eggs’, but they do not greatly help regional developmentTo build AI that serves the environment, regulations such as ‘power usage effectiveness’ and ‘power impact assessments’ are necessaryGiven the enormous social costs, consider placing responsibility on operators as in the United States

The three mega-projects that the government is ambitiously promotingsemiconductors, artificial intelligence (AI) data centers, and physical AI (robots)are the largest projects in national history, with about 1,500 trillion won to be invested across the three sectors. The goal of overcoming the growth slowdown, promoting balanced regional development, and achieving a great leap forward for the Republic of Korea raises expectations. However, some worry that the mega-projects will exacerbate the climate crisis by increasing greenhouse gas emissions. At the ‘919 Climate Justice March’ held nationwide on the 19th with the participation of 620 civic groups spanning environment, labor, human rights, religion, and youth, participants even called for halting the mega-projects.

Opposition to building AI data centers is fierce not only in Korea but also in the United States. In states such as New York and Texas, governors have declared moratoriums freezing new data center construction, and in Virginia and Arizona, extra-large data center projects have been withdrawn or delayed due to resident lawsuits and denials by local governments. Because data centers consume massive amounts of electricity and water, electricity rates can rise and water depletion is a concern. Vibrations and noise from cooling fans running 24 hours a day also plague residents. Data centers, essential infrastructure in the AI era, are being stigmatized as undesirable facilities.

We met Lee Heon-Seok, Policy Committee Member at Energy Justice Action, on the 15th, who has tracked the problems of data centers and proposed alternatives. He warned, “The data center construction plans within the mega-projects are greatly inflated beyond actual demand, creating excessive electricity demand,” and “This will lead to an increase in liquefied natural gas (LNG) power plants and higher operating rates at coal-fired power plants, both obstacles to carbon neutrality.” He added, “To build AI that does not harm the environment, appropriate regulations are needed, such as limiting data centers to electricity sourced from renewable energy.”

- Semiconductors are a field where Korea excels, so nurturing them as a future growth engine is understandable. But why are AI data centers included in the three mega-projects?

“Samsung Electronics and SK Hynix are in the semiconductor industry. Data centers similarly fit the industrial makeup of groups like SK and GS. These groups have energy affiliates such as oil and gas companies and also operate power plant businesses. SK is building a data center in the Ulsan Mipo National Industrial Complex, where it operates an LNG power plant. GS is promoting data center construction in Donghae, Gangwon, and runs a coal-fired power plant nearby. Data centers consume enormous amounts of electricity. When new data centers are built, more power plants must follow, letting these groups kill two birds with one stone.”

- Among the three mega-projects, you see especially many issues in the data center segment. Why is that?

“The semiconductor industry has tangible elementswhere to build a factory and at what scalebut data centers lack such substance. Sites have not been finalized, and, most importantly, usage demand for data centers is unclear. The data centers pushed as part of the mega-projects total 18.4 GW (gigawatts), but I believe more than half of that demand is inflated. Because of this phantom demand, we are being forced to build more power plants than necessary. This will trigger social conflicts over matters like power plant construction and electricity rates. That is why a proper review of data center electricity demand is needed.”

- For ordinary citizens, 18.4 GW may not be an intuitive scale.

“To build a 1 GW data center requires land area equal to dozens of soccer fields. It means 1 GW of electricity is needed, comparable to the output of one large nuclear reactor. So 18.4 GW is enormous.”

- Are AI data centers different from existing data centers?

“Most currently operating domestic data centers are used by financial institutions, internet portals, telecom companies, and large corporations for data management and website operations. Inside a data center, you see rows of racks that look like cabinets. Servers are stacked inside these racks. The racks going into new AI data centers are similar in volume to those in existing centers, but they weigh more than twice as much and use over ten times the electricity. They house entirely different kinds of servers. Typically, building a 1 GW-class AI data center costs around 60 trillion won. A single cutting-edge AI rack costs over 10 billion won, and a 1 GW center needs 3,000~4,000 such racks. Raising this funding is not easy. The massive number of graphics processing units (GPUs) required for data centers are also difficult to procure. In particular, these expensive racks are not like factory equipment that can be used for 10 or 20 years. At most they last 5~7 years before needing replacement. It is similar to the lifespan of ordinary computers.”

- What is the problem with the data center construction plan the government is currently pursuing?

“Because construction involves huge cost risks, data center operators cannot build facilities first and then tell customers, ‘Use ours.’ They must take pre-orders and sign contracts, then build accordingly. The plan to build 18.4 GW reflects the hopes of the government or companies, not contracted volumes. The problem is building power plants based only on corporate wish lists. It would be best to construct data centers and power plants together after eceiving orders and then operate them together. But while data centers can be built in 1~2 years, coal or gas-fired power plants take 6~7 years, and nuclear plants 13~14 years. That means power plants have to start first. If you build power plants trusting the massive data center plans and then those data centers do not materialize, the plants become unnecessary. The government plans to prepare electricity to match 18.4 GW as announced. Major confusion later will be unavoidable.”

- In the United States, public opposition to data centers has intensified and become a political issue. What is the situation in Korea?

“Data centers require enormous electricity and water, and noise is a serious issue, yet in Korea there is no discussion of creating appropriate regulations. Instead, local governments are saying, ‘We will support you, come to our town.’ As public opinion in the U.S. soured, President Donald Trump shifted to telling big tech to build data centers but generate the electricity for them themselves. In Korea, however, the government and public enterprises say they will take care of the power and more. But the government cannot shoulder matters involving vast social costselectricity, water, environmental issuesby itself. That only deepens conflicts. We should take note of the U.S.-style approach of having operators handle it themselves.”

- Ireland once actively attracted data centers from U.S. big tech as a matter of policy.

“The Ireland case can serve as a cautionary tale for us. Even before AI took off, Ireland's policy was to foster the data center industry. Data centers sprang up everywhere, and multiple problems emerged. Most notably, electricity rates rose sharply and power shortages became a concern. Data centers account for more than 20% of Ireland's total electricity consumption. Ireland supplies 40% of its total electricity from renewable energy. That is a fairly high share, but it still could not keep up with data center demand, so many new gas-fired power plants were recently built. Due to these issues, Ireland has begun to regulate.”

- You recently visited Korea's east coast regions where data centers are planned.

“Gangneung and Donghae are where SK and GS aim to build AI data centers. Another company besides these two conglomerates is also preparing a data center. In GS's case, the planned site is right next to the GS Donghae coal-fired power plant they operate. It looks clear they intend to draw electricity from their own coal plant. In Gangneung, there is the Anin coal-fired power plant, and near it a project entity called Gangneung DCPEF is pushing to build a 1 GW data center. As I understand it, the project is being pursued via project financing. Although this is a project costing tens of trillions of won, the company's name is unfamiliar, and it is unclear who will use the data center and who the real owner is. Because it is a small operator, even filing information disclosure requests is difficult when residents have questions about environmental impacts.”

- If east coast data centers are supplied with electricity from existing coal-fired power plants, is that not acceptable?

“The many coal-fired power plants built along the east coast currently lack transmission lines to send power to the Seoul metropolitan area. So their capacity factors average only 20~30%. In the past, domestic coal plants had capacity factors reaching 90%, but recently, because east coast plants run less, the nationwide average is below 50%. If data centers are built on the east coast, they will be able to draw power from nearby coal plants, and capacity factors will jump. Operators such as GS will earn more money, but carbon emissions will increase.”

- Last month, the government released a draft of the 12th Basic Plan for Long-Term Electricity Supply and Demand, projecting that the mega-projects will require an additional 26.6 GW of power. The challenge is how to procure an amount of electricity equivalent to 19 large nuclear reactors.

“Although the government says it will maximize renewable energy to cover the demand, the amount of power needed is so large that it will not be easy. In the end, I believe new nuclear or LNG power plants will be built, and utilization at coal-fired power plants will inevitably rise. That is the worst scenario for carbon neutrality.”

- The government and industry are likely planning vast data center capacity not only for domestic demand. Are big tech firms looking to build in Korea and use those centers because of opposition in the United States?

“Although the government does not state it explicitly, it is clear these will be used for leasing to overseas firms such as big tech. Data travel over internet lines, so from the perspective of U.S. big tech, it does not matter where the data center is located. Only costs matter. If using a data center in Korea is cheaper than in the United States, they may come. As a result, there is growing movement to build data centers in developing countries where costs are low. India and Malaysia are often mentioned. Although the hot climate is disadvantageous for cooling, very low land prices, labor costs, and electricity rates draw them there.”

- That is similar to advanced countries exporting polluting industries to developing countries.

- It is said that data centers are necessary to become an AI powerhouse.

“The core of the AI industry is not data centers but developing proprietary foundation models like ChatGPT. Owning many data centers does not mean possessing core AI technology. We do not call Ireland, which has many data centers, an AI powerhouse.”

- Some regions are actively trying to attract data centers.

“The problem is that politicians and others tout AI data centers as a ‘goose that lays golden eggs’. They say data centers will create local jobs and boost tax revenue. Because tens of trillions of won are invested to build them, rural communities mistakenly think all that money will flow to them. But most of the money goes to buying servers and chipsthat is, computers. The local economic effect appears only briefly during construction of the facility. Moreover, once completed, data centers do not require many people. Computers do the work, and a small number of maintenance, repair, and cleaning staff is enough. Tax revenue will increase, but given the electricity and water consumed, it is not enough to significantly help regional development.”

- Still, data centers are essential in the AI era. Is there a wise way to resolve the problems?

“The indicator showing the share of electricity used by core components such as chips and serversas opposed to cooling or lightingat a data center is ‘power usage effectiveness (PUE)’. If the PUE is around 1.2, it is a highly efficient data center. That means about 20% of total electricity goes to cooling and the like. At existing domestic data centers, the figure is 2.4. More electricity goes to cooling than to servers by a factor of 1.4, meaning the tail wags the dog. Even China has PUE regulations, allowing only highly efficient data centers to be built, but Korea still has no PUE standard. We need data centers. However, because their environmental impact is large, they must be built only as needed and in line with necessary regulations. Yet under the ‘Special Act on AI Data Centers’, if a data center is built outside the Seoul metropolitan area, it does not even have to undergo a power impact assessment. The environmental impact assessment rules are also ambiguous, so in most cases they are not conducted. First, we need proper environmental regulations. Currently, community briefings are not mandatory when building a data center. If operators conceal information, residents have no way to know. As a result, residents feel vague anxiety. This must also be fixed.”

- How should the power demand be covered?

“Use of renewable energy should be the principle. In China and Europe, there are rules that all new data centers above a certain size must use renewable energy. The National AI ComputingCenter in Gwangju, South Jeolla, Haenam recently broke ground. Right in front of the center is the largest solar power facility in the country. If connected like this, it is ideal. It will not be easy immediately, but with a phased approach it is possible. If it cannot be connected to a renewable complex next door, there are ways to buy renewable energy generated in other regions. Only by doing so can we achieve carbon neutrality.”

- The public, as well as the government and companies, has very high expectations for the AI industry.

“When problems with AI or data centers are raised, there is a tendency to treat it as taboo, as if it blocks national development. Such an attitude can make the problem worse. Despite many debates, we will eventually live in an era when everyone uses AI. Pointing out problems does not mean opposing AI; it means asking what electricity we will use even when we use AI, how we will protect the environment, and what social deliberation we will undertake. This involves not only environmental issues but also labor and personal data management. We need to find a way to establish in Korea ‘AI for people and the planet’ that does not destroy the environment and does not violate labor rights or personal data. Focusing only on technology development and profit may seem to speed things up, but it will run into greater resistance. I hope to see debate on achieving a proper AI transition through appropriate regulation.”