[Reporter’s Notebook] "Hot Blackwell" Arrives... Time to Address Safety Issues

"Korea faces even greater challenges after adopting Blackwell. This is because there is a significant lack of experience and know-how in building cooling facilities that can operate Blackwell properly in data centers."

A senior executive from an overseas big tech company I met informally last month expressed concerns ahead of the domestic introduction of Nvidia's high-performance graphics processing unit (GPU), Blackwell. Products like Blackwell, which emit more heat than existing GPUs, are scheduled to enter Korea this year, but the cooling technology in domestic data centers has seen little progress, and the legislation regarding data center management is not systematic.

What must follow closely like thread and needle with GPU adoption is a way to safely manage data centers housing thousands or tens of thousands of GPUs. If heat is not effectively dissipated, the chips themselves can burn out, and in the worst case, the data center could come to a halt.

To dispel these concerns, legislative and institutional improvements are essential. A bill to prevent digital disasters is currently pending in the National Assembly. The "Act on the Safety Management of Digital Disasters and Failures" (proposed by Representative Choi Hyung-doo) requires major operators to establish management plans annually, designate responsible persons, and allows the Minister of Science and ICT to conduct investigations in the event of a disaster. It essentially covers management measures throughout the entire digital disaster cycle, from prevention to preparedness and response.

However, the bill’s progress is sluggish. Companies oppose it, arguing that the obligations and penalties in the event of digital accidents are excessive. The tech industry is well aware that liquid cooling systems using coolant are essential to operate Blackwell stably. However, the installation cost of liquid cooling equipment is a burden, increasing by 40 to 60% compared to air cooling systems.

Everyone agrees on the importance of digital safety management. The bill discussions need to focus on prevention rather than post-incident punishment. Practical accident prevention measures, such as cooling technologies tailored to data center performance and step-by-step management plans, should also be included in the law.

Failure to operate AI technology safely causes direct harm not only to corporate management but also to the daily lives of citizens, as demonstrated by the "Kakao outage incident" caused by the SK C&C data center fire in 2022. At that time, not only the KakaoTalk messenger but also KakaoT and KakaoPay were completely paralyzed, causing chaos. Small business owners also suffered inconvenience as customer consultations and payments were disrupted. At a time when leading big tech companies are planning to build AI data centers handling vast amounts of customer information, safety must be the top priority. GPUs that cost tens of millions of won each must be operated without risk to deliver their full value.

This content was produced with the assistance of AI translation services.