Kakao has officially launched an artificial intelligence (AI) supercomputer with performance ranking among the top 30 in the world, starting operations last month. This supercomputer is the first large-scale system established through the government's AI infrastructure development initiative and is expected to significantly enhance the capabilities of Kakao's AI agents.
According to the AI industry on June 21, Kakao Cloud, the AI infrastructure subsidiary of Kakao, has completed the construction of a GPU cluster based on NVIDIA's latest graphics processing unit (GPU), the HGX B200 (Blackwell), through the Ministry of Science and ICT's GPU acquisition, construction, and operation support project. Testing began at the end of May.
The new supercomputer consists of 255 nodes, with a measured performance of 69.1 petaflops (PFLOPS) according to the High-Performance Linpack (HPL) benchmark, and a theoretical performance of 78.65 PFLOPS. This represents more than double the performance of the previous 32 PFLOPS supercomputer. PFLOPS is a unit that measures the ability to perform one quadrillion calculations per second.
When applying these figures to the global supercomputer ranking 'TOP500' as of November last year, it corresponds to a ranking of 33rd in the world. However, due to the recent acquisition of NVIDIA's latest GPUs by major countries in the U.S., Europe, the Middle East, and Asia, it is expected to rank around 40th as of June this year.
Kakao plans to utilize the computational resources for its collaboration with OpenAI, its own AI service 'Kanana', AI agent development, and internal business innovations.
Industry experts view this achievement as a significant milestone in the competition for AI infrastructure, indicating a shift in the industry's competitiveness from model development to securing and operating large-scale computational infrastructure. As a result, the competition among Kakao, NHN, Samsung Electronics, SK Telecom, and LG AI Research for next-generation supercomputers is expected to intensify.
NHN Cloud, which also participated in the government's GPU acquisition, construction, and operation support project, is reportedly finalizing its next-generation AI supercomputer based on the B200 and preparing for its launch. Industry analysts predict that NHN's new supercomputer will also achieve performance in the early 30s in the global rankings as of June.
In particular, Samsung Electronics is expected to secure performance in the 15th range globally for its next-generation AI supercomputer currently under construction. If realized, it would become the highest-performing supercomputer among private companies in South Korea. The specific performance details of the next-generation supercomputers from Samsung, NHN, and other major domestic AI firms are anticipated to be revealed in the fourth week of June.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.