"Even Those Are Secondhand" Only 2 Units at Korea's Top KAIST... High-Performance GPU 'Rotation' [AI Talent Development Red Light]

KAIST Has Only Two H100 Servers for National Projects
Korea University: "Researchers Must Share Scarce GPUs"
South Korea Labeled as 'AI Talent Exporter'
Ranks Fifth in AI Talent Outflow Among 36 OECD Countries
Companies Cite 'AI Talent Shortage' as Top Challenge

AFP Yonhap News

"Most GPUs (graphics processing units) owned by universities are low-end models. The high-performance GPU essential for cutting-edge artificial intelligence (AI) research, the NVIDIA H100 series, costs tens of millions of won per unit, making it difficult to purchase. In a single lab, multiple projects run simultaneously, and with government projects and personal research, the H100s are insufficient. Researchers write their names on waiting lists and just wait for their turn. Sharing the limited high-performance GPUs among many makes research challenging." (Choi Byung-ho, Professor at Korea University AI Research Institute)

One of the main reasons it is difficult to recruit AI professors in South Korea is the lack of research infrastructure at universities. The environment is so inadequate for developing skills that many do not even consider coming to universities. On the 10th, Professor Choi said, "Even if there are researchers, without infrastructure, they cannot produce results," adding, "The claim that DeepSeek has over 50,000 H100s exists because that scale is necessary to create foundation models like ‘V3,’ which underpins the recently released ‘R1.’"

Even the Korea Advanced Institute of Science and Technology (KAIST), regarded as the top science and engineering university in the country, has a poor AI research environment. Among about 40 GPU servers for national projects, only two are equipped with the high-end H100. A KAIST official said, "We recently received a GPU server as a donation, but it was secondhand."

Securing AI semiconductors for academic research is also considered important in the industry. Oh Seung-pil, CTO of KT, pointed out, "The reality is that a professor who can secure GPUs well is considered a competent professor." Choi Hong-seop, CEO of Maum AI, said, "For AI researchers, how many GPUs they have used is important for their career," adding, "After spending a year's revenue to secure 100 H100s, good researchers came in to build their careers."

The treatment of AI department professors is also poor. Kim Doo-hyun, a professor in the Department of Computer Engineering at Konkuk University and a member of the Talent and Infrastructure Subcommittee of the National AI Committee, said, "Although 10 AI graduate schools have produced a considerable number of personnel over several years, many leave Korea because salaries are lower compared to overseas companies."

South Korea is currently classified as a ‘net exporter of AI talent.’ According to an OECD survey, as of 2023, South Korea’s AI talent net migration index was -0.3. This index, compiled by the OECD’s AI Policy Observatory, is based on the net inflow and outflow of AI professionals per 10,000 people. A value below zero indicates a net outflow of talent. The data for the index is collected through the professional job-seeking social media platform LinkedIn. Among 36 surveyed countries, South Korea ranks fifth lowest after Israel, Greece, T?rkiye, and Hungary. In contrast, countries with the highest talent inflow include Luxembourg, Switzerland, and Germany.

The outflow of talent directly leads to labor shortages in companies. According to the 2023 AI Industry Status Survey conducted by the Software Policy Research Institute, the labor shortage rate in domestic AI companies reached 14.3%. Among AI personnel, the shortage rates were particularly high for AI system operators and managers (22.9%) and AI consultants (22.1%). The shortage rate refers to the ratio of currently lacking personnel to the total required personnel.

‘AI talent shortage’ was identified as the biggest challenge companies face. In the same survey, 81.9% of respondents agreed with the statement ‘There is a shortage of AI talent’ (44.9% strongly agree, 37% agree). When asked to rate the severity of the AI talent shortage on a 5-point scale, the average score was 4.25.

Industry experts emphasize the need for urgent response as AI semiconductor technology advances at an unprecedented pace. Moon Song-cheon, Professor Emeritus at KAIST Graduate School of Management, warned, "We must keep in mind that technology evolves at an astonishing speed," adding, "By the time the government secures tens of thousands of GPUs, the market will already consider the H100 and even Blackwell (NVIDIA’s next-generation AI chip) as outdated technology."