Exclusive: SK Telecom GPUs Underperform, Hitting National AI Timeline

-

Overview of the GPU Performance Issues

SK Telecom and Naver have acknowledged that the GPUs they leased to "National Representative AI" companies did not operate at normal speeds for a period. Both companies are part of the five consortia selected for "sovereign AI" development and also serve as GPU leasing providers to these consortia. The issue has raised concerns about the reliability and performance of the hardware used in critical AI development projects.

Details of the Problem

A source from the telecommunications industry revealed on the 19th that the GPUs SK Telecom leased to the National AI did not perform at normal speeds for about a month starting in August. This led to difficulties for the companies selected for the National AI in early August, as they struggled to properly train their foundation models during that time. The government had previously selected SK Telecom in July for a public offering to install 1,000 units of the latest B200 GPUs. As a result, SK Telecom was required to lease the GPUs to National AI companies from August to June of next year.

However, the GPUs provided by SK Telecom reportedly failed to operate at normal speeds from the outset. National AI companies found the GPU speeds provided by SK Telecom significantly slower than their existing GPUs, making it difficult to train AI models normally. Consequently, the Ministry of Science and ICT held a closed-door meeting with National AI companies in September to resolve the issue.

Resolution and Explanations

SK Telecom explained that the problem stemmed from incompatibility between each company’s development environment and their development code. A source from the Ministry of Science and ICT stated that there were initial difficulties, and optimization took time during the integration of each model with the GPU infrastructure. They added that the issue has now been resolved, and the GPUs are operating normally.

Some critics argue that National AI companies, which could not use the GPUs properly for a month due to the speed degradation, may face relative disadvantages. In the government-led first evaluation at the end of this year, one of the five National AI teams will be eliminated. Companies that leased SK Telecom’s GPUs faced a month of difficulties, putting them at a disadvantage.

SK Telecom's Response

Regarding the GPU speed degradation, an SK Telecom spokesperson stated that this was not an infrastructure issue on SK Telecom’s part, and the companies resolved it with guidance from NVIDIA. They added that they understand the period of subnormal speeds was less than a month.

Naver Cloud's Involvement

Meanwhile, Naver Cloud, another GPU leasing provider and a National AI participant, also reportedly provided GPUs that failed to operate at normal speeds for a certain period. Naver Cloud leased 1,000 units of NVIDIA’s H200 GPUs to National AI companies. However, similar issues reportedly prevented National AI companies from properly using the GPUs during the initial phase of the project.

Summary of the Situation

The incidents involving both SK Telecom and Naver Cloud highlight the challenges associated with large-scale GPU leasing for AI development. While both companies claim the issues have been resolved, the initial setbacks have raised questions about the preparedness and coordination required for such high-stakes projects. The impact on the National AI teams, particularly those facing elimination in the upcoming evaluation, underscores the importance of reliable infrastructure and seamless integration in AI development.

Key Points

  • SK Telecom and Naver Cloud faced GPU performance issues for National AI companies.
  • The problems led to difficulties in training AI models, affecting the development timelines.
  • Both companies attributed the issues to compatibility and integration challenges.
  • The Ministry of Science and ICT confirmed that the issues have since been resolved.
  • Critics believe the setbacks may have put some companies at a disadvantage in the evaluation process.
  • Naver Cloud also reported similar GPU performance issues for its leased units.

Comments

Post a Comment

Popular posts from this blog

Japan Firms Leverage Satellites and AI to Locate Abandoned Homes for Sale

-
Image
The Rise of AI-Powered Real Estate Discovery in Japan In Tokyo, a growing number of Japanese companies are leveraging satellite imagery and artificial intelligence to identify abandoned houses that could be repurposed or sold. This innovative approach is becoming increasingly vital as the country faces a rising number of deserted properties due to its aging population. Where Inc., a startup, offers a service that uses AI trained on tens of thousands of images to detect signs of aging roofs, such as rust and discoloration. These findings are then overlaid on satellite maps, helping users spot properties that may have been left unattended. One of Where Inc.'s users, Kotaro Yasue, recently discovered a two-story wooden house through this service. He was able to contact the owner via a real estate registry. It turned out that the house had been neglected for over a decade, and the owner was unsure how to proceed with its disposal. Yasue, who runs a house rental service company in G...
Read more

GEN Nigeria and Others Target Indigenous Firms with New Tech Fees

-
Image
Empowering Nigerian Entrepreneurs with AI Global Entrepreneurship Network (GEN Nigeria), in collaboration with Fidelity Bank Plc and other organizations, has called on young entrepreneurs in Nigeria to embrace new technologies, particularly Artificial Intelligence (AI), in their business operations. This initiative was highlighted during the Youth Entrepreneurship Dialogue, titled “Building the Future with AI: Technology, Talent, and Transformation Entrepreneurship,” which took place at the 2025 Global Entrepreneurship Week (GEW) in Lagos. Dr. Olawale Anifowose, the Managing Director of GEN Nigeria, emphasized that AI has become essential for local enterprises in Nigeria to boost productivity, drive innovation, and enhance competitiveness. According to Anifowose, this can be achieved by automating tasks, personalizing customer experiences through data analysis, and enabling entrepreneurs to develop and scale new products and services more efficiently. Enhancing Efficiency and Produc...
Read more

I Was Healthy Until a Dirty Habit Stopped My Heart

-
Image
A Rare Case of Cannabis-Induced Cardiac Arrest in a Young, Healthy Woman A 26-year-old woman from Georgia experienced a life-threatening cardiac arrest after smoking marijuana. The incident occurred when her fiancé discovered her unconscious on the bathroom floor, prompting immediate emergency medical intervention. Upon arrival at the hospital, doctors confirmed that she had suffered a cardiac arrest, where the heart suddenly stopped beating. The woman had no known pre-existing health conditions and tested negative for heart-related issues such as arrhythmia. However, she was found to have high levels of tetrahydrocannabinol (THC), the psychoactive compound in cannabis responsible for the "high" sensation. Doctors diagnosed her with ventricular fibrillation, a severe heart rhythm disorder that prevents the heart from effectively pumping blood and oxygen throughout the body. This led to what they described as a "cannabis-induced cardiac arrest." This case has rais...
Read more