LabUp CEO Shin Jeong-kyu: 'We Identify Problems Before the Market Opens, Aiming for 6x Revenue in 3 Years'

"LabUp not only solves customer problems but also identifies issues that customers may not yet recognize. We must define problems ahead of the market to provide solutions that can be immediately utilized when the market opens," said Shin Jeong-kyu, CEO of LabUp, during an interview at the company's office in Gangnam, Seoul, on August 28.

Shin emphasized the company's core competitiveness, noting that LabUp has proactively developed related technologies in anticipation of changes in the AI infrastructure market, which is shifting from model training to inference and then to agent-based AI. LabUp aims to go public on KOSDAQ in October and plans to accelerate its global market strategy with the funds raised from the IPO.

"We confirmed the high interest in the AI sector. As a company developing AI infrastructure systems, LabUp will supply optimized infrastructure to domestic and international AI companies and contribute to reducing the social costs associated with infrastructure construction and operation."

"I believe the market has recognized our ability to predict changes in AI infrastructure ahead of time and prepare the necessary technologies. Since our founding, we have been about four years ahead in technology and approximately one year ahead in commercialization. In the past, operating GPUs for large-scale model training was crucial, but with the emergence of ChatGPT, inference that can handle multiple models and users simultaneously has become a key capability. Recently, as reinforcement learning-based post-training has spread, the boundaries between training and inference are disappearing. LabUp prepared related platforms before these changes manifested as actual market demand."

"We submitted our securities registration statement last Tuesday. If all goes as planned, we expect to complete the KOSDAQ listing by the end of October."

"We plan to prioritize investments in R&D infrastructure and solution enhancement. To create a good infrastructure solution, it is essential to operate large-scale infrastructure and have firsthand experience in validating new equipment. Setting up one NVIDIA B200 system costs about 1.5 billion won, and the next-generation Vera Rubin system is expected to require about 6 billion won per rack."

"The biggest difference is that we possess end-to-end technology that encompasses the entire AI infrastructure. Solutions that combine various open-source tools often have limitations in organically optimizing the entire system due to the separation of infrastructure, machine learning operations, and inference layers. LabUp's Backend.AI can handle the reallocation of GPUs and the acquisition of additional resources on a single platform, even in situations where user demand suddenly increases."

"I cannot disclose specific figures for individual clients. However, I recall that in a proprietary AI foundation model project, we reduced the response time for failures to one-fortieth of the original. In a virtualized environment, we achieved performance levels about 97% of bare metal. Excluding resources needed for operating systems, we are pushing GPU performance to near hardware limits."

"LabUp's strength lies in its versatility. The AI semiconductor market is rapidly changing, and demand is diverse, making it difficult for a single company to dominate all areas. Currently, LabUp supports about 14 to 15 types of AI semiconductors. Since 2018, we have been responding to various AI semiconductors, starting with Google's TPU. We are also collaborating on the NPU verification projects of domestic companies like FuriosaAI and Rebellion."

"We plan to focus on Southeast Asia and the U.S. first, and then expand into Europe. The U.S. market is dominated by big tech companies, making it challenging for external solutions to enter, as large AI firms often have thousands of in-house infrastructure personnel. However, I believe we need to secure our unique entry routes before the market becomes more entrenched. In Southeast Asia, large neo-cloud providers are emerging rapidly. Given the fast growth of the market, I think it is crucial to establish partnerships with local players for early entry."

"In a word, it just works. In the AI infrastructure market, many solutions improve stability through customer testing after product release. LabUp can provide 'day-one ready' solutions that clients can implement immediately, based on our extensive experience, particularly in operating products in the enterprise market and encountering various exceptions from hardware to actual AI services."

"While LabUp has extensive experience in operating AI workloads for businesses and developers, we lack experience in large-scale AI services used by the general public. Through the Everyone's AI project, we hope to identify technical issues arising from user patterns and address them. I believe we can gain experience in discovering bottlenecks in large-scale services and optimizing AI systems and costs."

"Our goal is to establish a platform that manages and integrates AI agents distributed across various locations. In three years, the number of AI agents with human-level intelligence will far exceed that of humans. We anticipate that many office tasks will be handled by AI. LabUp aims to be a company that provides a 'playground' for AI to operate in an environment where humans and AI coexist. If the market grows as expected, we project that our revenue will increase sixfold in three years. LabUp's focus is on software; rather than owning cloud infrastructure directly, we will concentrate on software and intelligence to efficiently utilize any hardware."

Copyright ⓒ Aju Press All rights reserved.