XiiLab: Optimal Partner for the AI Highway, Cutting GPU Management Costs by 50%
We have 130 types of vision models. We also have technology that reduces GPU management costs by more than 50 percent. We are proud to be the best partner for the government-propelled AI highway.”
Yoon Se-hyuk, CEO of XiiLab, said in a recent interview with GDNet Korea, “We have technologies and solutions that optimize GPU usage and reduce costs.”
As President Lee Jae-myung presents three major AI initiatives as the nation’s core growth strategy and first pledge, there is growing public consensus that Korea should also accelerate the construction of AI highways. The government plans to invest 100 trillion won over the next five years to secure more than 50,000 GPUs and build AI highways through the establishment of data centers nationwide.
Yoon emphasized that XiiLab is the optimal partner for the government’s AI highway. Founded in 2010, XiiLab is a vision AI company. Based on AI-driven video analysis technology, the company operates in three areas: ▲ AI infrastructure ▲ Vision AI ▲ Digital Twin. It has partnered with global IT companies such as Nvidia, HPE, and Dell. Notably, the company has secured hyperscale infrastructure technology that enables stable operations by clustering a large number of Nvidia GPUs, reducing GPU operating costs. “We plan to secure technology to connect tens of thousands of Nvidia GPUs by next year,” Yoon said.
XiiLab has developed vision models since 2016 and currently holds 130 types, with plans to increase the number to 170 by the end of this year. For the so-called “AI Expressway,” comparable to the Gyeongbu Expressway during the era of industrialization, XiiLab provides a GPU-optimized solution called “AstraGo.” “As building the AI Expressway requires significant investment, adopting cost-effective solutions will also benefit the nation,” Yoon pointed out.
Yoon was appointed CEO in March this year. He graduated from Seoul National University with a degree in electrical and computer engineering and earned a master’s degree in computer vision. He previously worked at KT’s New Business Strategy Office and served as head of the Management Planning Department at Engenbio, where he played a key role in the company’s stock market listing. He is regarded as a convergence expert with both technical expertise and insights in finance and strategy.
Currently, XiiLab is focusing on discovering invisible value through AI under the slogan “Seeing the Unseen.
With the slogan “Touching the Untouchable,” the company is taking a leap forward as a physical AI enterprise that applies AI to real-world environments. By linking physical data in digital twin environments using VLM (Vision-Language Model) and NVIDIA’s Omniverse platforms, the company is drawing attention by simulating experiences beyond human reach through AI and presenting new value. Below is a Q&A session with CEO Yoon.
– You’re emphasizing physical AI…
“Physical AI refers to technologies and systems where AI models interact with real physical environments such as robots and manufacturing facilities, going beyond mere analysis. XiiLab is expanding its business scope by applying AI to manufacturing sites and practical industrial domains such as bio, autonomous driving, and robotics through digital twin and VLM technologies.”
– You’ve been working on Vision AI for a long time. How do you view the market prospects for Vision AI and Physical AI?
“Vision AI is growing at a global average CAGR of 21.5%. XiiLab is evolving into a physical AI company through the convergence of digital twin and VLM technologies, moving beyond video analysis in the Vision AI market. We have developed and secured the core engines that power physical AI. Our Vision AI solutions include XAIVA and VidiGo. They are used in large-scale manufacturing facilities to monitor safety equipment, detect risks, identify ultrafine foreign objects, and determine product defects. We are confident that AI technology will become a new standard as it deeply permeates industrial sites and personal daily lives.”
– What kind of solution is AstraGo, your core product? What value does it provide to companies or institutions?
“AstraGo is a solution that enables large-scale GPUs to operate efficiently. It reduces GPU management costs by more than 50 percent and configures a machine learning (ML) environment in just one minute. It is highly scalable, with compatibility across platforms such as HPE and Dell. AstraGo applies clustering technology that allows multiple GPUs to be shared by multiple users or pooled for a single purpose. Even if the number of GPUs or users increases, AI infrastructure can be operated stably without losing GPU efficiency. We are also upgrading AstraGo’s clustering technology. By next year, we plan to secure hyperscale infrastructure management technology to reliably operate tens of thousands of GPU clusters and support national AI GPU infrastructure expansion policies.”
– I’m also emphasizing the Digital Twin business…
“Last year, we secured the right to sell Nvidia’s digital twin platform, Omniverse. With Omniverse-based digital twin technology, we are advancing our own technologies, including 3D modeling, physical engine optimization, and scenario-based AI linkage that can simulate industrial data.
In particular, we are contributing to simulation-based process optimization and stability prediction by applying this technology to large domestic semiconductor and manufacturing lines. In addition to semiconductors, we are preparing to supply solutions to global automakers. Our digital twin combines our own virtual environment data generation solution XGen with the Omniverse platform, driving innovation across various industrial sites.
Beyond building environments, the XiiLab Digital Twin project plays a key role in the transformation toward physical AI that we are promoting. It provides simulations by linking physical data with digital twin environments and integrates with autonomous robotics. By connecting driving environment data with the digital twin, it not only derives optimal robot movement paths but also enables process optimization. Our strategy is to convert this into physical AI through physical engine linkage within the digital twin, and we are realizing this through the Omniverse platform.”
– What is the VLM technology you are developing yourself? How do you use it?
“Vision-Language Model (VLM) is a technology that explains video in text or finds video scenes that match text queries. XiiLab’s vision AI solution is called XAIVA. Using this solution, we have implemented accurate object recognition and customized event detection even in complex industrial video environments. This year, we plan to introduce core VLM technology that enables explanatory video analysis in autonomous driving, logistics, and manufacturing sites. We will expand the market by applying high-precision Vision AI to various industries.”
■ Product and Industry Applications
– The industrial use of the Vision AI solution “XAIVA” is increasing. I heard that the data is at least usable…
“XAIVA is an AI image analysis solution that enables large-scale image analysis in the construction and security industries through real-time object detection and behavioral analysis of CCTV. It is mainly applied to the manufacturing industry, but in the bio sector, it is used for GMP hygiene management, such as checking for safety gear and detecting exposure to foreign substances.
It is also applied in the cosmetics sector, where it is used to detect defects in cosmetic containers. This is possible because it is easy to create industry-specialized Vision AI models. Since 2017, we have collaborated with the Ministry of National Defense to conduct research on synthetic data generation technology. With synthetic data augmentation, we can complete desired Vision AI models even with insufficient data.”
– Please introduce the recently released “XAIVA On-Device” and “XAIVA Micro” products.
“XAIVA On-Device is an AI quality and hygiene management solution for industrial sites subject to bio and GMP (Good Manufacturing Practice) regulations. It integrates a GPU device and lightweight AI models in a kiosk, making it easy to deploy at sites. In addition, AI object detection and tracking technology can automatically check workers’ sanitary clothing and movements within one second. It increases on-site efficiency with 99 percent accuracy.
In particular, high-performance image analysis is possible even in low-end GPU environments by applying lightweight AI models to the on-device system. This is the major advantage and differentiator of XAIVA On-Device. It is specialized for GMP industries such as bio and pharmaceuticals.
In addition, “XAIVA Micro” is an ultra-precise AI image analysis solution for quality inspection in semiconductor and advanced manufacturing processes. In semiconductor part manufacturing, wafer alignment accuracy was achieved at less than 0.5 pixels. Each device handles 330 images with ultra-high-speed analysis capability of 3 ms or less. It greatly improves inspection accuracy and efficiency. It can detect more than 99.9 percent of minute defects such as contamination and scratches by scanning entire wafers in real time. We have applied XiiLab’s proprietary synthetic data generation technology to XAIVA Micro, enabling the development of high-precision inspection models with only small amounts of data.”
■ Business Performance at Home and Abroad
– What are some notable results from your recent business performance?
“XiiLab is achieving meaningful results in three areas of business. First, in the Digital Twin segment, we successfully carried out the construction of a semiconductor manufacturing process Digital Twin and achieved sales growth of more than 600% year-on-year. The Digital Twin business in semiconductors, which began as a pilot project (PoC) in 2022, transitioned to full-scale operations last year, and the process line was expanded. It will continue to expand this year. Using a platform based on Nvidia’s Omniverse, we combine AI simulation with physical engines and are pushing forward the transition to Physical AI.
In the AI infrastructure sector, the GPU optimization solution “AstraGo,” released last year, is expected to gain significant market traction this year. AstraGo has significantly strengthened GPU cluster management functions compared to existing products. It is being supplied mainly to large enterprises and conglomerates with large-scale data centers.
In the Vision AI segment, we are expanding sales in the semiconductor and bio industries based on our two new products, XAIVA On-Device and XAIVA Micro.”
■ Financial and Investor Perspectives
– Recently, you initiated a paid-in capital increase of about 18 billion KRW. What is the main purpose of this decision and your future growth strategy?
“A paid-in capital increase is a strategic decision to accelerate market expansion. Now is the right time to make such an investment. Specifically, we began increasing manpower and technological resources to internalize our business in line with rapid changes in the AI environment, such as large-scale national AI infrastructure investments, the shift of AI market trends from LLM to Vision AI, and increasing demand for Nvidia Omniverse platforms.
The funds secured will focus on ▲ strengthening VLM and Physical AI R&D ▲ expanding GPU cluster-based data centers ▲ advancing AstraGo hyperscale infrastructure technology ▲ and expanding overseas partnerships and sales networks. In addition to ensuring short-term profitability, we will pursue mid- to long-term technology internalization. With continuous technology validation and project expansion, we aim to provide customized AI support for complex industrial sites, lowering barriers to AI adoption for companies and driving substantial change.”
■ Future Plans and Shareholder Messages
– You said you would increase sales by more than 10 times in the next five years. Is that achievable? Isn’t it too challenging?
“Last year’s sales amounted to 9 billion KRW. The AI infrastructure market is expected to grow significantly both domestically and internationally. Our AstraGo solution is a cost-effective optimization solution for AI infrastructure. In addition, if Vision AI market growth, industrial application strategies, and Digital Twin business expansion progress as planned, we believe we can achieve growth that is 10 times higher than our current scale.”
– Lastly, if you have a message to shareholders and investors
“XiiLab is at a new inflection point. Through strategic investment and global expansion, we aim to go beyond Vision AI to become a global AI leader ushering in the era of Physical AI. We will focus on customer applications and deliver practical achievements in pursuit of sustainability and growth.
The public are the true owners of listed companies. As an AI company, we are constantly thinking about how we can benefit the country. What we can contribute through AI is making GPU use accessible and efficient for everyone. We believe we can achieve this best here in Korea.”
댓글
댓글 쓰기