※ Seoul National University of Science and Technology supports early-stage startups in the deep-tech (AI·Big Data) sector. Based on SeoulTech’s customized growth support program, IT Donga examines promising deep-tech startups that are achieving clear results in the market.


As artificial intelligence (AI) services permeate across industries, the importance of high-performance hardware infrastructure to power them is increasing daily. However, traditional GPU (Graphics Processing Unit) servers pose a significant economic burden, with initial deployment costs alone reaching tens of millions of KRW. They also generate enormous noise and heat and require a large installation space. As a result, many startups and small businesses are often discouraged at the threshold of introducing AI infrastructure.

Startup “XD NODE (CEO Jo Soo-jong)” is a deep-tech company that has focused precisely on this issue. Having mainly engaged in the distribution of high-performance GPU servers and infrastructure deployment, the company is now preparing to enter the market with an ultra-compact AI server equipped with a domestically produced NPU (Neural Processing Unit) specialized for AI inference. IT Donga spoke with XD NODE CEO Jo Soo-jong, who aims to inject a new perspective and vitality into the AI hardware market, to hear about the future of infrastructure innovation the company envisions.

- Please explain your career before founding the company and what led you to start XD NODE.

: Around 2017, I joined a certain partner firm that distributed NVIDIA products as a sales representative. At that time, before the AI boom, NVIDIA was mainly known as a graphics card company, and the enterprise GPU server segment received relatively little attention and generated almost no sales. In that environment, I witnessed firsthand in the field the entire process of the AI GPU server market emerging and growing explosively. Believing that AI would become the core of the Fourth Industrial Revolution, I decided to stake everything on it and founded the company in 2023. XD NODE now focuses on the distribution of NVIDIA GPUs and GPU-based servers, infrastructure deployment, maintenance, and technical support. Recently, we have also started developing our own products.

- What problems did you identify through your business experience?

: Small companies, such as startups, face physical and environmental constraints. Even if they want to introduce traditional high-performance GPU servers, they lack separate space for a server room and do not have enough specialized personnel to manage them. The price is also high. In addition, due to the characteristic enormous noise and heat generated by conventional GPU servers, using them in a normal office is virtually impossible. While working in the field and engaging in direct sales, I witnessed such situations countless times. Many customers ended up compromising with a workstation, but in most cases they were not fully satisfied in terms of performance or quietness. There was a need for a new alternative to solve their concerns.

- What solution have you prepared to overcome those limitations?

: One technology that has recently been attracting attention is the NPU, which features low power consumption. We are developing the “XDN-E100 mini server” that applies this technology. It is equipped with the MLA100, a 25W NPU from Korean AI semiconductor company MOBILINT. The key points of this product are its compact size and low power consumption. In addition to being more affordable than existing GPU servers, its noise, heat generation, and power consumption are dramatically reduced to the level of a standard PC. The size of the main unit is only about that of a couple of Korean dictionaries, so it can be placed directly on an office desk and used immediately.

- In what areas can an NPU-based ultra-compact server be used? How does it differ from existing GPU servers?

: This product is optimized not for AI “training,” which handles massive datasets, but for “inference,” which runs pre-trained models in actual services. It is highly suitable for areas that require real-time image inference, such as vision AI for autonomous driving, object recognition in games, and CCTV-based applications. Not every company needs to withstand massive heat and power consumption to fully utilize the maximum performance of GPU servers costing tens of millions of KRW. Our mini server is scheduled to be launched in the KRW 3 million price range, and if limited to inference tasks, it can be expected to deliver powerful utility comparable to GPU servers in the KRW 10 million range, helping to prevent unnecessary performance and cost waste.


- How did customers react when you introduced the product under development? Please also share your specific mass production schedule.

: When we presented the product concept to customers at the planning stage, the majority showed deep interest. In particular, as many as 30–40% gave an immediate response along the lines of “I want to buy it as soon as it comes out,” confirming the clarity of market demand. We are currently producing prototypes in collaboration with Advantech, a Taiwanese industrial PC specialist, and expect to ship the prototypes in early March. After thorough testing and optimization, if no major issues arise, we plan to launch full-scale mass production and sales in the second half of this year.

- You are using the SeoulTech Early Startup Package program while running a technology startup. How useful has it been?

: It has been extremely helpful for us as a startup. First, thanks to the grant support, we were able to secure excellent talent at the right time. Previously, we only considered loans as a means of financing the business, but through the program we came to understand the importance of investment attraction (IR) activities, which enabled more proactive and aggressive business development. In addition, thanks to the numerous seminars and conference opportunities arranged by SeoulTech, we were able to significantly expand our network with related institutions and partner companies. For early-stage companies aiming to take off, I would strongly recommend that they definitely apply for such support programs.

- What are XD NODE’s ultimate goals and vision in the market going forward?

: The existing server distribution market tends to operate on inertia. While AI technologies and products are growing explosively day by day, many existing market participants are unable to fully keep up with the pace of change. The AI-related hardware market is fundamentally different in nature from the traditional server market. Companies like XD NODE, which view the market from a new perspective, must take an active role in energizing the market.

Simply holding sales channels and engaging only in price competition cannot meet significantly heightened customer expectations and makes it difficult to survive for long. For the many innovative companies that intend to fully develop AI-era businesses, XD NODE will become the most reliable and straightforward technology partner.

Reporter Kim Young-woo, IT Donga (pengo@itdonga.com)

Popular News