“Next-generation AI infrastructure must deliver scalability and flexibility while simultaneously achieving three key benefits: energy savings, enhanced security, and operational efficiency. KTNF will overcome these challenges with its technological prowess and lead the innovation of Korea's future data centers."

The Artificial Intelligence Semiconductor Forum held its 13th breakfast lecture at the InterContinental Seoul Parnas in Samseong-dong, Seoul on the 12th.

At the event, KTNF CEO Lee Jung-yeon gave a lecture on the topic of “Next-Generation Data Center Architecture in the AI Era.”

CEO Lee Jung-yeon provided an in-depth explanation of the rapid advancements in AI technology, the resulting changes in data centers, and the company's vision for the future.

The representative first reviewed the current state of global AI competition.

Major countries such as the United States, China, and South Korea are accelerating the development of large-scale language models (LLMs). The US's GPT and Gemini models are ultra-large models with 1 to 2 trillion parameters, while China's mid-sized models centered on open source are mainstream, and South Korea's small models that emphasize efficiency are mainstream, he explained.

In particular, he mentioned that efficient models such as 'DeepSeek', which recently emerged, are emerging as a key driving force for the popularization of AI by achieving equivalent performance at a cost 97% lower than existing models.

AI workloads are broadly categorized into LLM, generative AI, and HPC (high-performance computing).

All of these require enormous computing power and memory bandwidth, pushing the limits of data center infrastructure.

For example, training the GPT-4 model requires 25,000 GPUs to operate for 100 days, and existing data centers often cannot accommodate even a few hundred GPU servers. Demand for AI is doubling every year, but infrastructure isn't keeping pace.

Regarding this, CEO Lee Jung-yeon emphasized, “Data centers in the AI era require not a simple upgrade, but a completely new design.”

Currently, NVIDIA monopolizes the GPU market, which is causing problems such as high prices and rising operating costs along with improved performance.

In response, open architectures and various AI accelerators (NPUs, TPUs, FPGAs, etc.) are being actively introduced.

In particular, the NPU that Korea is leading is specialized in neural network operations and is showing strengths in power efficiency and token generation performance.

Building an open ecosystem for data centers is also an important topic.

Open standards such as the Open Compute Project (OCP) and Universal Base Board (UBB) contribute to hardware architecture integration, cost reduction, and improved energy efficiency.

Designed to allow a variety of accelerators and servers to operate on the same baseboard, it creates an environment where both large and small businesses can participate in innovation.

Advances in network technologies such as NVLink, InfiniBand, and high-speed Ethernet for ultra-high-speed data transmission are also notable.

NVIDIA's MVL 72 platform integrates 72 GPUs into a single system, eliminating network bottlenecks between existing servers and enabling ultra-fast data transfer.

UA Link, developed jointly by global companies such as AMD, Intel, Meta, and Microsoft, is an open, high-speed connection technology that is leading the way in open AI infrastructure and industrial innovation.

Memory technology is also evolving rapidly. Next-generation memories such as GDDR7 and HBM4 pursue both bandwidth and economy, and the NVIDIA Rubin GPU, scheduled for release in 2026, is expected to use both types of memory in parallel.

New memory standards such as DDR6 and MR DIMM maximize processing and power efficiency.

Energy efficiency and cooling technology in data centers are also emerging as important issues.

With servers and cooling devices accounting for 80% of total power consumption, advanced cooling technologies such as liquid cooling systems, immersion cooling, and direct liquid cooling (DLC) are being introduced.

These technologies offer cooling efficiency up to 1,000 times greater than air cooling, playing a key role in energy savings and implementing eco-friendly data centers.

Security is also a key element of data centers. AI is changing the security paradigm, requiring a multi-layered security strategy that includes hardware supply chain security, BIOS integrity, and firmware verification.

“Supply chain security is the foundation of national competitiveness,” the representative said, emphasizing the importance of hardware security based on physical trust and next-generation cryptographic technologies (PUF, QRNG, PQC).

The data center of the future requires decentralization, high efficiency, and high-density infrastructure innovation.

Central, edge, and micro data centers must be organically linked to ensure both stability and scalability, and containerized data centers and rack-based micro data centers are emerging to provide flexible scalability and mobility.

CEO Lee Jung-yeon said, “KTNF is accelerating the development of micro data centers that can be quickly applied in the field by utilizing domestically produced AI semiconductors,” and added, “We will lead innovation in future data centers in Korea with our technological prowess.”