Over 100MW AIDC Built, GW-Scale Expansion
Starting GPUaaS with US Lambda Next Month
"To Propel Korea AI to G3 Level"
"We will take the lead in building an artificial intelligence (AI) infrastructure super highway (ultra-high-speed communication network) to enable South Korea to leap forward as an AI G3 (top 3 countries in AI)."
Yu Sang-yong, CEO of SK Telecom, stated at the SK AI Summit 2024 held at COEX in Seoul on the 4th, "Just as Korea rose to the ranks of an ICT powerhouse based on world-class information and communication technology (ICT) infrastructure, the success formula starting from infrastructure can also be applied in the AI era." Yu serves as the chairman of SK ICT, and the event is held over two days, from the 4th to the 5th.
SKT is accelerating the construction of AI infrastructure centered on ▲AI data centers (DC) ▲GPU cloud services (GPUaaS) ▲edge AI. First, it will start hyperscale AI data centers using more than 100 MW of power at domestic regional hubs. In the long term, the goal is to expand the scale to the GW level or higher and become an AI data center hub in the Asia-Pacific region. SKT explained that by combining SK’s high-efficiency next-generation semiconductors, immersion cooling energy solutions, and AI cluster operation capabilities, it can build economical and efficient AI data centers.
Next month, SKT plans to open an AIDC testbed in Pangyo. This site will introduce NVIDIA’s latest chips, SK Hynix’s high-bandwidth memory (HBM), advanced AI semiconductors, next-generation liquid cooling solutions, GPU virtualization solutions, and AI energy optimization technologies.
SKT expects that through GW-level AI data centers, it will attract new investments exceeding 50 trillion KRW, create over 550,000 jobs, and generate economic effects exceeding 175 trillion KRW. Through this, it plans to contribute to fostering advanced AI industries in the region. Additionally, it will convert the Gasan data center located in the metropolitan area into an AIDC and launch GPUaaS services that provide GPUs in a cloud form. Collaborating with the U.S. GPUaaS company Lambda, SKT will start services based on NVIDIA’s AI chip H100 from next month and introduce the H200 chip domestically for the first time in March next year, gradually expanding the volume. This will support companies in developing AI services at low cost without directly purchasing GPUs.
Starting next year, SKT will invest 100 billion KRW to build a ‘Korean-style sovereign AI’ by combining partner companies’ AIDC solutions such as Rebellion’s neural processing units (NPU) and SK Hynix’s HBM. SKT will establish the largest NPU farm in Korea, focusing on spreading the AI technology ecosystem not only in the public sector but also in the private sector.
Edge AI will be introduced using communication infrastructure. This technology bridges the gap between AIDC and on-device AI, offering advantages in terms of low latency, security, and privacy protection compared to AIDC, while enabling large-scale AI computations compared to on-device AI.
SKT is working with global partners to build AIDC using communication infrastructure and develop customized servers. It is also conducting proof-of-concept (PoC) projects with domestic companies to discover AI-specialized services in six areas including healthcare, AI robots, and AI CCTV.
Subsequently, SKT plans to collaborate with domestic and international AI companies to develop AIDC total solutions that combine AI semiconductors, data centers, and energy solutions, and enter the global market.
CEO Yu emphasized, "Until now, communication infrastructure was a battle of speed and capacity, but now the paradigm of network evolution must change," adding, "In the future, 6G will evolve into next-generation AI infrastructure that integrates communication and AI."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
