"Renting AI R&D Infrastructure" KT Launches 'Hyperscale AI Service'

[Asia Economy Reporter Seulgina Cho]A unified solution that allows renting artificial intelligence (AI) research and development (R&D) infrastructure has emerged.

KT announced on the 9th that it has launched the ‘KT Hyperscale AI Service,’ which supports AI research and commercialization infrastructure environments and operational consulting.

The KT Hyperscale AI Service consists of AI computing power, developer toolkits, optimization and consulting by expert groups, and data centers. Its biggest feature is that it is based on IPUs optimized for AI model training and inference, providing safety, speed, and scalability as needed.

KT is the first in Korea to introduce Intelligent Processing Units (IPUs), which offer high-performance AI computing services at a lower cost than existing infrastructure. Unlike GPUs designed for graphics processing, Graphcore’s IPUs are AI-dedicated processor chips optimized for parallel processing.

According to research from Purdue University in the United States, compared to existing AI infrastructure, IPUs showed 2 to 16 times improvement in computational efficiency and memory scalability. IPUs are over 60% cheaper than GPU-based configurations with the same setup and offer energy efficiency levels of 40 to 80%, which is expected to minimize energy consumption amid increasing AI R&D demand in the future.

The KT Hyperscale AI Service supports all currently used standard machine learning development frameworks and provides all related libraries free of charge. Additionally, in collaboration with Megazone Cloud, Korea’s number one managed service provider (MSP), it offers consulting services to optimize existing AI modeling for IPUs or improve AI models for faster inference.

Furthermore, the company emphasized that based on KT’s data centers optimized for high-density and large-capacity AI infrastructure, the efficiency of AI research and development can be maximized. To this end, KT has partnered with Megazone Cloud to establish a ‘Hyperscale AI Zone’ in KT’s Internet Data Centers (IDCs). KT’s IDCs play a role in securely protecting data with industry-standard certifications, encryption, and strict access control. KT, which started its IDC business in 1999, currently operates 14 IDCs nationwide, making it the largest operator in Korea.

Lee Mi-hee, Senior Vice President of KT Cloud & DX Business Division, said, “We hope the KT Hyperscale AI Service will be an opportunity for companies and institutions interested in AI but hesitant due to high infrastructure costs to jump into AI.” She added, “Based on Korea’s top-level backbone network and IDCs, we will continue to introduce solutions that lead the development of next-generation core technologies such as AI and blockchain.”

This content was produced with the assistance of AI translation services.