"Not Just Nvidia"… AI Companies Compete Fiercely at Conference (Comprehensive)

Ministry of Science and ICT Hosts 'AI Global Conference' on the 25th
Companies Announce Development of High-Performance, Low-Power AI Chips
Emphasis Also Placed on the Importance of Software and Data Centers

At the 'AI Global Conference' hosted by the Ministry of Science and ICT on the 25th, various domestic and international artificial intelligence (AI) companies introduced their products and presented their visions.

Minister Yoo Sang-im of the Ministry of Science and ICT is taking a commemorative photo with key participants at the 'AI Global Conference' held on the 25th at Lotte Hotel in Jung-gu, Seoul. Photo by the Ministry of Science and ICT

Discussions were held on various topics to respond to the AI era, ranging from the 'Sovereign AI' strategy to build national AI capabilities to the emerging essential infrastructure of 'AI Data Centers.'

Jung So-young, CEO of NVIDIA Korea, who gave the first presentation, announced that AI semiconductors with improved performance would be released annually. She showcased programs supporting physical AI, robotics, and autonomous driving research, including personal supercomputers DGX Spark and DGX Station, demonstrating NVIDIA's dominance in the AI ecosystem.

Keith Strier, Senior Vice President of AMD, a competitor of NVIDIA, explained the importance of Sovereign AI through a video presentation. He said, "Many participants are running on the Sovereign AI track, and there will be many winners," adding, "It can enhance public services or export AI capabilities through eco-friendly energy." He emphasized the need to promote AI infrastructure construction as part of national policy in preparation for the 'Sovereign 2.0' era.

Part of the presentation materials from NPU development company 'Rebellion'. Provided by the Ministry of Science and ICT.

Intel, the developer of the AI semiconductor 'Gaudi,' also appeared as a presenter and expressed its desire to become a partner in the Korean government's AI projects. Moon Jong-min, Managing Director of Intel Korea, stated, "For South Korea to become a strong AI nation, it is necessary to consider the costs and environmental impacts as much as the number of GPUs." He added, "Intel operates 15 environmentally friendly and energy-efficient data centers worldwide," and "We are continuously working to convert all energy used by 2030 to eco-friendly renewable energy."

He also mentioned, "The greater the digital divide, the more socio-economic problems it can cause," and said, "Intel is training responsible AI technology experts through over 20 countries and 20,000 organizations, and aims to spread this with the Korean government, companies, and academia." He continued, "Intel's strategy is to create economical, innovative, and safe AI," and "We co-developed AI with Naver a year ago, and many results have been published through papers and other means."

Kim Kwang-jung, Head of Business at Rebellion, which is developing inference AI semiconductors NPU, said, "We leave AI training to GPUs and focus on inference services that expand that part." He added, "For AI services to run smoothly, software is as important as hardware," and "Rebellion is investing heavily in the software field as well." NVIDIA has built the AI ecosystem by developing GPUs and distributing CUDA software, creating familiar workflows for engineers.

Baek Junho, CEO of FuriosaAI

Baek Jun-ho, CEO of FuriosaAI, founded in 2017 and once rumored to be sold to Meta, stated, "Our second-generation chip is currently being sampled with clients, and we expect cases of it being installed in general servers and applied to AI services." He added, "We are developing solutions that deliver sufficient performance for services while using less energy to reduce costs."

Cho Hyung-geun, Managing Director of startup More, which is developing low-cost, high-performance AI models, emphasized, "Making thousands of GPUs into a single system for ultra-large AI is the role of software."

He introduced collaboration cases with AMD, KT, and others. Cho said, "We built the world's first AMD GPU-based AI data center," and "We attracted over 100 customers through public cloud." He continued, "An AI model trained for four months on an infrastructure with 600 AMD MI250 GPU chips, with an investment of about 1.5 to 2 billion KRW, achieved performance similar to GPT-4o on various Korean language benchmarks and was released as open source," adding, "We will release image and video generation AI models within this year."

Startup More's presentation materials. Provided by the Ministry of Science and ICT.

NHN Cloud is responsible for building and operating various IT equipment within the national AI data center located in Gwangju City. Kang Min-soo, Chief Infrastructure Officer (CIO) of NHN Cloud, said, "There are about 1,000 NVIDIA H100s here, which generate a lot of heat, so the cooling method is important." He explained about AI data centers, "We need to adopt a hybrid cooling method by adding water cooling to the existing air cooling," and "The floor must withstand a load of 2 tons per square meter, and the ceiling height must be over 7 meters."

He added, "From piping installation to fire safety facilities, AI data centers will be very different from previous data centers," and said, "We are putting a lot of thought into leading the AI field in the future based on our experience and know-how."