Elice Group, an AI full-stack company, said on Wednesday it will work with Arista Networks Korea to pursue the country's first implementation of Ethernet-based large-scale GPU clustering.

The two companies signed a memorandum of understanding the previous day at Elice Group's headquarters for cooperation on next-generation AI network technology. The agreement was driven by efforts to shift large-scale AI infrastructure, which has relied on expensive dedicated communications networks, to Ethernet, a general-purpose network standard, to improve the economics of AI data centres and lower barriers to entry for South Korea's AI industry.

The companies will focus on building AI networks based on RoCEv2 technology. RoCEv2 is a technology that maximises data transmission speeds and eliminates bottlenecks on existing general-purpose Ethernet networks to support large-scale AI computing. Elice Group will first introduce a RoCEv2-based high-performance network environment to Elice Cloud and its AI portable modular data centre (PMDC). Arista Networks Korea will be responsible for high-performance network design and technical support.

Park Jeong-guk (박정국), chief technology officer at Elice Group, said, "Through cooperation with Arista, we will present an economical and efficient AI infrastructure standard and take the lead so that companies at home and abroad can achieve AI transformation more easily and quickly." Kim Se-jin (김세진), head of Arista Networks Korea, said, "Cooperation with Elice Group is an important opportunity to realise innovation in RoCEv2-based AI data centre networks," and added, "We will continue to expand technical cooperation in the Korean market."