Kim Jae-won (김재원), CEO of Elice Group, at an IPO briefing on Sept. 30. [Photo: DigitalToday reporter Seulgi Son]

Full-stack AI company Elice Group is seeking to use an initial public offering as a springboard to target the global AI infrastructure market with a "Korean-style neo-cloud" strategy.

It plans to convert a total of 130 MW of secured power capacity across the Seoul metropolitan area, the southwest and the southeast into AI computing resources, while using its own cloud software to raise GPU utilisation and improve profitability. The share of its cloud business in total revenue rose to 76 percent in the first half of this year from 7 percent in 2023.

Kim Jae-won (김재원), CEO of Elice Group, said at an IPO press briefing in Seoul's Yeouido on Sept. 30 that the company will quickly convert domestic power capacity into AI computing resources and expand its business into the global neo-cloud market based on that foundation.

Elice Group has secured power capacity of 10 MW in the metropolitan area, 40 MW in the southwest and 80 MW in the southeast, for a total of 130 MW. It plans to avoid constraints on power supply and data-centre permits in the metropolitan area by building modular AI data centres in regions with spare electricity and deploying the latest graphics processing units to convert them into AI computing resources. It plans to further expand infrastructure by region in line with customer demand.

The core of its business model is to build data centres cheaply and quickly to bring forward the payback point for investment. The company said its modular AI data centres can typically be built in about three months, compared with 24 to 36 months for conventional data centres. It also said total cost of ownership per 1 MW can be cut by about 40 percent.

It also aims to raise GPU utilisation with its self-developed GPU virtualisation software, ECI. ECI reduces idle resources by optimising GPU virtualisation, clustering and resource allocation. Kim cited low construction costs, high GPU utilisation and faster investment recovery as strengths to compete in global markets.

Kim cited cloud and software capabilities as differentiators versus rivals. Elice Group developed its own cloud early after its 2015 founding for AI education practice, and has since expanded its technology into GPU cloud and infrastructure. "We started from computer science and our core is software," Kim said. "Depending on customer demand, we can provide a tailored offering from ECI to infrastructure that combines it with AI PMDC."

Performance is also being driven by AI cloud growth. Consolidated revenue for the first half of this year was 34.4 billion won, up 202.2 percent from a year earlier, and operating profit was 10.3 billion won. The operating margin was 30.1 percent. EBITDA was 23.2 billion won, up 11.3 times from a year earlier, and the EBITDA margin was 67.5 percent.

The company is also pursuing a business to sell AI computing resources secured in South Korea to overseas customers. Under the plan, it converts domestic power capacity into AI computing resources and allows overseas customers to use GPUs remotely, and Elice Group is holding related discussions with multiple foreign customers. The company expects related revenue could begin as early as the first half of 2027. Over the mid to long term, it also aims to target the sovereign AI market by supplying its AI PMDC itself locally to match each country's power, security and regulatory environment.

Elice Group will offer 2,222,000 shares in the IPO, all newly issued. The indicative price range is 70,400 to 90,500 won per share. The planned offering size is 156.4 billion won to 201.1 billion won, and the expected market capitalisation is 782.1 billion won to 1.01 trillion won. It aims to list on KOSDAQ in October after a public subscription on Oct. 7-8. Mirae Asset Securities is the lead underwriter and Samsung Securities is a joint underwriter.

Keyword

#Elice Group #IPO #KOSDAQ #ECI #Mirae Asset Securities
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.