Elice's Kosdaq debut shows a 389-times covered book is no floor for AI neoclouds

Elice Group, a Seoul company that began life as a coding school and now rents out graphics processors from its own modular data centres, made its Kosdaq debut on Tuesday 20 October and closed at 61,500 won, 32.04% below its offering price of 90,500 won. The pricing had been set at the very top of a marketed range of 70,400 to 90,500 won after institutional bookbuilding came in 388.84 times covered and the retail tranche drew more than 10 trillion won in subscription deposits. On Wednesday the stock reversed, rising by the daily limit of 29.92% to 79,900 won, which still left it 11.71% under the price paid by investors in the offering.

The deal raised about 201.1bn won (roughly $151m) and was the largest initial public offering on the Kosdaq this year. In isolation it is a modest transaction. What made it a subject of attention in New York and Sydney as well as Seoul is timing: it was the first meaningful test of public appetite for an AI infrastructure listing since a run of withdrawals in the sector, and the answer, over two sessions, was a violent "maybe".

What investors were buying

Elice was founded in 2015 by three doctoral students from KAIST, the Korean science and technology institute. It started as an AI and coding education platform and has since moved into GPU cloud services, modular data centre construction and enterprise AI projects. It now reports two segments, AI cloud and AI transformation, and cloud services made up 76.2% of revenue in the first half of 2026.

The growth is real on the company's own figures. Revenue for the whole of 2025 was 39.5bn won with an operating profit of 5.2bn won. In the first six months of 2026 alone it booked 34.3bn won of revenue and 10.3bn won of operating profit, almost double the previous full year's profit. The first-half numbers supplied with the listing were unaudited. Borrowings rose from 5.1bn won in 2024 to 29.8bn won in 2025, a reminder that even a small neocloud is a capital-hungry business.

The proceeds are meant for regional edge data centres and expansion abroad, with the United States, Japan and Singapore named as overseas bases. At home the company plans to grow its Seoul-area site from 4MW to 10MW by the end of the year and to build 40MW in the Honam region. Its pitch rests on speed: it says its modular facilities can be put up in a matter of months rather than the two to three years typical of conventional builds. Those are company claims, not independently verified figures.

At the offering price the company was worth about 1.0054trn won. A Kiwoom Securities analyst, Kim Hak-jun, put the marketed range at between 16.85 and 22.64 times trailing EV/EBITDA, and cautioned that the market still needed to see customers broaden as government projects give way to private-sector AI cloud contracts.

Why the book did not hold

The first day's trading flows show who wanted out. Institutions were net sellers of 35.364bn won and foreign investors sold a net 9.046bn won, while retail buyers absorbed the difference with net purchases of 46.342bn won. In other words, much of the institutional demand that produced a 389-times covered book was demand for an allocation, not for a long-term holding. Heavy oversubscription in Korean bookbuilds often reflects investors asking for far more stock than they expect to receive; a large multiple is a measure of how scarce the paper looked, not a commitment to own it at any price.

The backdrop did the rest. Neocloud shares had been sliding for weeks on worries about debt-funded expansion, with US-listed CoreWeave more than halving from its May peak. A price range set when comparables were richer was always going to look generous by listing day.

The reversal on Wednesday was equally instructive. Foreign investors turned modest net buyers early in the session, and by 9.25am Seoul time the stock was already trading at its upper limit. Nothing about Elice's business changed between Tuesday and Wednesday. What changed was the price, and with it the arithmetic: at 61,500 won the company was valued well below the multiple that Kiwoom had described, which brought in buyers who had been unwilling to pay the offering price.

A thinner pipeline for AI infrastructure

Elice listed into a market that has become selective about the asset class. Earlier in October the Nvidia-backed Australian operator Firmus withdrew its planned ASX listing a day after its bookbuild closed, with its board saying the available terms would not appropriately reflect the strength of its business. Blackstone-backed AirTrunk has delayed a planned real estate investment trust listing, and SoftBank affiliate SB Energy has shelved a planned US flotation. Korean IPOs have raised roughly $1.4bn so far this year, compared with more than $3bn in each of the three previous years.

Albert Yong, chief investment officer at Petra Capital Management, summed up the mood: "Sentiment toward AI infrastructure has been highly volatile and often disconnected from underlying fundamentals."

Both observations apply with particular force to neoclouds. These companies borrow or raise equity to buy chips whose economic life is short, and they sell capacity to customers whose demand is hard to forecast more than a year out. Public investors are being asked to price a capital cycle, and they are doing so with wide error bars. A small company with a clean income statement, such as Elice, is not immune: in a stressed market it simply becomes the most liquid way to express a view on the whole group.

What the two days tell issuers

For companies preparing to list, the lesson is less about valuation than about construction. A book built on allocation hunting can collapse once trading starts, and a deal priced at the top of a range set weeks earlier leaves no cushion if comparables fall in the meantime. Bankers on the next AI infrastructure offerings will be under pressure to price closer to where peers trade on the day, to favour long-only anchors over fast money, and to keep the deal size proportionate to genuine demand.

Our view

Elice's debut was not a verdict on one Korean company. It was a reading of the market's tolerance for AI infrastructure risk, and that reading is unstable. A 32% fall followed by a limit-up day is what happens when investors agree a sector matters but cannot agree what it is worth.

For founders of capital-intensive AI businesses, the message is plain: public markets remain open, but only at prices that leave room for disappointment. Pricing at the top of a range to capture headline demand looks clever on the evening of the bookbuild and expensive the following morning. Private investors holding neocloud stakes at late-stage marks should assume that listing prices, when they come, will be set by the most nervous buyers in the room, not the most enthusiastic.

What to watch: whether Elice can trade back to and hold its offering price on the strength of customer wins outside government work, how the next US neocloud reports land with debt investors, and whether the operators that stepped back from listings this autumn return with smaller, cheaper deals.

Sources

• TechNode Global: Elice shares fall in KOSDAQ debut after $151M IPO
• BigGo Finance: South Korea's Elice Group plunges 32% on KOSDAQ debut despite $7.7 billion subscription frenzy
• Dealroom: Korea's biggest 2026 IPO stumbles, Elice falls 30% despite 389x-covered book
• SmartCompany: Firmus shelves $44 billion IPO plan citing market volatility

Filed under AI. Written by Invest Times Editorial under our editorial standards. Commentary and analysis for information only, not investment advice.