Why Did Korea's Biggest IPO of the Year Crash 32% on Day One?

Elice Group's 32% collapse on its first day of trading was less a verdict on one company than a repricing of the whole "neocloud" model, after a week in which the market suddenly asked whether renting out borrowed GPUs can actually make money. The Seoul-based AI cloud firm priced its Kosdaq IPO at 90,500 won, closed its first day at 61,500 won, and then rebounded by the daily limit of nearly 30% to 79,900 won on day two, which still leaves it about 12% below the offer price. The sharp swing in both directions suggests that traders are reacting to credit headlines out of New York at least as much as to anything in Elice's own numbers, and that is exactly why the exceptions matter before anyone draws a lesson from it.

반응형

What Actually Happened to Elice on Its Debut?

Elice listed on the Kosdaq on October 20, 2026 after raising about 201.1 billion won, the largest Kosdaq IPO of the year and, according to Reuters citing the company's filing, second in Korea in 2026 only to K Bank's March listing. Demand going in looked overwhelming: institutional bookbuilding drew 2,367 investors and a roughly 389-to-1 competition ratio, while the retail tranche was about 408 times covered with roughly 10.3 trillion won in subscription deposits. Under the rules for new listings, the first-day price can range from 60% to 400% of the offer price, so a quadrupling was theoretically possible and the floor sat at about 54,300 won. Instead the stock opened at 75,700 won, traded as low as about 58,000 won according to Bloomberg-compiled data, and closed at 61,500 won, down 32.04%. Bloomberg characterized it as the worst debut on record for a Korean IPO of $100 million or more, and calculated that retail buyers of the 25% retail allocation lost roughly $12 million on the day.

Metric	Figure
Offer price / market cap at offer	90,500 won / about 1.0 trillion won
Day 1 close (Oct 20)	61,500 won (-32.04%), market cap about 684 billion won
Day 2 close (Oct 21)	79,900 won (+29.92%), still about 12% below offer
Revenue / operating profit, 2025	39.5 billion won / 5.2 billion won
Revenue / operating profit, H1 2026	34.4 billion won / 10.3 billion won
Public-sector (NIPA-related) share of H1 2026 revenue	about 65.6%

Why Did a Hot Deal Fall Apart in Two Weeks?

The most commonly cited explanation is timing. Bookbuilding wrapped up at the end of September and the retail subscription ran October 7 and 8, so the price was effectively set before the market's mood changed. Then CoreWeave, the US neocloud that Korean analysts treat as Elice's closest comparable, hit a credit wall. On October 14 its shares fell roughly 15% to 16% as its 9.625% notes due 2032 traded at yields of about 15%, according to market reports based on CNBC, and S&P rates the company B+, below investment grade. CEO Mike Intrator called the reaction "irrational" in an internal memo obtained by CNBC, arguing that contracted backlog covers the company's debt, but the damage spread: Oracle fell nearly 9%, Nebius about 10% and Dell about 11% that day. Other neocloud setbacks added to the unease, including Microsoft's reported cancellation of a data center deal with Nscale and Firmus shelving its planned ASX listing. Local reporting adds technical factors specific to Elice's debut: roughly 69% of institutional allocations carried no lock-up commitment, so allocated shares could be sold from the first minute, and pre-IPO investors holding about 41% of the shares become free to sell starting one month after listing.

The market is no longer paying for AI data center growth on faith; it now wants proof that leased GPUs generate cash after interest and depreciation.

Is Elice Really a Korean CoreWeave?

Only partly, and the differences cut both ways. Elice began in 2015 as an AI coding-education platform founded by three KAIST doctoral students, and its cloud arm grew out of the server management experience it gained running that business. The company says its container-based portable modular data centers, or PMDCs, can be deployed in about three months compared with three to five years for a conventional AI data center, and company materials claim initial build costs up to 50% lower; those are management claims rather than independently audited figures. On the government side, Elice was named in June 2026, alongside Naver Cloud and Samsung SDS, as an operator for a 2.08 trillion won program to secure 9,704 advanced GPUs, and it is responsible for building 2,560 Nvidia B300 units, about a quarter of the total. That public-sector backing is what separates it from CoreWeave's heavily bond-financed, hyperscaler-dependent model, and a Seoul fund manager quoted by Bloomberg argued its finances differ enough that it need not be seen as particularly risky.

The same concentration is also the risk. Roughly two-thirds of first-half revenue came from NIPA-related government projects, which means the private-market demand that would validate the commercial case is still thin. At the offer price the company was valued at about 25 times its 2025 revenue (our calculation), and even at the first-day close it was still about 17 times, which is why a stock that looks cheap against a 1 trillion won debut valuation can still look demanding against a business that earned 10.3 billion won in operating profit in six months.

⚠️ Don't Read Too Much Into One Trading Day

A 32% fall followed by a 30% jump is the signature of a thinly floated new listing, not a clean signal about fundamentals: only about 0.81% of shares were in the initial float, institutions without lock-ups could sell immediately, and the day-two rally was attributed by local reports to easing neocloud credit worries after CoreWeave recovered overnight rather than to any new company news. Elice's financials also remain young, with operating profit having fallen 37% in 2025 before surging in the first half of 2026, so it is unclear whether profitability can be sustained. Nothing here is a view on whether the stock will go up or down from here.

How Do You Judge a GPU-Rental Business?

The skeptics' question is simple: if you buy GPUs with debt, rent them out, and watch them lose value quickly, does the contract revenue outrun interest and depreciation? CoreWeave shows how thin the margin for error can be, with its second-quarter interest expense reportedly large relative to adjusted operating income. For a company like Elice the same test applies at smaller scale. The balance sheet figures that matter are short-term borrowings, which rose from 3.0 billion won in 2024 to 28.0 billion won in 2025, tangible asset purchases of 63.8 billion won in 2025 versus 22.7 billion won a year earlier, and the 30.5 billion won of customer prepayments that cushion near-term cash flow.

✅ A Quick Checklist for Evaluating Neocloud Stocks

• How much of revenue comes from government or one anchor customer versus diversified private contracts?
• Are GPUs funded by high-yield debt, and how does interest expense compare with operating income?
• How fast are GPUs depreciated, and do contract lengths cover the useful life?
• When do lock-ups on pre-IPO and institutional shares expire?

What Should Investors Watch Next?

Three dates and data points are worth tracking. The first is CoreWeave's third-quarter results, expected in November, which will show whether the credit scare reflects real stress or sentiment. The second is the end of the one-month lock-up window for Elice's pre-IPO holders, which could add supply regardless of how the business performs. The third is whether Elice can show private-sector contracts building alongside its government work. Koreanalysis has previously covered how AI infrastructure spending is stressing balance sheets across the sector, and a closer look at how Korean listed companies exposed to GPU demand are financing capacity would be a natural next piece in this coverage area.

Frequently Asked Questions

Q1. Why did Elice fall 32% on its Kosdaq debut?

A. Industry commentary points to a mix of factors: CoreWeave's credit scare hit AI infrastructure sentiment between pricing and listing, institutional allocations without lock-ups could be sold immediately, and the offer valuation of about 1 trillion won looked demanding against the company's earnings base.

Q2. What is a neocloud?

A. A neocloud is a cloud provider focused on renting out GPU computing capacity for AI workloads, rather than offering the broad service menu of the large hyperscalers. CoreWeave is the best-known example, and Elice is often described as Korea's closest equivalent.

Q3. Does the day-two rebound mean the selloff was a mistake?

A. Not necessarily. The 29.92% rebound left the stock about 12% below its offer price, and it was attributed locally to easing neocloud credit fears rather than new company news. A single pair of trading days on a small float says little about long-term value.

The market is no longer paying for AI data center growth on faith; it now wants proof that leased GPUs generate cash after interest and depreciation.