Elice's Book Was 389 Times Covered. The Stock Closed Down 32% on Day One.

Two numbers came out of the same Korean IPO within a week of each other. Retail investors subscribed to Elice Inc.'s offering 408.26 times over, committing more than 10 trillion won in deposits to chase 201.1 billion won of stock. On Tuesday the shares closed at 61,400 won, 32 percent below the 90,500 won offer price.

The explanation on offer everywhere is that AI sentiment turned. It did turn. It does not account for a gap this size on a book that institutions covered 388.84 times, and it does not explain the decision that actually produced the loss, which was to price at the very top of the range with the selloff already two weeks old. The subscription multiple and the closing price were measuring two different populations, and only one of them had to put up money that stayed put.

A Korean subscription multiple counts deposits, not holders

In a Korean retail IPO allocation, investors wire a deposit against the shares they request. The deposit is refunded within days. The allocation is pro-rated, so requesting more gets you more, and because the money comes back almost immediately, the cost of requesting a very large number is close to the interest forgone for a few days.

That mechanism produces a number that looks like conviction and functions like a lottery entry. Elice's 10 trillion won in retail deposits is roughly fifty times what the company was actually raising. Nobody involved intended to buy 10 trillion won of Elice. They intended to maximise their slice of an allocation they expected to flip.

Institutional book-building is less extreme but points the same way: 2,367 institutions produced 388.84 times coverage, according to Seoul Economic Daily. A book covered 389 times does not mean 389 times the demand exists at the offer price. It means the indications were submitted at a price where allocation was free money, and indications are not commitments to hold.

So the question a syndicate has to answer is not how many multiples the book was covered. It is how many of those multiples survive contact with the opening auction. On Tuesday the answer was: not enough to hold 90,500 won, and not enough to hold 75,700 won either, which is where the stock opened, 16 percent underwater before a single session had been completed.

The company's numbers got better, not worse

The convenient version of this story has investors recoiling from a weak business. They did not, because the business is not weak on the figures disclosed.

Elice's prospectus showed first-half 2026 revenue of 34.4 billion won, up roughly 200 percent year over year, and operating profit of 10.3 billion won. That half-year operating profit is about double the 5.2 billion won Elice earned across all of 2025, against full-year 2025 revenue of 39.5 billion won. EBITDA rose more than elevenfold to 23.2 billion won. The company founded in 2015 by three KAIST doctoral students as a coding education platform had, on paper, just converted a pivot into GPU cloud and modular data centres into real operating leverage.

Douglas Kim, writing on Smartkarma, projected an 83 percent sales increase for the full year and an operating margin near 33 percent. Those are not the numbers of a company the market decided was broken between Friday and Tuesday.

Which is the point. If fundamentals improved and demand was nominally 400 times covered, the 32 percent break has to come from somewhere else. It came from the price.

The float was always going to be thin

Elice sold 2.22 million new shares into a company with roughly 11.1 million shares outstanding, implying a post-listing market capitalisation of about 1.005 trillion won at the offer price. The new issue was therefore around a fifth of the share count.

A small float does two things at once, and they pull in opposite directions during the deal and on debut. Before listing, it inflates the subscription multiple mechanically: the same won of demand chasing fewer shares produces a bigger number, and that bigger number is then reported as evidence of appetite. After listing, it removes the depth that would absorb selling. The flippers who subscribed for an allocation they never meant to keep all arrive at the opening auction at once, and there is no natural holder base to sell into, because the natural holders were diluted to a sliver of the register.

By the close, the implied market capitalisation had fallen from above 1 trillion won to roughly 700 billion. The Korea Exchange data shows the shape of the day plainly: open 75,700, a high of 77,700, 65,800 by ten in the morning, and a low and close of 61,400. There was no moment when buyers stepped in.

Pricing at the top was a decision, made in plain view

The indicative range was 70,400 to 90,500 won. The syndicate took the top.

That decision was not made in ignorance of the market. Nvidia-backed Firmus had abandoned an Australian listing less than two weeks earlier, having targeted as much as $5.5 billion at an equity valuation near $30 billion, close to triple its August private round of $10.5 billion, before citing volatility and insufficient demand. Citi's Peter Lee has been pointing to rising rates as the source of pressure on data-centre names. The IMF warned in October about weaker AI profitability and rising leverage producing sharp corrections.

All of that was known. Pricing at 90,500 won anyway is defensible only if you believe the subscription book told you something about where the stock would clear. It did not, and the gap between 90,500 and 61,400 is roughly the size of that belief.

Had the deal priced at the bottom of its range, at 70,400, Tuesday's close would have represented a 13 percent break rather than 32. The company would have raised about 45 billion won less and kept a functioning aftermarket. That trade-off is the one the subscription number talked the syndicate out of making.

The case that it is simply the sector

The strongest objection is that none of this mechanism matters, because the whole Asian AI infrastructure trade repriced in October and Elice listed into it. Firmus pulled. Singapore's DayOne is reported to be weighing a listing of up to $5 billion in November. The Kospi, where Samsung Electronics and SK Hynix together account for about half the index, has lost about a fifth of its value in the past month. On this reading, Elice was a leaf in a storm and the book-building detail is noise.

Two facts resist it.

The first is the Kosdaq itself, where Elice actually listed, fell only about one percent on Tuesday. Samsung slipped 1.6 percent. This was not an index dragging a constituent down 32 percent. The move was idiosyncratic on a day the market was close to flat.

The second is timing. The sector explanation treats the selloff as a shock that arrived between pricing and trading. It did not arrive; it was already there, visible to everyone, when the range was set at the top. A story about sentiment has to explain why the people closest to the deal read the demand signal as strong enough to justify the maximum price at precisely the moment the sector was repricing. The answer is that they were reading a number built out of refundable deposits.

The honest concession is that both things are true and they compound. A thinner market punished a pricing error that a frothier one would have absorbed. But the error was available to be punished, and it was made on the basis of a demand signal that had been structurally inflated before the selloff started.

Elice raised its money. The 201.1 billion won is in the company and earmarked for domestic data-centre build-out and expansion into the United States, Japan and Singapore, and a business tripling revenue with a 30 percent operating margin will probably put it to work. The people who lost on Tuesday were the ones allocated stock at 90,500 won, and the mechanism that set that price is still running.

The test comes in November. If DayOne lists, watch which number the syndicate anchors on. A subscription multiple in the hundreds will be available again, because the mechanics that produce it have not changed, and it will mean exactly as much as it meant this week. The figure worth reading is the one nobody publishes: how much of the book intends to own the shares on day two.

Sources

• Data-Center Firm Elice Plunges 32% in Debut After Korea IPO — Bloomberg
• Elice shares plunge 28% on debut as investor doubts over AI trade deepen — Investing.com
• Elice Plunges 30% in KOSDAQ Trading Debut — Seoul Economic Daily
• Elice Shares Tumble 32% on Kosdaq Debut Amid AI Sector Concerns
• Elice IPO Prices at Top of Range Ahead of Kosdaq Debut After Firmus Collapse
• South Korea's AI Infrastructure Firm Elice Group Plunges ~30% in KOSDAQ Debut — BigGo Finance, citing Korea Exchange
• After CoreWeave's Slide, Korea AI Cloud IPO Elice Sinks 32% — FourWeekMBA
• Elice IPO Raises $151M for AI Data Center Expansion
• Elice shares plunge 30% on KOSDAQ debut as AI valuations cool
• Elice Group shares fall in KOSDAQ debut after 201.1 billion won IPO — Traders Union, citing Reuters