Elice IPO Sinks 32% in Kosdaq Debut Amid AI Data Centre Doubts

A record debut, for the wrong reasons

South Korean AI cloud provider Elice Group had the largest IPO on the Kosdaq this year. Its first day of trading went badly anyway. The shares were priced at 90,500 won and closed on Tuesday, 20 October, at 61,500 won, down 32.04%.3 They fell further during the session. Bloomberg reported an intraday drop of as much as 38%.4 Other trading data put the low at 56,500 won, about 37.6% below the offer price.19

If the closing loss stands, it will be the worst first day on record for a Korean IPO that raised at least $100 million. The previous worst was W-Scope Chungju Plant, which fell 30% in 2022.15 Bloomberg's data reaches a similar conclusion, describing the debut as the worst ever for a sizable Korean listing.4

Early coverage reported anything from a 28% to a 33% decline.6211 The spread reflects different moments in a volatile session, not real disagreement. The numbers that matter are the 32% close and a low of nearly 38%.

The demand that didn't last

The fall stands out because the book-building had gone so well. Elice set its price at the top of its 70,400–90,500 won range.6 Institutional orders came in at 388.84 times the shares available, from 2,367 domestic and foreign institutions.10 Retail investors subscribed 408.26 times their allocation.6 About 10.3 trillion won was placed in subscription deposits.8 The company sold 2.22 million new shares and raised about 201.1 billion won, or roughly $151 million.12

At the offer price, Elice was valued at about 1.005 trillion won.17 By the first close, that had fallen to around 680–684 billion won.19 Reports differ on the dollar figure. One converted the trillion-won valuation to about $700 million.10 Another put the market cap at roughly $1.3 billion.7 The won amounts are consistent across reports, and at current exchange rates the lower dollar range fits better. The $1.3 billion figure looks like an outlier.

The gap between huge oversubscription and a collapse after listing is the main lesson here. Korean IPO subscriptions are known for inflated order books. Investors bid far more than they expect to get, because allocations are scaled back. A ratio of 400 to 1 shows how many people wanted an allocation. It does not show demand at 90,500 won once anyone can sell. Seoul analysts quoted in the coverage made a similar point. They said the habit of expecting new listings to double in price is ending, and that oversubscription is no longer turning into gains after listing. One investment adviser said retail buyers may have assumed IPO stocks only go up.15

Timing: priced before the sell-off

The calendar worked against Elice. The IPO was priced on 6 October, before AI-related stocks sold off.2 By the listing date, sentiment toward neoclouds had soured. Neoclouds are companies that rent GPU capacity to AI developers.

CoreWeave, the US neocloud Elice is most often compared with, was at the centre of that shift. Korean reporting said CoreWeave had funded most of its GPU purchases and data centre construction with debt. As its bond yields rose, its shares fell more than 25%.3 Its stock has more than halved since its May peak.8 HSBC wrote in a note that CoreWeave's problems are not going away.7 A Gartner report predicted that 80% of neoclouds could stop operating independently by 2030. The firm said AI revenue will not be enough to give profitable returns on all planned capital spending.7

The wider IPO pipeline shows the same caution. This month, Nvidia-backed Firmus shelved its Australian listing. Blackstone-backed AirTrunk postponed a Singapore REIT offering. SoftBank's SB Energy unit had filed for a US IPO but also stepped back.2 Nscale has reportedly faced pushback on its target valuation.8 Ke Yan of Shenton Research said investors have stopped asking whether to build and started asking who is paying for the capacity, and for how long. He expects new listings across the region to carry a risk premium until AI labs show they can pay for their compute commitments.2

The weakness also worked in the other direction. Elice's debut weighed on US data centre stocks during Tuesday trading. CoreWeave dropped 4% soon after the open before recovering, and Iren fell about 4%.7 Elice is a small company, but markets read it as a test of sentiment toward AI infrastructure.7

Was the valuation the problem?

AI sentiment is the obvious explanation, but it is not the full picture. Elice's own numbers raise valuation questions. At the IPO price, the company was valued at about 25.5 times its 2025 sales of 39.5 billion won. It was valued at 14.6 times an annualised version of its first-half 2026 revenue.1910 Korea JoongAng Daily reported that people in the industry are openly asking whether the company was priced too aggressively. One startup executive said institutional investors' expectations had been too high.3

The growth figures are strong. First-half revenue roughly tripled to 34.4 billion won. Operating profit was 10.3 billion won, compared with a loss a year earlier. EBITDA rose more than elevenfold to 23.2 billion won.8 Analyst Douglas Kim forecast an 83% sales increase for 2026 and an operating margin of about 33%.15 Unlike many neoclouds, Elice is already profitable.

The operating base is still small, though. The company runs a single 10-megawatt data centre near Seoul. It has secured 130 megawatts of power capacity, most of which has not been developed.17 In the first half it reported 97.4 billion won in total capital and 35.5 billion won in short-term borrowings.3 Analysts have also pointed to how concentrated its revenue is. Coverage of the debut raised questions about its reliance on NIPA, a state agency, as a customer.19

The market's message, then, is not that Elice is a bad business. A trillion-won valuation assumed the expansion would be carried out flawlessly, at a time when investors are discounting exactly that kind of plan.

What the money is for

The company's plans for the proceeds are the bet investors are now judging. Elice says it will spend most of the money on portable modular data centres and related infrastructure.2 It plans regional edge facilities outside the Seoul area, where power connections are easier to get. It will expand abroad through subsidiaries in the US, Japan and Singapore.12 Its branded modular system, AI PMDC, can supposedly be deployed in under three months, against 24 to 36 months for a conventional build, at up to 40% lower initial cost.12 Those claims come from the company and have not been independently benchmarked.12

Chief executive Kim Jae-won said the company is going all-in on becoming a full-stack AI infrastructure leader globally. He described the IPO as a way to get the financial firepower to do that.8 Elice has the cash regardless of the share price. But a weak debut could make future fundraising or talks with overseas partners harder.

A day-two rebound, and why it doesn't change much

On Wednesday, the shares opened at 69,500 won. Within about ten minutes they reached the daily limit of 79,900 won and stayed there until the close.3 That is a rise of roughly 30% in one session. It still leaves the stock about 12% below the offer price. The wider Korean market also recovered: the benchmark index rose more than 3% as worries about AI spending eased.3

Some commentators argue the bounce shows a reshuffling of positions, not a final judgement on AI demand. On the first day, retail investors absorbed the selling, while foreign and institutional investors sold. The next day, the roles reversed.9 Others argue for patience. Chung-Ang University professor Lee Jae-sung recalled that predictions of falling semiconductor demand after the "DeepSeek shock" turned out to be wrong.3 Kim Do-joon of Zian Investment Management said Elice's finances are quite different from CoreWeave's heavily debt-funded model.8

Those points are reasonable, but they don't change the main conclusion. The question for Elice is no longer whether investors will back AI infrastructure. It is whether the price they paid made sense. Investors are no longer pricing AI infrastructure on growth alone. They want evidence of cash flow, profitability and the ability to earn back capital spending.

For Elice, that comes down to two things it can be measured on: how full its data centres are and whether it wins overseas orders. For the region's IPO pipeline, Elice now serves as a warning. A huge order book can disappear on the first day of trading. Neocloud listings priced before the October sell-off will probably have to accept lower valuations or wait.