Elice IPO Sinks 32% on Kosdaq Debut as AI Data Centre Fears Grow

A record-breaking flop in Seoul

One of South Korea's most heavily oversubscribed share sales of the year has turned into a warning about the AI infrastructure trade. Elice, a Seoul-based neocloud provider, started trading on the Kosdaq on Tuesday 20 October. Its shares fell as much as 38% during the session and finished the day 32% down, according to Bloomberg1. Bloomberg's data shows this was the worst first day on record for a sizable Korean IPO1.

The company sold its shares at 90,500 won each, which gave it a market value of more than 1 trillion won after listing. On the first day the stock dropped as low as 58,000 won before recovering slightly to close at 61,500 won. That move cut the company's implied value to about 700 billion won.

Outlets reported different headline figures. The Telegraph led with the 38% intraday low and gave a closing price of 63,600 won3. CNBC reported the shares "last trading" at 63,600 won and described the fall as roughly 30%15. The Japan Times, writing earlier in the session, cited a 33% drop5. One trade publication wrongly described 61,500 won as the offer price6. Taken together, Bloomberg's account seems the most reliable: a low around 58,000 won, a close at 61,500 won, and a 32% loss on the day.

Huge demand, then a collapse

The drop stands out because of how strong demand was beforehand. Elice raised 201 billion won, about $151 million, at the top of its marketed range, and the retail tranche was about 400 times oversubscribed1. A company filing showed institutional orders worth 388.84 times the shares on offer15. Korean retail investors put roughly 10.3 trillion won into subscription deposits hoping to get an allocation.

The size of the deal adds to the damage. Elice's offering was the largest on the Kosdaq this year, and in Korea only K Bank's $337 million listing in March was bigger. Mirae Asset Securities and Samsung Securities were the lead underwriters. Seoul's IPO market was already weak before this. Bloomberg data shows first-time listings there have raised $1.4 billion in 2026, compared with more than $3 billion in each of the previous three years. A strong debut could have helped revive that market. This one is more likely to make it harder.

The reaction among retail investors was harsh. Bloomberg described posts on Toss Securities' app showing an Alice-in-Wonderland character smashing ants with a hammer. "Ants" is local slang for small retail investors, and the play on the company's name was deliberate. One investor said it was the first time they had ever lost money on an IPO. Jo Hyunik of Iroom Investment Advisors suggested some retail buyers had wrongly assumed IPO stocks always go up.

Profitable growth was not enough

The main lesson from the Elice debut is that solid financials no longer protect an AI infrastructure listing. The prospectus showed first-half revenue up 200% to 34.4 billion won, a move into operating profit of 10.3 billion won, and EBITDA up more than elevenfold to 23.2 billion won. Smartkarma analyst Douglas Kim forecast an 83% jump in 2026 sales and an operating margin of about 33%. Those numbers do not look like a cash-burning company sold on hope.

The business model is what worries investors. Elice buys high-performance GPUs and liquid-cooled server racks, puts them in modular data centres, and rents the compute and networking capacity to AI customers. It also sells AI-based education and business software15. The company plans to spend most of the IPO money on portable modular data centres and related infrastructure15. Chief executive Jae Won Kim said the company is "going all-in" on becoming a full-stack AI infrastructure leader globally, and that the listing provides the funds to do it.

Commentators disagree about how much risk that carries. Kim Dojoon of Zian Investment Management argued that Elice's finances are different from CoreWeave's and that investors need not see it as especially risky. He added that it is still unclear whether the current profitability will last. Kim Hak-joon of Kiwoom Securities said the bigger problem is that nobody can reliably estimate AI demand, now or in the future15.

A timing mismatch

Timing explains much of the gap between the order book and the trading floor. Analysts pointed out that Elice priced its IPO on 6 October, before the latest sell-off in AI stocks15. The offer price reflected one market and the shares opened into a much weaker one.

A lot changed in those two weeks. CNBC reported that the trouble for neoclouds began this month with disappointing results from private AI model developers. Anthropic's quarterly revenue growth reportedly slowed from 140% in the second quarter to 45% in the third. A Wolfe analyst said OpenAI's revenue came in $20 billion below expectations7. Gartner also published research suggesting that 80% of neoclouds could stop operating independently by 2030. Its chief forecaster told CNBC that compute capacity currently exceeds demand7.

Industry leaders and policymakers added to the pressure. The Telegraph reported that AI stocks fell sharply after Broadcom chief executive Hock Tan called the market "a casino" at a conference in Singapore2. The same report said the IMF warned that an AI stock crash could set off a wider financial crisis. IMF deputy director Athanasios Vamvakidis said global exposure to US AI assets is at its highest ever2. Higher borrowing costs, linked to persistent inflation, are the common factor. They raise the cost of building debt-funded capacity and make investors more doubtful it will pay off15.

Other data centre listings stall

Elice is the latest in a series of setbacks for data centre companies trying to go public, and the details vary between reports. Bloomberg reported that Australia's Firmus Grid cancelled a planned IPO over valuation concerns, AirTrunk is delaying a listing, and Nscale is facing pushback on its target valuation. Other coverage added that SoftBank unit SB Energy has delayed its plans for a US filing. The figures for Firmus differ. The Telegraph described it as a £20bn float3, while another account gave a $5 billion target. One article called SB Energy Japanese6 and another listed it as a US business. Whatever the exact numbers, the reports agree on the trend: data centre IPOs across the region are being postponed or cancelled.

CoreWeave is the reference point for most of these companies. Bloomberg noted that its shares have more than halved since their May peak and that its bonds fell sharply last week. HSBC analyst Abhishek Shukla, who has a reduce rating on CoreWeave, wrote that problems in debt markets and slowing LLM revenues are making the structural concerns about neoclouds worse7.

The selling spreads

Elice's fall did not stay within one stock. The Telegraph reported that the benchmark Kospi closed down 4.2%3, and other coverage said Samsung Electronics and SK Hynix fell 3% and 4%. The Telegraph also noted that the Kospi has lost about a third of its value since its June record3. In the US, CNBC reported that CoreWeave fell 4% shortly after the open before recovering to trade 1.6% higher, while Iren dropped about 4% and Nebius slipped about 1%7.

CNBC made the point that Elice is small. The deal raised only about $150 million and the company is worth roughly $1.3 billion7. On its own it is not significant. As a gauge of sentiment, it matters a great deal.

What comes next

The clearest summary of the episode came from Ke Yan of Shenton Research. He said the AI infrastructure trade has moved from "build at any cost" to "who is paying and for how long"15. He added that until the AI labs renting capacity show they can pay for their compute commitments, new listings in the region will probably be priced with that risk built in15.

That reading fits the evidence. Elice was not a weak company caught out. It was a growing, profitable company whose valuation was set before the market started questioning whether AI customers will keep paying. Its first-day result is a guide to how future deals will be priced. Neocloud listings will now need to show who their customers are and how long their contracts run, not just how much they are growing. Companies still preparing listings, including those that have recently delayed, should expect investors to demand lower prices.