The offering was 388 times subscribed. Then it traded.

Elice was founded in 2015 as an AI education platform and now also runs cloud services and data centres built on graphics processors¹. Its book-building drew 2,367 institutions and was 388.84 times subscribed; the retail tranche was 408.26 times subscribed and attracted more than 10 trillion won in deposits¹. It priced at 90,500 won, the top of its range¹.

At 9:30 on Tuesday morning in Seoul the stock traded at 63,000 won, 30.39% below the offering price, and touched 61,000 won¹. The market value implied at pricing, about 1 trillion won, fell to around 700 billion¹.

The wider tape. The same morning, Samsung Electronics closed 3% lower and SK Hynix 4.2% lower, and the Kospi fell 3.7% to its lowest close since 7 April². In Europe, Siltronic fell 6%, Soitec 5.7%, Aixtron 5.2%, and ASML 3%². A State Street strategist put the investor question in one line: whether “every dollar of AI capex earns more than it costs to finance”³.

The company’s own numbers. Elice reported revenue of 39.5 billion won and operating profit of 5.2 billion won last year, and 34.4 billion won of revenue with 10.3 billion won of operating profit in the first half of 2026¹. Those are real figures for a real business. They were simply worth less on Tuesday than the subscription queue had suggested.

The lesson is not that AI is a bubble or that it is not. It is that an oversubscribed order book is a count of people who wanted an allocation at a fixed price. The first trade is the first time anyone has to agree on a price with someone who is selling.