Korean AI firm Elice falls 32% on debut despite profits and 400x retail demand

Elice, a Korean AI service company, listed with an IPO price of 68,000 won. Its first-day close was 46,300 won, a drop of about 32%. The hosts also converted the figures, noting the stock opened at 90,500 won and closed at 61,500 won in their terms.

The company raised about $151 million at the top of its offering range, and retail subscriptions were more than 400 times oversubscribed. First-half revenue rose about 200%, and operating profit was positive. Citing Bloomberg data, the hosts said the decline was the worst opening-day performance on record for a sizable Korean IPO.

The hosts pointed out that investors were not buying. One guest said the episode reflected the current market mood more than the specifics of this company. He argued the drop came from insiders and early holders selling rather than from investor buying, describing a sell-off mindset of 'this looks worse, we want out.'

The IPO was priced on October 6, and the discussion noted that the AI-related mood shifted afterward. One guest stressed that strong subscription demand does not translate into actual buying after listing. Another explained that the offering price was not adjusted down because doing so would have spooked the stock.

The guests cited Fermi withdrawing its IPO and Air Trunk delaying its own. They judged that Elice's drop shows listing now is risky. The host said this is not a good time to take a neocloud public and that investors cannot be expected to step in naturally.

Reports said some large investors and funds would not buy without a steep discount. Deutsche Bank, Bank of America and Goldman were mentioned as involved in the process. US neocloud stocks were also reported down that morning.

HSBC cut its rating on CoreWeave to 'Reduce' while keeping its target price at $39. The host noted that sell ratings are rare on Wall Street. HSBC sees structural problems across the neocloud business and cites the scale of hyperscalers as a key reason.

The report puts 2027 annual revenue per megawatt at about $12 million for neoclouds, versus $17 million to $18 million for hyperscalers. It argues hyperscalers have more profitable short-term contracts, a general-purpose cloud business, and many smaller customers using the full technology stack. Neoclouds, by contrast, depend on large customers that mainly buy commoditized bare-metal instances.

The host noted that bulls such as Adam Johnson are favorable toward CoreWeave, expecting AI demand to surge. They view the current volatility as a short-term phenomenon.

The host said the debt whose yields spiked is not debt CoreWeave currently holds, so the rate jump, though a bad sign, is not really CoreWeave's core problem. A guest countered that if the company had to issue new debt, that would become a real problem.

The bull case holds that CoreWeave will earn many times what it has to pay out. The argument adds that its long-term leases are backed by strong contracts and pose no problem as long as demand does not slow.

Elice is not a neocloud buried under debt, the hosts stressed. The company posted an operating profit, yet its shares still fell about 30% on the first day of trading. That led to the view that the sell-off reflected market sentiment rather than the company's earnings.

The debate also turned to the debt. The hosts noted that the debt whose yields spiked is not debt CoreWeave currently holds, so the rate jump is a bad signal but not its core problem. One guest added that if CoreWeave had to issue new debt, that would become a real problem. Bulls such as Adam Johnson argue the company will out-earn its payments by a wide margin, backed by long-term contracts that hold up as long as demand does not slow.

A report says internal Amazon emails show the company is ordering and stockpiling chips before the data centers to house them are ready. The host described this as a departure from past practice. Previously Amazon ordered chips without taking delivery, which deferred payment, but tight chip supply and fast-moving buildouts have changed that approach, so the spending now sits on the balance sheet as an asset.

Because many of these chips wait on power connections, the host explained that costs pile up on the books while no revenue has yet been recognized.

AI skeptic Ed Zitron read the situation as Nvidia pushing chips into Amazon's warehouses, suggesting Amazon may not receive chips at all unless it buys them. The host said it is not certain this reading is exactly right, but that it deserves serious consideration.

From a CFO's perspective, the company must carry unrecognized expenses on its financial statements while remaining highly uncertain when those costs will convert into revenue, given the power constraints.

According to BI reporting, at a September internal Amazon meeting AWS CFO John Felton joked to the infrastructure chief that he had pushed the company into negative free cash flow. The infrastructure chief acknowledged that spending on AWS, infrastructure, cloud and chips is very expensive and keeps free cash flow negative, but replied that this is the same kind of bet the company made with cloud computing and its delivery systems.

Amazon is reportedly expected to spend about $220 billion in capital expenditures this year, the most among the hyperscalers. The host noted that when the delivery network and AWS were being built, people called both crazy, yet both became major revenue sources. Still, the early AWS bet was seen as smaller in risk and cost than the current spending.