AI infrastructure enters the "calculation stage"! The market no longer only looks at capital expenditure; cash flow and investment returns have become new challenges.

According to Zhihui Finance APP, as investors increasingly question whether the tens of billions of dollars invested in data centers can translate into sustainable profits, the artificial intelligence (AI) boom may be entering a more stringent phase.

The disastrous stock performance of South Korean AI infrastructure provider Elice on its IPO day serves as the latest warning sign. This Seoul-based new cloud service company (neocloud) saw its share price plunge 32% on its first day of trading on Tuesday, despite pricing at the top end of its IPO range and raising 201 billion Korean won ($151 million). However, the stock rebounded by about 30% on Wednesday.

With borrowing costs on the rise and mounting uncertainty regarding the long-term returns on large-scale AI investments, investor sentiment toward AI infrastructure firms has shifted. In addition to Elice, several other AI infrastructure companies have recently experienced setbacks. Most notably, CoreWeave (CRWV.US) saw its stock tumble 25% last week, its worst performance since November 2025. At the same time, yields on its junk-rated bonds soared to 15%, typically seen as a distress signal, while credit default swap pricing reflects a more than 60% probability of default within five years.

Nscale is also facing investor opposition to its IPO valuation targets. According to sources, the Nvidia (NVDA.US)-backed data center developer has been seeking an IPO valuation of around $30 billion. However, some of these sources note that a group of investors recently urged the company to consider a range of $16-$19 billion, believing this would be more attractive to buyers.

Additionally, Blackstone’s data center operator AirTrunk has decided to postpone its planned Singapore real estate investment trust (REIT) listing to the first or second quarter of 2027. AirTrunk operates a hyperscale platform across Australia, Japan, Singapore, Malaysia, Hong Kong, and India. In 2024, it was acquired by Blackstone and CPP Investments for AUD 24 billion, the largest recorded data center transaction globally at the time. The REIT was originally positioned as a “robust path” to invest in AI infrastructure—assets already in stable operation, tenants are large-scale cloud computing firms, and dividend yields are robust.

Reportedly, AirTrunk considered raising its pre-offering yield target from 5.5% to about 6% to attract investors, yet was still unsuccessful. The AirTrunk REIT only offered less than 100 basis points of excess return, which many investors deemed insufficient compensation for an illiquid single-industry vehicle with currency, tenant concentration, and construction risks.

Earlier, Firmus, an Australia-based AI data center operator backed by Nvidia, had planned a listing at a valuation of around $30 billion, expected to be one of Australia’s largest IPOs. However, due to inflated valuations, insufficient investor subscription demand, and concerns over the massive capital needs of AI infrastructure, this high-profile IPO plan was ultimately cancelled on October 9.

These signals highlight an overall shift in sentiment towards AI investment deals, suggesting that investors may be scrutinizing ambitious expansion plans and profitability paths more rigorously.

Dan Ives, tech analyst at Yorkville Ives, said, “This is a moment of reality check for tech stocks. Investors are trying to more clearly distinguish winners from losers and assess if there are any weak links in the technology AI supply chain.” He added: “Investor patience is gradually wearing thin, and the intensity of scrutiny will only increase, especially for companies burdened with large amounts of debt, such as some new cloud service providers.”

With borrowing costs climbing and growing concerns about debt-driven expansion, investors may be forced to differentiate between two types of companies: those that can generate cash flow from the AI boom, and those spending heavily in anticipation of future demand.

Billy Leung of Global X ETFs said, “AI investment deals have shifted from a stage where the mere size of capex announcements would be positively received by the market, to a stage where investors are now asking which companies can really generate sufficient returns from these investments.”

He further pointed out, “The market is differentiating between two kinds of companies—one group needs capital just to sustain infrastructure buildout, and the other is able to earn revenue as development continues.” He noted that the most vulnerable are new cloud service providers whose expansion rate exceeds their cash flow capacity, especially those highly reliant on external financing.

John Blank, chief equity strategist at Zacks Investment Research, noted that CoreWeave is one of the largest new generation cloud service providers and “a typical case for the whole event.” He said, “This is pressure faced by the entire market, not just by speculative behavior.”

Those enterprises building computational power primarily through increased borrowing may face shrinking profit margins if financing costs rise faster than revenue growth. This pressure becomes particularly acute if organizations incur mounting construction and depreciation costs before their facilities reach full utilization.

This shift is especially important for Asian markets, as Asia’s tech markets have significant exposure to the semiconductor and memory chip industries—which are central to the global rush into AI investment. John Blank said, “US semiconductor and AI-related companies are already experiencing overbuilding. In Korea, the extent of overbuilding is even greater.”

South Korea’s SK Hynix and Samsung Electronics have been among the biggest beneficiaries of surging demand for advanced memory chips, but as investors reevaluate the sustainability of AI spending, both companies’ share prices have begun to come under pressure.

Billy Leung noted, “Asian markets are more directly linked to hardware and memory chips in AI investment deals. Therefore, when investors begin to question AI capex and its financing, Korea and North Asia’s tech markets will experience a disproportionate impact.” He added that profit expectations in Korea and Taiwan remain on the rise, indicating that recent selloffs reflect deteriorating sentiment and valuation compression—not a sudden collapse in demand.

HSBC CEO Georges Elhedery warned that investors and lenders must become more discerning when deciding which data center projects deserve financing. He said, “You have to examine the quality of lease agreements. You have to consider contract pricing. Not all contracts reach a high enough quality to give you sufficient confidence to back them. Some projects are likely to face greater challenges.”