AI 거품이 가장 먼저 터질 곳은 엔비디아가 아니라 잘 알려지지 않은 네오클라우드 대출기관들일 수 있다는 분석이 나왔다.

Nobody panics about a Nebius or an Applied Digital the way they panic about Nvidia. That's exactly the problem.

You've heard the big AI bubble story a dozen times: sky-high valuations at OpenAI, Nvidia's chip backlog, Microsoft's capital spending. What you haven't heard enough about is the layer underneath all of it, the neocloud firms that borrowed tens of billions of dollars to build the data centers those giants rent by the month. According to the BBC, firms like CoreWeave, Nebius, IREN, Applied Digital and the British firm Nscale are the ones actually holding the debt, and they're the first to be hit if AI spending so much as pauses.

CoreWeave is the clearest case. The company started as a crypto mining operation before pivoting to rent out Nvidia GPUs, and it now carries roughly $50 billion in debt. Its bonds due in 2031 have traded with yields surging past 16%, and credit default swaps on the company have implied a nearly 63% probability of default within five years. That's not a stock wobble. It's bond markets pricing CoreWeave the way they would price a company already in trouble, even though it still has a large backlog of contracts with AI labs.

Applied Digital presents a different kind of warning sign. The company launched a $2.35 billion senior secured notes offering to fund new data center campuses, on top of debt obligations already over $2 billion. Its interest payments are now running close to its entire annual revenue. That is the arithmetic of a company betting its whole balance sheet on AI demand arriving on schedule, with almost no room to absorb a delay.

Nscale was supposed to be Britain's neocloud success story, backed by Nvidia and heading toward a US listing that it hoped would value the company at $30 billion. Then Microsoft walked away from a major data center leasing agreement with the firm, citing power supply constraints that could persist through 2030. Investors reportedly began pricing Nscale's IPO closer to $16 billion. Half the valuation was gone because one customer pulled out of one lease.

That's the real exposure. Unlike Nvidia, which sells chips to practically every AI company on earth, or Microsoft, which can absorb a slow quarter from its cloud division without blinking, these neocloud firms often depend on a small number of massive, long-term contracts with the same handful of hyperscalers. Lose one of those contracts and the balance sheet doesn't bend. It breaks.

Nebius and IREN haven't been spared. Both stocks fell sharply alongside CoreWeave and Applied Digital through October and into November, part of a slide that left the group down nearly 30% from mid-October highs. Fears about AI infrastructure spending spread from boardroom caution into bond and equity prices. Gartner has estimated that more than $3.5 trillion in AI infrastructure spending globally could face impairment risk by 2030 if demand doesn't keep pace with construction.

Gartner's forecast for the neocloud sector specifically is blunter still: by 2030, the firm expects 80% of neocloud companies to stop operating independently. Even Elon Musk weighed in publicly, suggesting that the roughly 200 neoclouds now competing for AI workloads could face what he called guppy level death and consolidation.

AI demand isn't fake, and Nvidia and OpenAI aren't about to stumble. The household names are the best positioned to survive a slowdown because they have diversified revenue and the balance sheets to absorb a bad quarter. The neoclouds don't. They borrowed heavily, locked themselves into single-customer dependency, and bet that AI spending would keep accelerating in a straight line long enough to pay it all back.

Pension funds, insurance companies and university endowments are among the investors supplying the private credit behind much of this data center debt. Lenders have been slicing those loans into tranches and selling them on, which makes the actual holders of the risk harder to trace than the headline debt figures suggest. If a CoreWeave or an Applied Digital stumbles badly enough to default, the damage won't stay contained to one stock ticker. It runs through the credit markets that financed the whole buildout, and those markets don't care how famous the company's name is.