Rothschild downplays AI cloud giants, warns of $3 trillion in off-balance-sheet commitments.

The narrative of a booming AI computing infrastructure is facing systemic challenges from the credit market.

Rothschild & Co. Redburn, an independent research firm under the Rothschild Group, initiated coverage of CoreWeave and Nebius with "sell" ratings for both. Their core assessment is that while the stock market is still actively pricing in AI growth, the credit market has begun to price in risks that the stock market has largely ignored.

Meanwhile, Michael Bury, known as the "Big Short," estimates that leading tech giants collectively hold more than $3 trillion in off-balance-sheet commitments, prompting scrutiny of the sustainability of their expansion capabilities.

In a report released Monday, Rothschild & Co. Redburn stated that the profitability and valuation of new cloud companies are "cyclically inflated," and that Nebius's stock price already reflects the most optimistic outcome of the ongoing revaluation of computing power economics, facing the greatest downside risk; CoreWeave's risk-reward ratio "still lacks sufficient appeal."

The agency also pointed out that after taking off-balance-sheet commitments into account, hyperscale cloud vendors have higher actual economic leverage, which may limit their ability to provide financial support for the next phase of large-scale AI infrastructure construction.

The aforementioned concerns did not immediately suppress market sentiment. On Monday, Nebius closed up 4.14% and CoreWeave closed up 5%, with retail investor sentiment on both stocks in the "bullish" zone, a clear departure from the cautious stance of research institutions.

Rothschild initiates coverage: New cloud leaders both "sold".

This coverage by Rothschild & Co. Redburn includes several key companies in the field of AI computing infrastructure.

The agency gave "sell" ratings to CoreWeave and Nebius; and "neutral" ratings to six companies that are transitioning from Bitcoin mining to AI computing infrastructure: Core Scientific, TeraWulf, Applied Digital, Cipher Mining, Hut 8, and IREN.

The agency's differentiated ratings for the two types of companies reflect its differing assessments of the risks associated with their business models.

New cloud companies that directly engage in GPU leasing bear higher equipment depreciation, financing costs, and order fulfillment pressures; while mining companies that rely on their own power resources and data center capacity to transform have a more flexible path to enter the AI infrastructure track and have relatively less valuation concerns.

Rothschild also pointed out that traditional real estate investment trusts "deserve particular attention as indirect beneficiaries that have not yet received sufficient attention."

The logic behind price increases supports the stock price, but financing costs are eroding profit margins.

The common logic supporting the stock prices of CoreWeave and Nebius is the continued expansion of demand for AI training and inference, while GPU clusters, power, and data center capacity that can be delivered in a timely manner remain scarce.

On September 17, Nebius announced a price increase of approximately 20% for some GPU on-demand rentals, effective October 1. This marks the second price hike in three months. CoreWeave also stated that it is signing new contracts at higher prices. Nebius's stock price has risen by approximately 167% year-to-date.

However, Rothschild's core question points to the quality of profitability: After covering equipment depreciation, operating expenses, and financing costs, how much return can GPU rental income ultimately leave for shareholders?

CoreWeave reported revenue of $2.575 billion in the second quarter, with net interest expenses of $640 million, highlighting financing costs as a key variable affecting profitability.

The agency believes that the new cloud competition in the AI inference era will place greater emphasis on order quality, prepayment arrangements, equipment utilization, and return on capital after covering hardware upgrades and financing costs.

The credit market is sending signals that are quite different from those of the stock market.

Rothschild & Co. Redburn stated that the credit market is "sending more important signals." The credit market encompasses instruments such as corporate bonds, bank loans, private lending, data center project financing, and credit default swaps (CDS).

Equity investors are focused on the growth potential of AI, while creditors are more concerned with whether a company's future cash flow can cover interest and principal payments. Rising credit spreads and CDS costs typically indicate that investors are demanding higher risk compensation.

The agency cited the example of the $18 billion loan related to Oracle's Project Jupiter data center, with a quote of approximately 89 to 91 cents per dollar face value, reflecting investors' cautious attitude towards project progress and financing risks.

According to the agency's logical deduction, after hyperscale cloud vendors include off-balance-sheet commitments, their actual economic leverage is higher. Once their expansion capabilities are constrained by tighter financing conditions, the pressure will be transmitted to the growth and valuation of CoreWeave and Nebius through slower new orders, renewal negotiations, and counterparty credit risk.

Bury estimates AI companies have committed over $3 trillion, revealing a mismatch between capital recovery and hardware upgrades.

Rothschild's warning about the credit market echoes the latest assessment by Michael Bury, known as the "Big Short."

According to reports, Burry estimates that Amazon, Meta, Alphabet, Microsoft, and Oracle collectively hold nearly $1.2 trillion in uncommitted lease commitments and over $1.5 trillion in procurement commitments. Adding exposures to special purpose entities, guarantees, and contingent support, the total exceeds $3 trillion.

Burry made it clear that he did not accuse anyone of fraud. His core concern was that long-term leases and procurement expenditures were already locked in, while the economic value of equipment, technical requirements, and future demand could change much faster, leaving companies facing a structural mismatch between capital recovery and hardware upgrades.

It is worth noting that this $3 trillion is an aggregate estimate of various intertemporal commitments and cannot be directly equated with incurred losses; its investment implications mainly lie in assessing the expansion capabilities of AI data center giants, which requires examining future cash outflow obligations in the financial statement footnotes.

For investors, Rothschild's report does not represent order cancellations or a reduction in the company's operating guidance, but it may prompt some institutional investors to examine order growth in conjunction with cash recovery capabilities and to re-compare the differences in return structures among computing power operators, data center lessors, and power resource holders.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.