The growth narrative surrounding AI computing infrastructure is now facing a systemic challenge originating from the credit markets.
Rothschild & Co. Redburn, an independent research arm of the Rothschild group, has initiated coverage on both CoreWeave and Nebius with "Sell" ratings. The core thesis is that while stock markets continue to price in AI growth optimistically, credit markets are already beginning to price in risks that equity investors have largely overlooked. In parallel, famed investor Michael Burry, known for his "Big Short" bet against subprime mortgages, estimates that leading tech giants hold a combined total exceeding $3 trillion in off-balance-sheet commitments, casting doubt on the sustainability of their expansion capabilities.
In a report published on Monday, Rothschild & Co. Redburn argued that the earnings and valuations of new cloud companies are subject to "cyclical exaggeration". The report noted that Nebius's share price already reflects the most optimistic outcome of an ongoing repricing of computing economics, exposing it to the greatest downside risk. For CoreWeave, the firm stated that its risk-reward profile "remains insufficiently attractive". The research house also pointed out that when off-balance-sheet commitments are factored in, hyperscale cloud providers are operating with significantly higher effective financial leverage, which could constrain their ability to fund the next phase of large-scale AI infrastructure development.
These concerns did not immediately dampen market sentiment. On Monday, Nebius closed up 4.14% while CoreWeave rose 5%, with retail sentiment for both remaining firmly in "bullish" territory — a notable divergence from the cautious stance adopted by the research institution.
Rothschild's Initial Coverage: "Sell" Ratings on Both New Cloud Leaders
Rothschild & Co. Redburn's coverage universe includes several core companies in the AI computing infrastructure sector. The firm issued "Sell" ratings for both CoreWeave and Nebius, while assigning "Neutral" ratings to six companies transitioning from Bitcoin mining to AI infrastructure: Core Scientific, TeraWulf, Applied Digital, Cipher Mining, Hut 8, and IREN.
The differentiated ratings reflect the firm's assessment of varying business model risks. New cloud companies engaged directly in GPU leasing bear heavier burdens from equipment depreciation, financing costs, and contract delivery pressures. Mining companies leveraging their own power resources and data center capacity, by contrast, enjoy greater flexibility in entering the AI infrastructure space, with comparatively fewer valuation concerns. The report also noted that traditional real estate investment trusts (REITs) "deserve particular attention as yet underappreciated indirect beneficiaries".
Pricing Power Supports Share Prices, But Financing Costs Erode Profit Margins
The common thesis underpinning CoreWeave and Nebius share prices is the sustained expansion of AI training and inference demand, alongside the continued scarcity of promptly deliverable GPU clusters, power supplies, and data center capacity. Nebius announced on September 17 a price increase of approximately 20% for certain on-demand GPU rentals effective October 1 — the second hike within three months. CoreWeave has also indicated it is signing new contracts at higher rates. Nebius shares have gained roughly 167% year-to-date.
However, Rothschild's central line of questioning targets earnings quality: after GPU rental revenue covers equipment depreciation, operating expenses, and financing costs, how much return ultimately remains for shareholders? CoreWeave reported second-quarter revenue of $2.575 billion, with net interest expenses of $640 million over the same period — financing costs have become a critical variable affecting profitability. The firm believes that competition in the AI inference era will place greater emphasis on contract quality, prepayment arrangements, equipment utilization rates, and capital returns after accounting for hardware refresh cycles and financing costs.
Credit Markets Send a Signal Distinctly Different from Equities
Rothschild & Co. Redburn stated that credit markets "are sending the more important signal". The credit market encompasses corporate bonds, bank loans, private credit, data center project financing, and credit default swaps (CDS), among other instruments. While equity investors focus on AI growth potential, creditors are more concerned with whether future cash flows can cover interest and principal. Rising credit spreads and CDS costs typically indicate that investors are demanding higher risk compensation.
As a concrete example, the firm cited the $18 billion loan associated with Oracle's Project Jupiter data center, which is being quoted at approximately 89 to 91 cents per dollar of face value — reflecting investors' cautious stance on project execution and financing risks. Following this logic, hyperscale cloud providers effectively carry higher leverage once off-balance-sheet commitments are included. Should their expansion capacity become constrained by tighter financing conditions, the pressure would transmit to CoreWeave and Nebius's growth and valuations through slower new order flow, weakened renewal pricing power, and counterparty credit risk.
Burry Estimates Over $3 Trillion in AI Commitments, Exposing a Mismatch Between Capital Recovery and Hardware Refresh
Rothschild's credit market warning echoes the latest assessment from Michael Burry, known for the "Big Short" trade. According to reports, Burry estimates that Amazon, Meta, Alphabet, Microsoft, and Oracle together hold nearly $1.2 trillion in lease commitments not yet begun, over $1.5 trillion in purchase commitments, plus exposures from special purpose entities, guarantees, and contingent support — bringing the total above $3 trillion.
Burry has made clear he is not alleging fraud. His central concern is that long-term leases and procurement expenditures are already locked in, while the pace of change in equipment economic value, technological requirements, and future demand may be faster — creating a structural mismatch between capital recovery and hardware refresh cycles. It is worth noting that this $3 trillion figure is an aggregated estimate across multiple types of multi-period commitments and cannot be directly equated with realized losses. Its investment implication lies primarily in the need to examine future cash expenditure obligations disclosed in financial statement footnotes when assessing the expansion capacity of AI data center giants.
For investors, Rothschild's report does not represent order cancellations or downward guidance revisions by the companies involved. However, it may prompt some institutional investors to evaluate order growth in conjunction with capital recovery capacity, and to re-examine the differences in return structures between compute operators, data center landlords, and power resource holders.
Comments