The data center bond market is experiencing a widespread sell-off. Gartner predicts that 80% of emerging cloud computing companies will cease independent operations by 2030; meanwhile, the Chief Investment Officer of Allianz Group warns that data center bonds are facing a moment of "survival of the fittest."

According to a report by The Financial Times on Thursday, from the special financing bonds for large data centers leased to Meta, Microsoft, and NVIDIA, to the debt issued by emerging cloud computing companies such as CoreWeave, prices have fallen into a pressured range, reflecting investors' deep concerns about the sustainability of returns on infrastructure investments in AI.

The direct triggers for this round of selling include: OpenAI postponing IPO to next year, a slowdown in revenue growth during the Anthropic quarter, and OpenAI revealing actual annualized revenue that was far lower than previous expectations last week. These factors have intensified market doubts about AI's ability to generate cash flows, leading to a coordinated decline in both stocks and bonds. NVIDIA's market value evaporated by approximately $270 billion over the past week, with CoreWeave's stock price falling by about 17% within that period, while its competitor Nebius saw a decline of 18%.

The concerns of asset management firms are spreading. Ludovic Subran, the Chief Investment Officer of Allianz Group in Germany, stated: "Data center bonds are facing a moment of 'survival of the fittest,' and the weakest credit entities will struggle to continue. The surge in credit spreads caused by power and construction delays indicates that data center debt has previously been too lenient in pricing for execution risks."

Bond prices have tumbled across the board, with some reaching distressed levels.

The scope of this round of sales is broader than previously anticipated, no longer limited to emerging cloud computing companies, but has spread to multiple data center financing projects backed by large tech giants as tenants.

The bonds issued by the special purpose vehicle Beignet Investor, which is jointly funded by Blue Owl and Meta, for the Hyperion project have fallen from above par at the beginning of July to around 92 cents; the bonds for two projects in Louisiana and Texas (the latter leased by NVIDIA), developed by Hut in August, have fallen to around 94 cents and 92 cents respectively; the bonds for the “Project Excalibur” project of Blackstone's QTS data center in Georgia (with Microsoft as the lessee) have fallen below 88 cents.

Emerging cloud computing companies are facing even more severe challenges. The bonds issued by CoreWeave have fallen below 80 cents—a level that is generally regarded by the market as a symbolic threshold for distressed debt, with yields approaching 15% on Thursday. According to a previous report by the UK's Financial Times, a loan of about $18 billion related to Oracle's Project Jupiter data center in New Mexico has also slipped into a troubled zone due to delays in obtaining power supply permits amidst local opposition; the construction bonds for Oracle's AI campus in Michigan have also dropped from face value to 87 cents.

Trillion-dollar gap: AI revenue is far from sufficient to cover capital expenditures

The underlying logic behind the shift in market sentiment lies in the vast gap between AI infrastructure investment and the expected returns.

Technology consulting firm Gartner released a report this week predicting that due to a gap of trillions of dollars between the pace of capital expenditure and the expected returns of AI, 80% of emerging cloud computing companies will “cease independent operations” before 2030.

According to Gartner, data center operators such as AI need to generate a cumulative revenue of 8.5 trillion US dollars by the end of 2030 in order to cover capital and financing costs and achieve a reasonable return; however, their forecast is that the service expenditures from the consumer and enterprise sides, which flow towards data center providers, AI, "will not exceed 3.6 trillion US dollars." Gartner warns: "If AI infrastructure investments continue on their current trajectory, there may be over 4 trillion US dollars in assets at risk of impairment by 2032."

NVIDIA CEO Jensen Huang said in an interview with Bloomberg on Thursday that he believes CoreWeave does not require financial support from NVIDIA. "I am optimistic; their cost of capital is actually decreasing," he stated, and added that the strength of CoreWeave "is improving every day."

Regulatory headwinds and circular financing, with overlapping risks that are difficult to ignore

In addition to fundamental pressures, two types of structural risks are attracting increasing attention from investors.

One of the obstacles is political resistance at the local level. The backlash against data centers has spread across various regions in the United States, sparked by concerns about water consumption, noise, and air pollution, which has led to the halting of some projects. New York State and Texas have imposed strict restrictions on new data center projects, and other states are also considering similar measures. Columbia Threadneedle The head of U.S. investment-grade credit Tom Murphy stated that the opposition to data centers "is an issue that we are closely monitoring. I wouldn't say that this has changed our investment decisions, but it certainly makes those data centers that have already been built and put into operation more valuable."

The second is the risk of revolving financing. A recent report by the Bank for International Settlements pointed out that "a considerable proportion of financing and business relationships in the AI sector are self-referential in nature"—that is, customers and suppliers finance each other. Once a problem arises in one link of this structural feature, it may trigger a chain reaction, making it more difficult to predict the spread of risks.

Market differentiation: Operational assets are relatively resistant to declines

Despite the overall pressure, there are still signs of differentiation within the market.

The stance indicated by Tom Murphy suggests that data center assets that have already been built and put into operation may, in the current adverse environment, gain a relative premium due to reduced supply. This is in line with the logic reported by Gartner – new projects face triple pressures of licensing delays, rising financing costs, and uncertain revenue, while the scarcity of existing high-quality assets will become even more prominent.

For bond investors, the most critical question at present is: as the yield on corporate bonds such as CoreWeave approaches 15%, has the pricing of distressed assets fully reflected the execution risks and industry cycle risks? The answer to this question may gradually emerge in the revenue data from the demand side of OpenAI, Anthropic, and other AI over the coming months.