The bond market is sending a signal to the builders of the AI data center: we intend to raise the interest rates on your credit cards and may even impose some new restrictions on credit limits.
This metaphor comes from Christopher Wood, the head of global equity strategy at JPMorgan Chase. In a report on Friday, he wrote: “The most likely way for the arms race in capital spending to end is when credit cards are taken away.”
This has not yet happened, but the credit market has already been disturbed by the new political landscape—under Democratic governors in key states, data centers face more obstacles, and at the same time, Congress is also controlled by Democrats.
A typical example of debt pressure is neocloud Company CoreWeave. The yield on its bonds, which mature in 2032 with a coupon rate of 9.625%, has risen to over 16%. Deutsche Bank stated that the five-year credit default swap for CoreWeave soared above 1,289 basis points on Thursday, breaking through the October high. On Friday, CDS fell by 8% to around 1,155 basis points.
Jim Reid, the head of global macro and thematic research at Deutsche Bank, stated in a report on Friday: "We have received many warnings from the credit markets." He added that the correlation between bonds and stocks is now stronger, "so we still need to see whether the credit markets will deteriorate further."
Investors are demanding higher compensation to hedge against the risk of default even from companies with the strongest balance sheets and the highest credit ratings, including Nvidia, Broadcom, and Amazon.com.
Nvidia Debt-related CDS has risen by 76% in the past month. Broadcom's CDS has increased by 49%, and Amazon's CDS has increased by 44%. However, in terms of absolute levels, these CDS are still relatively low; the yield spread for Nvidia is about 139 basis points, while for Oracle it is about 298 basis points, and for CoreWeave it is about 1,100 basis points. A significant increase from a lower starting point does not necessarily mean the same level of credit risk.
Nevertheless, the increase in CDS indicates that the market desires more default protection even for the largest and most financially sound debt issuers. The rise in CDS also implies that future financing costs may be higher.
The market is now digesting the possibility that as Democratic governors take office, data center expenditures in Ohio, Texas, and Georgia—three states with substantial planned capacity—may be postponed or canceled, at least in part.
Goldman Sachs stated that the suspension order could affect 45% of the planned production capacity in the United States, with a large portion of this being in Texas. Texas' incoming governor Gina Hinojosa indicated that she is not opposed to data centers, but hopes to suspend construction until community protection measures are passed by the state legislature.
However, this does not mean that the debt financing of AI is collapsing. According to Goldman Sachs, since October 29th, there have been 5 transactions of AI bonds, with a total amount of 67 billion US dollars, bringing this year's total debt issuance by hyperscale cloud service providers to 296 billion US dollars. Goldman Sachs expects that next year, the debt issuance by hyperscale cloud service providers will reach 440 billion US dollars, in addition to which there will be 300 billion US dollars in project-level issuances, covering investment-grade and high-yield bonds.
The Trump administration, on the other hand, vows to support data centers. Kevin Hassett, the director of the National Economic Council of the United States, told CNBC on Friday that despite the Democratic Party's heavy victory in the elections, he still expects data center growth to continue as predicted.
What will happen next?
The biggest question at present is how quickly AI's expenditures need to generate sufficient cash flow to reassure hyperscale cloud service providers about their future capital expenditure plans. For neocloud company, if there are signs of integration or reduction in capital expenditures, it may lead to a decrease in data center computing power contracts or shorter contract terms, or delays in upfront payments.
Goldman Sachs analysts stated that hyperscalable cloud service providers reported negative free cash flow in the third quarter, as capital expenditures exceeded operating cash flow.
This line states: 'Delayed cash realization and increased procurement commitments – which we consider to be additional and significant uses of cash, in addition to capital expenditures – are key risks for further borrowing, depending on market conditions.'
The peak of debt issuance may be approaching. Bank of America analysts estimate that the scale of AI related credit issuance will grow by 31% from 2026 to 2027, reaching $685 billion, including both public and private debt; it is then expected to decline by 23% in 2028 to $531 billion.
As for risks, Bank of America stated in the report: 'There are downside risks to these forecasts, as the 'Blue Wave' could lead to a slowdown in AI infrastructure projects, thereby reducing the demand for capital expenditures and lowering the financing needs in the bond market.'
– CNBC's Nick Wells and Michael Bloom have contributed to this report.
