Bond issuance by five major hyperscalers surges

CoreWeave bonds fall below 80% of face value

Circular financing among AI firms tops 50%

Bank of Korea warns shock transmission may be stronger than in past cycles

Google's headquarters in Mountain View, California. [Getty Images]
Google's headquarters in Mountain View, California. [Getty Images]

Concerns are mounting over financial risks stemming from the AI sector as global AI companies ramp up debt-funded investment, according to a new analysis. The five major US hyperscalers — Alphabet, Microsoft, Amazon, Meta and Oracle — issued $230 billion in corporate bonds in the first nine months of this year, more than double their total for all of last year. Circular financing, in which AI companies invest in one another and then purchase each other's products, is estimated to account for more than 50 percent of their funding activity.

Kim Yun-gyeong, a senior official at the Bank of Korea's Research Department, and colleagues released the findings Tuesday in a report titled "Key Characteristics and Risk Assessment of AI Industry Financing," warning that "voices calling for caution over overinvestment, rapid leverage expansion and the side effects of circular financing have been growing of late."

The report found that hyperscalers have rapidly expanded external fundraising, led by long-term corporate bond issuance. The $230 billion in bonds the five companies issued in the first nine months of this year represents more than double their full-year issuance last year — equivalent to 33 percent of US government bond issuance (maturities of two years or more) over the same period. Their share of the outstanding US investment-grade corporate bond market more than doubled in a year, rising from 2 percent to 4.8 percent.

Markets expect the companies' capital expenditure to outpace operating profit next year, as investment costs have been rising faster than revenue.

Neo-cloud companies — AI-specialized cloud providers with relatively lower credit ratings — have raised funds through GPU-backed credit facilities and convertible bonds. As of the second quarter, the ratio of total debt to EBITDA stood at 0.6 to 5.6 times for hyperscalers, compared with 10.6 times for CoreWeave and 121.5 times for Nebius, reflecting their far heavier reliance on debt.

Market wariness is also rising. The price of CoreWeave's bonds maturing in 2032 fell below 80 percent of face value on Oct. 14, while the credit default swap premium jumped 2.7 percentage points from the end of September.

Off-balance-sheet financing — which does not readily appear in financial statements — is also on the rise. Such arrangements involve keeping debt off the books through special purpose vehicles, joint ventures, or uncommenced leases and purchase commitments.

The outstanding balance of joint-venture project financing bonds issued by hyperscalers is projected to grow from more than $108 billion this year to $300 billion next year. CoreWeave's uncommenced lease obligations stood at $35.5 billion at the end of the second quarter, exceeding Nvidia's $25 billion.

Circular transactions among AI companies are also expanding. Hyperscalers and AI chip manufacturers invest in AI labs, which in turn purchase cloud services and chips from those same companies. The Bank for International Settlements estimates that circular financing accounts for more than 50 percent of such activity. Circular financing with major AI labs such as OpenAI and Anthropic is also estimated to make up 40 percent of hyperscalers' outstanding cloud order backlogs.

The report warned that under this structure, the AI industry's expected returns may be overstated while risks are underestimated, as the ultimate location of risk becomes difficult to identify.

The report noted, however, that the current situation differs from the dot-com bubble. During that era, actual utilization of fiber-optic cable capacity was only 2 to 5 percent, whereas the vacancy rate at major North American data centers stood at just 1.4 percent in the first half of this year, with more than 80 percent of capacity under construction already leased — suggesting that future revenue is relatively predictable.

"While the sound financial position of large leading companies means the likelihood of AI-driven financial risks materializing is not high, vulnerabilities are gradually accumulating amid rising long-term interest rates," the report said. "Particular attention should be paid to the potential risks and spillover effects of neo-cloud companies and others with limited financial capacity and low information transparency."

The report also emphasized that South Korea, as one of the biggest beneficiaries of the AI boom, faces heightened exposure. "Due to South Korea's deep integration with global supply chains, the transmission of real and financial shocks from the AI sector could be amplified beyond what was seen in past episodes," it said, urging continuous monitoring of accumulating vulnerabilities and the potential for risks to materialize.


kimstar@heraldcorp.com