Despite a series of earnings surprises from major U.S. tech companies in the second quarter, Wall Street's concerns over AI investments are shifting towards warnings about 'hidden debt' not reflected on financial statements. The intertwining of special purpose vehicle (SPV) financing, credit guarantees among competitors, and cyclical investments in semiconductor firms is reigniting fears of an AI bubble.
According to the IT industry on August 5, hyperscalers like Oracle and Meta are using SPV structures to finance data center construction, effectively moving related debt off their balance sheets.
Oracle has formed SPVs with firms such as Crusoe, Blue Owl, and Vantage to build data centers, financing them through long-term leases after completion. It is estimated that Oracle has raised $66 billion through these SPVs.
Meta has also established SPVs with Blue Owl and Bayou Investors to raise $30 billion, which is not recorded as debt on its financial statements, allowing the company to raise an additional $30 billion in the corporate bond market shortly thereafter. Moody's has pointed out that the renewal and termination costs of such lease agreements may not be fully reflected under current accounting rules, suggesting that disclosures do not fully capture the actual financial risks faced by AI companies.
A new method of mutual credit guarantees among competitors has also emerged. On August 1, Google guaranteed a $15 billion loan to support Anthropic's construction of a 1.6-gigawatt data center in Texas. This structure supports competitors' capital investments using 'other people's credit' rather than its own equity or debt.
NVIDIA is reportedly discussing a similar arrangement, potentially guaranteeing up to $250 billion for OpenAI's 10-gigawatt data center project in Ohio. Analysts suggest that this cyclical structure, where NVIDIA sells GPUs to AI startups it has invested in and reinvests that revenue back into those companies, is inflating demand and prices for AI semiconductors.
This trend is not limited to U.S. tech giants. Similar patterns are emerging in South Korea.
Naver disclosed on August 3 the structure of its 'AI Factory' Phase 1 (200MW) project, which it is pursuing with NVIDIA and Brookfield. Brookfield is set to raise up to $9 billion to form an SPV that will directly purchase and own GPUs and data center equipment. Naver will receive computing resources from the SPV through a wholly-owned operating subsidiary, paying usage fees and reselling the services for profit. This approach mirrors the Oracle and Meta SPV structure, shifting initial investment and borrowing burdens off the balance sheet.
Industry analysts warn that such complex financing structures are contributing to instability in the semiconductor market, intertwined with the AI bubble. Recently, South Korea's stock market has seen significant fluctuations, particularly among major semiconductor firms like Samsung Electronics and SK Hynix, with concerns about AI-related credit risks from the U.S. being cited as a contributing factor.
A credit rating agency official stated, 'The more debt accumulates off the balance sheet, the harder it becomes to predict where the risk will transfer in the event of a crisis. A similar pattern of astronomical investments occurred during the dot-com bubble, ultimately leading to significant losses for many investors.'
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.