Another AI bubble theory… $120 billion of investment not recorded on big tech balance sheets
공유하기
- US big tech companies are raising funds for AI data center infrastructure through special purpose vehicles (SPVs), and related debt is not being reflected on company financial statements.
- This financial structure is advantageous for defending credit ratings, but in practice increases reliance on the private credit loan market and could spread lending risk across the financial markets.
- The FT warned that if AI demand falls short of expectations, it is unclear who would bear the related debt and loss risk, and this could shock Wall Street.
- The article was summarized using an artificial intelligence-based language model.
- Due to the nature of the technology, key content in the text may be excluded or different from the facts.
Meta·OpenAI and other US tech companies
Using special purpose vehicles
Investment in data centers, GPUs, etc.
Favorable for big tech credit but
Concerns about risk transferring to Wall Street
Meta, OpenAI and other major US big tech firms have been criticized for raising funds needed to build AI data centers through special purpose vehicles (SPVs), resulting in a substantial amount of related debt not being reflected on their corporate financial statements. Because this debt does not outwardly appear as big tech liabilities, it can be easily overlooked, but it could ultimately become a risk to financial markets.
According to the Financial Times (FT) on the 24th (local time), US big tech received large investments from SPVs to build data centers. US financial firms including PIMCO, BlackRock, Apollo, Blue Owl Capital and JPMorgan Chase supplied at least $120 billion (including equity investments) to SPVs. The structure is that the SPV owns physical assets such as data center sites, buildings and graphics processing units (GPUs), and big tech leases and uses them. The industry says using SPVs keeps borrowing off the big tech accounting books, which is advantageous for defending credit ratings.
Meta's planned 'Hyperion' data center in Louisiana is a representative example. In October, Meta established a $30 billion SPV called 'Benye Investor' with Blue Owl. $27 billion was borrowed from PIMCO, BlackRock, Apollo and others, and $3 billion was funded by Blue Owl's equity. The FT reported, "With this borrowing structure Meta has $30 billion of debt that was not on its financial statements," and "a few weeks later (Meta) found it easier to raise an additional $30 billion in the corporate bond market."
xAI, an AI startup founded by Elon Musk, Tesla's chief executive officer (CEO), is also reportedly preparing to use an SPV to raise about $20 billion in funds that would include $12.5 billion of debt. This SPV plans to buy NVIDIA GPUs with the raised funds and lease them to xAI.
As the funding needs for AI infrastructure among big tech rapidly increase, this structure could spread further. As a result, analysts say data center construction through this method is heavily dependent on the $1.7 trillion private credit loan market. There is a high possibility that lending could become concentrated on a small number of core AI firms, increasing risk. For example, OpenAI has currently signed long-term computing contracts worth more than $1.4 trillion. If OpenAI falters, multiple data center lenders would be exposed to risk simultaneously.
On Wall Street, more complex products have emerged that securitize data center-related debt into asset-backed securities (ABS). These bundled and securitized AI-related debts have spread data center loan risk to a broader investor base, including asset managers and pension funds. The FT warned, "If AI demand falls short of expectations, it is unclear who will be held accountable and who will bear the risks and losses," and "if financial stress hits AI operators in the future, it is difficult to predict how shocks would propagate across Wall Street."
Donghyun Kim Reporter 3code@hankyung.com
