An analysis said U.S. big tech corporations pouring astronomical sums into artificial intelligence (AI) used special purpose vehicles (SPVs) to keep as much as $120 billion (about 170 trillion won) in liability from appearing on their financial statements.
Big tech is issuing large bonds and taking on debt to invest in AI data centers, high-performance chips, and cloud facilities, and they enlisted Wall Street investment banks and asset managers so that this expense would not be recorded as liability on financial statements. As Wall Street and big tech intertwine over fundraising, there are concerns that if future AI demand falls or an AI bubble bursts, the shock could spread beyond the tech industry to the U.S. financial sector.
The Financial Times (FT) reported on the 24th that Meta, the parent of Facebook and Instagram; xAI, the AI startup of Tesla CEO Elon Musk; Oracle; and data center operator CoreWeave used such complex financial transactions to transfer AI investment-related liability off their balance sheets.
When these big tech corporations set up SPVs, Wall Street financial firms such as Pimco, BlackRock, Apollo, Blue Owl, and JPMorgan provide funding by purchasing stocks or bonds issued by the SPVs.
These off-balance-sheet financing deals also have the effect of making a corporation's reported performance look better than it really is, giving big tech the advantage of avoiding credit-rating downgrades, but they can be abused to hide AI investment risks, FT noted.
If AI corporations face financial difficulties in the future, the crisis could also spread to the U.S. financial industry in unpredictable ways. A senior Wall Street executive told FT, "Just 18 months ago, it was unimaginable, but now it has become common for tens of billions of dollars to flow into SPVs to finance data centers."
Data analyzed by FT shows Oracle has been the most aggressive in using these complex financial structures. Oracle borrowed $66 billion (about 96 trillion won) for AI through SPVs. Oracle is working with builders and financial firms such as Blue Owl, Vantage, Crusoe, and Related Digital to build data centers in Texas, Wisconsin, and New Mexico. Each SPV owns a data center, and Oracle leases the facilities.
FT explained that if a default occurs, investors who lent the money can exercise claims only against physical assets such as data center sites, equipment, and chips, and cannot hold Oracle, the facility manager, responsible for repayment.
In Oct., Meta set up an SPV called "Beignet Investor" to raise funds for the Hyperion data center planned in Louisiana. The SPV took out $27 billion in loans from Pimco, BlackRock, and Apollo, and raised $3 billion in Blue Owl equity, securing a total of $30 billion (about 44 trillion won). Although Meta effectively borrowed $30 billion to build the data center, thanks to the transaction through the SPV, the liability is not immediately reflected on the financial statements.
xAI also arranged $20 billion (about 29 trillion won) needed to purchase Nvidia graphics processing units (GPUs) through a separate SPV, and CoreWeave borrowed $2.6 billion (about 3.7 trillion won) via an SPV to fund performance of its contract with OpenAI.
FT assessed that the new forms of borrowing that emerged amid the AI boom carry high risk, in line with the rapid growth in related demand.
According to investment bank UBS, as of early this year, funds borrowed in the big tech private credit market totaled about $450 billion (about 650 trillion won), up $100 billion from a year earlier. UBS analyzed that $125 billion (about 181 trillion won) has flowed this year alone into project financing, a form of long-term infrastructure finance such as data center construction.
FT reported, "Big tech's AI data center construction heavily relies on the private credit market, which has rapidly expanded to about $1.7 trillion, but this market is riddled with risks such as asset price inflation, illiquidity, and borrower concentration." It also warned, "For now, big tech's financial capacity and credit ratings are solid, so the burden is not large, but the situation could change at any time."
There are also concerns that if SPV liabilities spread at the same time, the financial markets could be shaken. If multiple AI companies borrow through SPVs, even a small market shock could simultaneously transfer to private credit funds.
Another risk factor is that the current AI investment boom is led by a handful of companies, including OpenAI, the developer of ChatGPT, so weakness at an individual company could hit the entire industry. OpenAI alone has signed more than $1.4 trillion in long-term computing contracts. AI corporations including OpenAI also face uncertainties such as AI-related power shortages, regulatory changes, and the rapid obsolescence of AI chips, the key collateral assets.
A finance-industry source familiar with data center financing transactions told FT, "Risky loans and potential credit risks have already piled up in the private credit market," adding, "With the added risk of AI investment-related uncertainty turning loans sour, the next few years will be very interesting—and worrisome."
Not all big tech relies on SPV liabilities, however. Google, Microsoft (MS), and Amazon are choosing to use internal cash or borrow directly to build and expand data centers.