By Zhang Yaqi

To stay ahead in the expensive artificial intelligence (AI) arms race while maintaining pristine financial statements, Silicon Valley tech giants are using complex financial instruments to move massive infrastructure spending off their balance sheets.

According to a Financial Times analysis on the 24th, tech companies including Meta, xAI, Oracle, and data center operator CoreWeave have utilized Special Purpose Vehicles (SPVs) to shift over $120 billion in data center financing debt to Wall Street investors. This strategy protects the credit ratings of the tech giants but has also raised market concerns about opaque risks and potential financial contagion.

In this financing frenzy, financial institutions like Pimco, BlackRock, Apollo, Blue Owl Capital, and banks such as JPMorgan have injected at least $120 billion into computing infrastructure projects structured through SPVs, via debt and equity. This innovative financing structure allows companies like Meta and Oracle to secure the enormous funds needed to build AI data centers without significantly increasing their on-book debt. This practice, nearly unimaginable 18 years ago, has now become an industry norm.

However, this financing boom may obscure the actual risks borne by tech conglomerates. While the SPV structure nominally isolates the debt, who ultimately pays the bill if AI demand falls short of expectations remains uncertain. Market participants worry that if AI operators face financial stress in the future, such pressure could spread unpredictably through the SPV structure across Wall Street and even the private credit market. Morgan Stanley previously estimated that tech companies' AI plans require about $1.5 trillion in external financing, suggesting the scale of such financing models is likely to continue expanding.

The "Perfect" Balance Sheet Game

Silicon Valley giants have long been known for their ample cash flow and minimal debt, granting them excellent credit ratings and investor trust. However, the race for advanced AI computing power is forcing these tech groups to shoulder unprecedented borrowing pressure. To avoid showing excessive leverage on their balance sheets, thereby protecting credit ratings and beautifying financial metrics, bringing in private capital through off-balance-sheet structures has become the preferred choice.

Reportedly, a senior executive at a major financing institution stated that due to their strong credit profiles, the tech industry can access more capital than any other sector. The basic logic of this structure is: tech companies do not borrow directly; instead, they raise funds through an SPV to build a data center and subsequently sign a lease agreement with it. In case of default, lenders' recourse is typically limited to the assets under the SPV's name—the data center, land, and chips—not the tech parent company managing these sites.

The Specific Maneuvers of the Giants

Meta completed a landmark deal last October. By partnering with New York financing firm Blue Owl Capital to establish an SPV named "Beignet Investor," Meta raised $30 billion for its planned Hyperion facility in Louisiana, including approximately $27 billion in loans from institutions like Pimco, BlackRock, and Apollo. This transaction allowed Meta to effectively borrow $30 billion without showing any debt on its balance sheet, paving the way for it to refinance another $30 billion in the corporate bond market a few weeks later.

Oracle is also actively leveraging third parties to construct large debt deals. The tech group led by Larry Ellison has entered into several agreements with partners like Crusoe and Blue Owl Capital. Among these, Blue Owl and JPMorgan invested about $13 billion into an SPV holding its Texas data centers, including $10 billion in debt financing. Furthermore, the company arranged two package financings for multiple data center projects across Texas, Wisconsin, and New Mexico, with scales of approximately $38 billion and $18 billion respectively. In these cases, Oracle agreed to lease the facilities from the SPVs.

Additionally, as part of its $20 billion financing round, Elon Musk's startup xAI adopted a similar structure, which includes up to $12.5 billion in debt. The SPV will use these funds to purchase Nvidia Graphics Processing Units (GPUs) and lease them to xAI.

Concerns in the Private Credit Market

As private capital investors eagerly seek to participate in the AI boom, "project finance" deals focused on long-term infrastructure financing have surged. According to UBS data, by early 2025, tech companies had borrowed approximately $450 billion from private funds, an increase of $100 billion year-on-year. Since the beginning of this year, about $125 billion has flowed into project financings similar to the Meta and Blue Owl deals.

This trend has heightened market concerns about the private credit industry, which has rapidly ballooned to $1.7 trillion, focusing on soaring asset valuations, illiquidity, and borrower concentration risks. A banker close to data center financing transactions noted that risky loans and potential credit risks have already accumulated in private credit. Since the AI data center boom heavily relies on a handful of clients like OpenAI (which has made long-term computing commitments exceeding $1.4 trillion in the industry), if a major tenant encounters problems, lenders to multiple data centers could face correlated exposure.

Moreover, these investors face uncertainties regarding power access, risks from AI regulatory changes, or hardware obsolescence due to technological iterations. Wall Street has even begun experimenting with securitizing AI debt, bundling loans for sale to a broader range of investors (such as asset managers and pension funds), with current transaction volumes in the billions of dollars.

Risk Exposure and Diverging Strategies

Although the financing structure aims to isolate risk, in many cases, the ultimate financial risk often still falls on the tech company leasing the site if demand for AI services declines, impairing the facility's value. For example, in the "Beignet Investor" case, Meta owns a 20% stake in the SPV and provided other investors with a "residual value guarantee." This means if the data center's value falls below a certain level and Meta decides not to renew the lease, it must repay the SPV investors.

Matthew Mish, Head of Public and Private Credit Strategy at UBS, pointed out that while most investors consider having ultimate exposure to "hyperscalers" a good thing, SPV financing actually increases the tech companies' outstanding liabilities, implying that the overall credit quality of these enterprises might be worse than current models suggest.

Not all giants have joined this off-balance-sheet financing trend. Companies like Google, Microsoft, and Amazon, which had mature data center businesses before the AI boom, currently still finance construction primarily through cash or direct bond issuance and have not disclosed any significant SPV financing plans. This divergence in market strategies reflects different players' varying considerations for risk control when facing this expensive AI wager.