To fund the astonishingly costly AI arms race, tech giants are partnering with Wall Street to engage in off-balance-sheet financing in unprecedented ways. These innovative financial arrangements aim to shift astronomical levels of debt and risk off balance sheets to appease investors concerned about an AI bubble.
According to media reports on Wednesday, the details of three massive transactions that have recently come to light reveal the core trends of this movement. The financing plans of Meta, OpenAI, and Musk's xAI are raising hundreds of billions of dollars for AI infrastructure through complex joint ventures, special purpose entities, and agreements with extraordinary guarantees. These transactions are not only highly profitable but also carry considerable risk due to their novel structures.
The direct motivation behind this series of operations is that the soaring costs of the AI race pose a threat to the financial health of these giants. Previously, after Meta CEO Mark Zuckerberg warned that AI spending would increase significantly, the company's market value evaporated by about $300 billion in just a few days, highlighting the market's sensitivity and concerns. As a result, tech companies are eager to offload risks, while financial institutions, in pursuit of high returns, are also worried about the performance of these complex transactions once the AI frenzy subsides.
These groundbreaking financing schemes are reshaping Wall Street's trading landscape with their massive scale and complex structures. An analysis of the following three transactions clearly demonstrates how tech and financial giants are engaging in a high-risk capital game around AI.
<h2>Meta's &#34;Frankenstein&#34; Financing: Exchanging Guarantees for Off-Balance-Sheet Debt</h2>
The financing plan designed by Meta for a massive data center named Hyperion in Louisiana has been described by industry insiders as a &#34;Frankenstein&#34; financing that combines elements of private equity, project financing, and investment-grade bonds. Since Meta issued $30 billion in bonds last October, doubling its debt overnight, it urgently needed a financing method that would not increase its own liabilities.
This plan was orchestrated by Morgan Stanley, with the core idea being to have &#34;others&#34; borrow money for the Hyperion project. Specifically, the fund management company Blue Owl Capital invested about $3 billion to acquire 80% of the private equity in the data center joint venture (named Beignet Investor), while Meta retained a 20% stake with its initial investment of $1.3 billion. Subsequently, the joint venture completed the remaining financing by issuing $27 billion in bonds maturing in 2049, of which Pimco purchased $18 billion. The key point is that this massive debt is recorded on the balance sheet of the joint venture Beignet, not Meta.
This arrangement comes at a steep price. The interest rate on these bonds is as high as 6.58%, significantly above the 5.5% yield of bonds from similar companies to Meta. To ensure that this leasing agreement is not counted as long-term debt, Meta retained the right to exit the lease every four years in the contract. In exchange, Meta provided an &#34;unusual guarantee&#34;: if Meta exercises its exit right, it promises to compensate investors for all losses. Alexey Teplukhin, a managing director at Blue Owl, candidly stated, &#34;What we are pursuing is to achieve equity-like returns with fixed-income risk.&#34;
<h2>OpenAI's &#34;Scale&#34; Dilemma: Massive Loans Test Bank Underwriting Capabilities</h2>
The Stargate data center project, developed by OpenAI, Oracle, and SoftBank, has a financing structure that is not complicated in itself, but its unprecedented scale is challenging Wall Street's underwriting limits. The total cost of the project is as high as $38 billion, constructed by data center developer Vantage in Texas and Wisconsin.
As OpenAI, being a startup, cannot bear such a massive loan, and Oracle's credit rating is relatively low among tech giants, financing is conducted through traditional project loan models. Oracle signed a 15-year lease, with the rent used to repay the loan, while the loan itself is secured by the data center assets. According to transaction materials, for this loan named Jacquard, JP Morgan and Mitsubishi UFJ Financial Group formed a super syndicate of over 30 banks, a scale far exceeding the norm.
This five-year loan has an interest rate of about 6.4%, nearly two percentage points higher than the yields on bonds from Oracle's peers. Despite the attractive returns, banks are struggling to sell this debt to various investors to diversify risks. However, since JP Morgan only obtained a credit rating for this loan from a relatively small rating agency, Kroll, this debt cannot be sold to collateralized loan obligation (CLO) managers who control about $1.3 trillion in assets, reflecting the market's dilemma in assessing the unknown technology risks associated with AI.
<h2>xAI's &#34;Chip&#34; Financing: A Value Bet Under High Leverage</h2>
Elon Musk has also designed an aggressive financing scheme for his AI startup xAI, aimed at purchasing chips for the super data center codenamed Colossus 2. According to previous media reports, purchasing just 300,000 NVIDIA chips requires $18 billion. To meet this demand, Musk's ally Antonio Gracias established a financing tool named Valor Compute Infrastructure through his investment firm Valor Equity Partners.
This tool specializes in purchasing chips by selling private equity and leveraging billions of dollars in debt from private credit funds. According to media reports citing informed sources, private credit giant Apollo Global Management is assisting in arranging this debt. This transaction could ultimately raise $7.5 billion in equity and $12.5 billion in debt.
The risk structure of this financing is highly speculative. While the rent paid by xAI will be used to repay the five-year debt, the returns for equity investors will largely depend on the residual value of these chips after five years. Apollo is pitching the debt to investors at an interest rate as high as 10.5%, which may also include additional returns based on chip value performance. Some analysts are concerned that this model, where NVIDIA customers finance the purchase of its products through high leverage, may be creating a market bubble.
<h2>Trillion-Dollar Funding Gap: The AI Boom May &#34;Squeeze&#34; the Credit Market</h2>
The emergence of these transactions is driven by the enormous demand for capital in the AI industry. Pimco's Chief Investment Officer Dan Ivascyn warned, &#34;We haven't experienced a prolonged period of economic weakness for a long time, which often breeds more complexity and complacency.&#34; He pointed out that the scale of debt transactions flooding the market currently far exceeds any previous credit cycle.
JP Morgan's team of strategists has also issued a warning, believing that the construction boom of AI data centers over the next five years will require at least $5 trillion, which will &#34;squeeze&#34; every credit market. According to their estimates, even if the investment-grade bond market, leveraged finance market, and asset-backed securities market fully commit, there will still be a massive funding gap of $1.4 trillion.
This gap will ultimately need to be filled by large-scale private credit and even government funding. In such a macroeconomic context, the innovative financing methods that tech giants are employing may just be beginning

Wallstreetcn

Meta Platforms

The giants are "finding ways" for off-balance-sheet financing! These three "huge AI financings" are so "innovative" that the entire Wall Street is watching