Summary of Key Points
The five major cloud giants (Amazon, Microsoft, Google, Meta, and Oracle) are facing substantial capital expenditure pressures in their bid to dominate the AI data center market. The growth rate of their cash flows (23%) far lags behind the high costs associated with building data centers (70%). To prevent debt from weighing down their financial statements, they have employed various strategies to hide their debt, such as using shell companies, leasing arrangements, and long-term contracts. Wall Street private equity firms (like Blue Owl and BlackRock), which have withdrawn due to regulatory constraints on traditional banks, have become the primary financiers of these debts. The article discusses whether this pattern could lead to an "AI-era subprime crisis." Experts believe that, given the current scale of the issue and strong demand from end-users, a systemic risk is unlikely, but there are potential concerns, such as the rapid depreciation of collateral (graphics cards) and the concentration of financial flows.
Why Do the Giants Hide Their Debt?
AI data centers are extremely costly to operate. It takes a decade or more to build a data center and see returns, involving the purchase of graphics cards, the construction of facilities, and the signing of long-term power supply contracts. The capital expenditure growth rate of the five giants (70%) significantly exceeds their operating cash flow growth rate (23%). If this trend continues, they could face financial difficulties by 2027.
The route of issuing public debt is not feasible for several reasons:
1. Limited market appetite: The subscription multiples for public debt (the amount of money interested in purchasing bonds relative to the issuance) have been declining; for example, Amazon's ratio has dropped from 5.3 to 1.6, indicating a more selective market.
2. Breaking shareholder expectations: In the past, the giants maintained zero debt and engaged in substantial share repurchases. Now, by reducing repurchases and borrowing heavily, they may deter funds that value stable dividends from investing in their stocks.
3. Impact on credit ratings: Increased debt can lead to lower ratings. Oracle, for instance, is on the brink of being downgraded to a junk status, and large buyers like insurance companies and pension funds only invest in investment-grade bonds.
Therefore, the giants must find ways to borrow money that are less noticeable, effectively hiding their debt.
The Giants' Debt-Hiding Tactics
Each company uses different methods to keep debt off their financial statements:
- Meta: Uses shell companies to isolate debt. When building data centers, Meta establishes a series of subsidiaries, holding only 20% of the equity. This way, Meta does not have to include the subsidiaries' debts (e.g., $27.3 billion) in its own financial statements. However, Meta agrees to lease agreements and residual value guarantees; if the lease is not renewed, Meta will cover the difference, effectively repaying the debt, though it is not reflected on its balance sheet.
- Microsoft: Leases data centers but actually buys them in installments over the life of the facility. The debt is recorded as "other current liabilities," which may make the total debt appear lower, but the lease debt has more than doubled (to $62.9 billion).
- Google: Provides guarantees for third-party loans, only recording a small portion of the debt on its balance sheet. For example, a guarantee of $43.8 billion is only recorded as $800 million.
- Amazon & Oracle: Utilize long-term leases to defer the recognition of debt. Amazon has signed a lease for $106.3 billion, and Oracle for $260 billion, but these leases have not yet begun, so no debt is reflected in the current financial statements. The debt will be gradually recognized over time.
Wall Street Private Equity Becomes the Main Financier
Traditional banks are withdrawing from financing AI infrastructure due to strict regulations. The Basel III framework requires banks to use more of their own capital (shareholder funds) when granting long-term, large loans, making it unprofitable for them. As a result, private equity firms (often backed by patient investors like pension funds and insurance companies) have stepped in to fund these projects. These firms are attracted by the stable, long-term rental income from data centers, with the giants being reliable tenants.
Examples include Blue Owl and BlackRock, which purchased land for data centers long before the AI boom and are now directly investing in related projects (e.g., holding 80% of Meta's shell companies' equity). The private equity credit market has seen AI-related loans grow from zero to $200 billion in three years, accounting for nearly 8% of the total, with an expected increase to $800 billion in the next three years (more than the software industry has accumulated in ten years).
Could AI Debt Trigger a Subprime Crisis?
There are potential risks:
1. Mismatch in collateral: Debt terms are typically 20 years, while graphics cards depreciate significantly within five years (H100 cards lose 55% of their value after three years). Giants may extend the depreciation period to reduce their costs.
2. Concentration of financial flows: Private equity firms often fund each other; for example, if Blue Owl issues bonds and PIMCO buys them all, a redemption crisis (e.g., if Blue Owl's funds redeem 40% but only receive 15% of the principal) could spread the risk.
3. Impact on ordinary investors: Private equity funds often use pension money, and any issues could affect retirement accounts.
However, experts believe that a systemic crisis is unlikely:
- Small scale: The scale of AI debt is negligible compared to the real estate market during the 2008 subprime crisis.
- Strong demand: The growth rate of Microsoft's Azure services (43%) indicates that businesses are indeed investing in AI infrastructure, which should generate returns (with an ROIC of nearly 30%).
- Strict bank regulation: Banks have lower leverage levels, so they are less likely to cause a financial crisis like the subprime mortgage crisis.
AI Changes the Game
Tech companies have shifted from a "light-asset" model (focusing on coding and software sales) to a "heavy-asset" model, where they invest in chips, power, and data centers using future cash flows to fund current construction. This shift reflects bets on the future of AGI (Artificial General Intelligence). If AI demand continues to grow, these investments will be profitable. However, the risk lies in whether these assumptions hold true; if demand slows or technology evolves, the debt could become non-performing.
In summary, the debt strategies associated with AI infrastructure represent a form of financial innovation in Silicon Valley and Wall Street. While these practices are currently controllable, there is a need to be cautious of potential bubbles, as the risks are merely being shifted rather than eliminated.