Summary of Key Points
By August 2026, the five major American tech giants (such as Apple, Google, and Microsoft) had issued $208.1 billion in corporate bonds, which is 12 times the amount for the entire year of 2024. The funds were primarily used to build data centers required for AI development. Strangely enough, a significant portion of this debt “disappeared” from their financial statements—these companies manipulated the records to keep the debt off their own balance sheets. This raises three critical questions: Why do they hide the debt? How did Wall Street become the “boss” of the data centers? And what are the long-term risks of relying on rapidly obsolescing GPUs as collateral for these debts? Could this lead to an “AI version of the subprime mortgage crisis”?
1. Why Does the Debt “Disappear” Suddenly? — For a “Better Financial Report”
You can think of a company’s balance sheet as a personal “financial health check”: assets represent what you own, liabilities represent what you owe, and net assets are the difference between the two. If liabilities are too high, investors may see the company as risky, causing stock prices to fall, and banks may be reluctant to offer low-interest loans.
How do the giants make the debt disappear? Simply put, they use an intermediary: they set up a special purpose entity (SPV, or Special Purpose Vehicle) to borrow the money to build the data centers, and then rent the facilities from the SPV. This way, the debt is recorded on the SPV’s books, not on the giants’ own financial statements.
For example, imagine you want to buy a house for $1 million but don’t want the mortgage to affect your credit. You could have a friend set up a company to take out the loan, and you pay rent to the company each month. You live in the house, but the mortgage is not in your name. That’s exactly the strategy the giants use. The advantage is that their financial statements show lower liabilities, which reassures investors and allows them to continue borrowing at low costs to expand their AI businesses.
2. How Did Wall Street Become the “Decider” in Data Centers? — From Lenders to Landlords
Building data centers is expensive, and the giants don’t want to take on the debt themselves. So, Wall Street’s investment banks, private equity firms, and REITs (Real Estate Investment Trusts) stepped in: they funded the establishment of SPVs to build the data centers, and the tech giants pay rent monthly.
As a result, Wall Street became the “landlords” of these data centers, owning the assets while the tech giants are merely tenants. For instance, if Microsoft wants to build 10 AI data centers, it doesn’t have to borrow the money itself; it can have Goldman Sachs set up an SPV to build them, and Microsoft pays rent annually. Wall Street not only earns rent but can also package the data center assets into financial products to sell to other investors (such as pension funds and insurance companies), generating additional profits.
Now that AI data centers are the “infrastructure” of the AI industry, whoever controls the data centers holds a significant advantage. By doing this, Wall Street has transformed from a mere lender into a key player in the AI ecosystem, thus gaining more influence.
3. Long-Term Debt on Obsolete GPUs — A Time Bomb
The debt borrowed by the tech giants through SPVs typically has a term of 20 to 30 years, but GPUs (the core components of AI systems) are updated very quickly. The most advanced GPUs today may become obsolete in 5 to 6 years (for example, the chips used in GPT-4 might not be compatible with future versions of GPT).
This is like taking out a 30-year loan for a computer that will become useless in 5 years, but you still have to pay the remaining 25 years of the loan. The risks are:
- If the GPUs become obsolete, the value of the data centers will plummet;
- The tech giants may not want to renew their leases, leading to a loss of rental income for Wall Street;
- If the SPV, which holds the debt, is unable to repay, it could trigger a debt default.
Even worse, Wall Street has packaged these data center debts into financial products sold to many investors, meaning that a default could have a domino effect across the financial system.
4. Is an “AI Version of the Subprime Mortgage Crisis” Emerging? — Caution, but Not Yet at a Crisis Level
The subprime mortgage crisis occurred in 2008 when banks lent to people with poor credit and packaged those mortgages into financial products. When the borrowers couldn’t repay, the entire financial system collapsed.
There are similarities between the current situation and the subprime crisis:
- Both involve transferring risks: the giants shift the debt to SPVs, and Wall Street sells the SPV’s assets to investors;
- The value of the assets depends on future expectations: subprime mortgages relied on rising housing prices, while the current situation depends on the perpetual demand for AI and the durability of GPUs.
However, there are differences: the demand for AI is real (e.g., large models and autonomous vehicles rely on data centers), whereas the subprime crisis was fueled by loans to people unable to repay. Nevertheless, if AI development slows down or GPUs are updated too quickly, the risks could materialize. We are still in the early stages, but there are already signs of potential problems—just as no one thought housing prices would decline before the subprime crisis, no one is certain that AI demand will always remain strong.
In Conclusion
The giants hide their debt to improve their financial appearance, and Wall Street has seized control of the AI infrastructure. However, the contradiction between long-term debt and short-term technological obsolescence could lead to quietly accumulating risks—like an unexploded bomb that needs to be monitored for when it might detonate.