Summary of Key Points
NVIDIA, in collaboration with six major financial institutions including Blackstone and Goldman Sachs, has established a $500 billion computing power financing platform. This platform uses a “Special Purpose Vehicle (SPV)” to help customers purchase chips and build data centers, allowing them to afford the substantial costs of AI infrastructure without impacting their own balance sheets. At the same time, NVIDIA repositions its chips from being rapidly depreciating hardware to assets that can generate long-term revenue, similar to commercial real estate, in an attempt to attract more investors. This move has sparked discussions about “narrative magic” and potential risks: some warn that it resembles the tactics used during the subprime mortgage crisis, while others argue that crises may arise from unexpected directions when everyone is focused on the risks.
How Does the Financing Platform Work?
The core concept of this platform can be explained using the example of buying a house:
If you want to buy a house but don’t want to take on the mortgage, you create an “SPV” (a shell company) that obtains the loan. You then sign a 20-year lease agreement with the SPV, committing to make monthly rent payments. Since the bank sees that the tenant is reliable, it is willing to grant the loan, and this debt does not appear on your personal financial statements.
Applying this to NVIDIA’s approach:
1. The six institutions lend money to the SPV, which is established specifically for this purpose.
2. The SPV uses the funds to purchase NVIDIA chips and rents them to customers in need of computing power (such as AI companies).
3. Customers pay rent monthly, and the SPV uses these payments to repay the loans and interest to the institutions.
4. NVIDIA acts as a guarantor: if the chips cannot be rented out at the end of the lease, it will compensate for 25% of the remaining value.
In simple terms, NVIDIA’s chips are purchased with money facilitated by NVIDIA itself, and ultimately become part of NVIDIA’s revenue—solving the customers’ funding issues while keeping its own financial statements clean.
Why Doesn’t NVIDIA Borrow Money Directly?
NVIDIA is cautious about damaging its image as a company with “abundant cash flow.” AI infrastructure is an expensive endeavor, with total costs estimated at $5.5 trillion (more than Japan’s annual GDP), and 80% of that needs to be financed through borrowing. Customers want to buy NVIDIA’s chips but don’t have the funds; NVIDIA wants to sell them, yet it cannot lend directly because its reputation requires it to appear financially strong. If its balance sheet suddenly shows several hundred billion in debt, rating agencies might downgrade it, and shareholders would become anxious, damaging its image.
Thus, traditional Wall Street tactics are employed: the debt is transferred through the SPV, allowing NVIDIA to sell chips without taking on direct debt. This is essentially a form of “manufacturer financing”—similar to how a car dealership recommends a loan from a finance company for the car purchase, only on a much larger scale.
Chips as Assets?
NVIDIA’s narrative redefines GPUs from rapidly depreciating hardware to assets that can generate revenue, similar to commercial real estate or toll booths:
- Longer usage life: Many customers are still using the A100 chips from 2020, and some have extended their leases to 2029 (a nine-year term).
- **NVIDIA’s “computing power rent”:” Just as the best land generates the highest rents, NVIDIA, holding the most advanced GPUs, can profit from the surge in global computing demand (similar to Ricardo’s theory of land rent).
This reframe is designed to attract new investors, such as pension and insurance funds that seek stable returns. They may not invest in tech stocks but are willing to buy assets that generate steady income.
What Are the Risks?
Berri from “The Big Short” warns about the dangers of “bull market tails,” suggesting that the use of structured and unnatural credit arrangements can be particularly risky. In a bull market, the focus is often on selling real assets, but as it nears its end, the products sold may be less valuable—similar to how mortgage-backed securities (MBS) during the subprime crisis were backed by poor-quality mortgages, now replaced by chips.
However, there are opposing views: if everyone constantly discusses risks and anticipates the worst-case scenarios, the crisis might arise from unexpected sources (e.g., a sudden decline in AI demand or the emergence of new technologies that replace GPUs).
In summary, this financing platform is NVIDIA’s latest attempt to continue its narrative around AI. Whether the underlying credit structure is sound depends on whether future computing power demand can support it. After all, even the best narratives rely on real-world demand to be effective.
(The entire analysis avoids technical jargon and uses everyday examples to explain complex financial concepts in a clear and accessible manner, making it understandable even for non-financial professionals.)