Summary of the Core Content
This article discusses a phenomenon that challenges industry conventions: NVIDIA, in collaboration with six top Wall Street financial institutions managing over ten trillion dollars in assets, has rebranded GPUs (which typically depreciate within 3-5 years) as “digital real estate” and launched a computing power financing platform. This allows customers to purchase GPUs via loans, similar to purchasing a house with a mortgage. The financial institutions provide the funds, and the customers make monthly payments, with NVIDIA acting as the intermediary. Essentially, this approach transforms GPUs from rapidly depreciating hardware into assets that can be mortgaged over the long term. However, there is a significant time mismatch risk: the depreciation period of GPUs (3-5 years) does not align with the loan terms (20-30 years), shifting the risk to the financial institutions and the capital behind them. The market is skeptical about this move; on the day of the signing, NVIDIA’s stock price fell, and the price of credit insurance soared, indicating that Wall Street is essentially buying insurance for itself while entering into these agreements.
Detailed Analysis
1. Computing Power Mortgages: What Are the Plans of All Three Parties?
You can think of this as a “GPU version of a mortgage”:
- Financial Institutions (the six giants): With low global interest rates, trillions of dollars are seeking stable returns. By packaging GPUs as “digital real estate,” they can generate long-term, stable cash flows (similar to rent income), which meets their needs—recreating the success of data center REITs (Real Estate Investment Trusts) by turning heavy assets into investable financial products.
- Customers (cloud providers, AI companies): Building a data center with thousands of GPUs can cost billions; paying for it all at once would deplete their cash flow (for example, Google’s cash flow turned negative in the second quarter of 2026). With this financing model, they can spread the cost over monthly payments and build the infrastructure without exhausting their resources.
- NVIDIA: On the surface, NVIDIA acts as a facilitator, but it benefits the most: ① Customers can only purchase NVIDIA’s GPUs through loans, ensuring sales; ② Its CUDA ecosystem and financing services bind customers to its products, making it difficult for them to switch suppliers; ③ It becomes the “organizer of capital” for the AI infrastructure, with funds circulating within its ecosystem. Even more strategically, NVIDIA provides financial institutions with a 25% residual value guarantee, claiming that GPUs will retain their value while secretly fearing depreciation risks.
NVIDIA’s revenue model has also changed: it has shifted from selling chips for a one-time fee to a dual-model approach of selling chips and offering financing services, evolving from a product company to a “product + financial platform” company.
2. Can GPUs Be Used as Collateral?
For collateral to be effective, it must meet four criteria: stable cash flow, liquidity, value preservation, and a depreciation period that matches the loan term. Commercial real estate meets these criteria (stable rent, mature secondary markets, scarce land). However, what about GPUs?
- They勉强 fit the criteria: Computing power can generate income (like rent), and GPUs can be transferred (NVIDIA promises support for this).
- Fatal Weaknesses: Chips are updated every 3-5 years, and there is no guarantee of residual value; the depreciation period is significantly shorter than the loan term.
How does NVIDIA address these weaknesses? It reframes the issue by claiming that computing power is essential infrastructure (like electricity or the internet), arguing that the need for it in the AI era justifies its value. However, this effectively shifts the risk from customers and NVIDIA to financial institutions. If new chips emerge in three years and old GPUs become worthless, customers may be unable to repay their loans, transferring the risk to these institutions and the capital behind them.
3. Wall Street Buying Insurance While Signing Agreements: What Risks Does the Market Detect?
On the day of the signing, NVIDIA’s stock price dropped by 2.8%, and the price of credit default swaps (CDS) soared by 6 basis points, indicating that the market believes NVIDIA’s default risk has increased. Why?
- A Vicious Cycle of Financing: Money flows from financial institutions to customers, who then buy NVIDIA chips, boosting NVIDIA’s revenue and maintaining its valuation, allowing it to raise further funds. This cycle is self-sustaining (similar to the internet bubble in 2000, when Cisco lent money to customers to purchase its own equipment). If real demand fails to keep up, the entire chain could collapse.
- Time Mismatch: GPUs become obsolete in 3 years, but loans must be repaid over 20-30 years, relying on the assumption that AI will generate profits. The market is skeptical about this assumption and is buying insurance as a precaution.
4. Spending Faster Than Earning: How Large Is the Gap Between Asset and Revenue Streams?
The gap between spending (on assets) and revenue generation is enormous:
- Asset Side: In 2026, five tech giants (Microsoft, Google, etc.) spent a total of $750 billion on capital expenditures, with anticipated AI infrastructure investments of $3.5 trillion over the next three years—creating an ever-growing demand for computing power.
- Revenue Side: OpenAI’s revenue in 2025 was $13 billion, and Anthropic’s is around $60 billion, far from covering the required $3 trillion in investment. Moreover, more investment means higher revenue thresholds (due to expensive chips and limited electricity).
- Hidden Debt: The off-balance-sheet GPU procurement contracts and data center leasing commitments of these giants total $1.65 trillion, exceeding their on-balance-sheet debt (Meta’s hidden debt is 2.8 times its book value). Their cash flows are strained: Google had a negative cash flow of $5.9 billion in the second quarter, and Meta’s stock price plummeted by 91%. They rely on borrowing to cover their expenses.
5. The Competition Between China and the US: Financial Leverage vs. Autonomous Systems
China and the US have taken completely different approaches to AI infrastructure:
- US (NVIDIA Model): Using financial engineering to amplify demand through third-party debt, securitization, and residual value guarantees. The advantage is speed (faster capital accumulation and faster infrastructure deployment), but the downside is high leverage (risky as it depends on AI revenue).
- China (Autonomous Route): Relying on domestic chips, state investment, and long-term commitments (e.g., DeepSeek’s financing with five-year financial investor lock-ups and direct state fund participation). The advantage is stability (no external debt, reduced risk), but the downside is slower progress (it takes time to catch up with NVIDIA’s technology).
The core constraint for both approaches is the same: ultimately, they must rely on actual AI revenue to repay their investments. The US bets on the patience of the capital market, while China relies on its technological advancement.
Final Reminder: Watch These Three Indicators
There’s no need to rush to conclude that a bubble has formed, but these three indicators are worth monitoring:
1. NVIDIA’s CDS prices (the higher they rise, the more concerned the market is about default risks).
2. When the free cash flows of the five tech giants turn positive (indicating that revenue will cover investments).
3. When the financing platforms start to shrink (indicating that financial institutions are becoming cautious).
The essence of financialization is to “discount the future into the present.” Securitizing GPUs assigns a value to AI’s potential, but whether this value is fair can only be determined by time. Don’t mistake extending payment periods for actual repayment of debts.