Summary of Key Points
Recently, Guangdong introduced the province's first "Token Loan," a new type of loan product aimed at companies in the computing power industry. This loan does not rely on traditional collateral such as land or factories; instead, it assesses creditworthiness based on data such as the consumption of computing power tokens (e.g., the number of times large models are called), the value of computing power service contracts, and the amount of commission settlements. Multiple banks are accelerating their involvement in the computing power finance sector. However, experts believe that this model of using computing power to prove creditworthiness is still in its exploratory phase and can only serve as a supplement to traditional credit systems at present. This is due to risks such as data fraud and a disconnect between revenue and expenses, which need to be addressed before the technology can become widely adopted.
1. What exactly is a Token Loan? Similar but Different from "Sales Revenue Loans"
Many people compare Token Loans to "sales revenue loans" in e-commerce, where the latter use sales figures to demonstrate a company's operational capability. However, there are significant differences:
- The "revenue" in sales revenue loans represents actual cash received (e.g., money from product sales);
- Token consumption is merely an indicator of business activity (e.g., the number of times a company uses computing power to process data), which indicates that the service is being utilized but does not necessarily mean that money has been earned.
For example, an AI company may consume 1,000 computing power tokens per day, but the customers may not have made payments yet, or the payment amount may not cover the cost of using the computing power. Therefore, high token consumption does not equate to actual revenue generation.
2. Why Are Banks Competing for Computing Power Loans?
Banks are actively entering the computing power finance market for several reasons:
- The AI industry needs funding: Computing power is a critical resource in the digital age, but many AI companies have limited assets and struggle to obtain traditional loans.
- Targeting innovative customers: Future technology companies will be important clients for banks. By offering computing power loans, banks can secure these valuable customers early on and enhance their digital finance offerings.
- Policy and market momentum: Governments are supporting banks in serving AI companies, addressing their financing needs, and the rapid growth of the AI industry is driving this trend.
As Zeng Gang, the director of Tianfu Liyian Financial Research Institute, stated: "This represents an exploration of how technology finance can adapt to the AI industry, helping to overcome the lack of collateral for these companies."
3. Computing Power in the Credit System: Moving from "Asset-Based" to "Activity-Based" Evaluation
Banks' credit evaluation methods are evolving:
- In the past: They focused on tangible assets like land, factories, and equipment.
- In the era of innovation: They began to accept intangible assets such as patents and trademarks for collateral.
- Now: They include computing power data (e.g., token consumption and contract values) in their evaluations, shifting from a static balance sheet to a dynamic assessment of business activity.
Dong Ximiao, the chief economist at Zhaolian, explained that token consumption directly reflects the frequency of large model usage, indicating whether a company's products have market demand and helping banks assess the company's growth potential, rather than just looking at past assets.
4. Challenges with Computing Power Loans
Despite being innovative, computing power loans face several challenges:
- Data fraud: Token data is provided by the companies or platforms themselves, which could lead to inflated usage figures without independent third-party verification.
- Disconnection between revenue and expenses: High token consumption does not necessarily mean high profits. For example, frequent use of computing power may result in high costs and slow customer payments, leading to increased expenses without increased earnings.
- Business volatility: The AI industry is cyclical, with orders potentially canceling out suddenly, causing a sharp drop in token consumption and leading to misjudgments by banks.
- Lack of standards: Different companies use different definitions and calculation methods for tokens, making it difficult for banks to compare and evaluate them.
If these issues are not resolved, the widespread adoption of computing power loans could be risky.
5. Future Directions: Serving as a Supplement, Not a Replacement for Traditional Credit
Experts agree that computing power loans are currently in a pilot phase and cannot replace traditional credit systems. Dong Ximiao noted: "In the future, they will serve as an additional dimension for evaluating the creditworthiness of technology companies. Banks will consider both traditional assets and cash flows, as well as computing power data." To become widely adopted, three issues need to be addressed:
- Standardization of data: Uniform definitions and calculation methods for tokens are necessary.
- Verification of authenticity: Independent third-party audits are required to ensure the accuracy of the data.
- Risk management models: Banks need to develop models to distinguish between genuine business activities and fake traffic.
In summary, computing power loans are more like a "test field" at present, and it will take time for them to mature and become a mainstream solution.