(Yicai) Aug. 17 -- Guangdong province has recently launched the “Token Loan,” which is the southern Chinese province’s first dedicated financial product for the “token economy,” aimed at better serving computing-power companies by using factors such as token output and consumption, computing-service contract values and other business data to provide firms with access to credit.

As the artificial intelligence industry continues to expand rapidly, computing power has become a core productive resource in the digital age. To better meet the financing needs of companies in the computing-power sector, several banks have been introducing specialized lending products for computing hardware companies since last year.

Often marketed under names such as “Computing-Power Loans,” some of these products incorporate computing-power metrics into their credit assessments to provide differentiated financing support for firms across the computing-power value chain, including chipmakers, server manufacturers, data centers and optical-module suppliers.

Bank of China, China CITIC Bank and Bank of Guangzhou, for example, rolled out new financial products at a press conference on Aug. 14. Bank of China released the “BOC · Computing Power Token Loan,” which provides a credit line of up to CNY30 million (USD4.4 million), with a maximum tenor of three years.

Unlike earlier computing-power lending products, which focused primarily on asset-heavy firms involved in computing infrastructure, the “BOC · Computing Power Token Loan” targets three areas, namely computing-power supply, applications and services. It consists of three sub-products, the Computing Token Supply Loan, the Computing Token Application Loan and the Computing Token Service Loan, covering all types of micro, small, medium-sized and large enterprises across the computing-power sector.

Alternative Collateral

What makes the “Token Loan” special is that it uses computing-token output and consumption, the value of computing-service contracts and the volume of token commission settlements as key indicators for credit assessment. Each time a company uses computing resources or processes a batch of data, it generates token consumption. Higher token consumption indicates more active use of computing services.

Some market observers compare the model to the “transaction-based lending” that emerged in the early days of e-commerce. An online merchant may have few tangible assets to pledge as collateral, but its transaction volume on an e-commerce platform can demonstrate the strength of its business operations, allowing a bank to extend credit based on that data. The same logic is now being applied to computing-power firms. They may lack traditional collateral, but their token activity, computing-service contracts and computing-related commissions can demonstrate the strength of their operations.

“What banks value is the ability of token consumption to provide a direct view into the actual level of business activity at AI companies,” said Dong Ximiao, chief economist at CMB-China Unicom Consumption Finance and deputy director of the Shanghai Institution for Finance and Development. Traditional bank lending relies heavily on collateral such as real estate, while many AI firms operate asset-light businesses.

Token consumption directly reflects the frequency and intensity with which large AI models are being used, making it a key indicator of customer activity, market acceptance of products and business sustainability.

This approach helps banks shift their risk-management focus from a static assessment of “assets and liabilities” toward a dynamic focus on “operating activity and cash flows,” Dong said. This could enable banks to more accurately assess a company’s stage of development and growth potential.

Credit Risks

Although incorporating computing power into credit assessment is increasingly viewed as an inevitable trend, the concept of using “computing power as proof of creditworthiness” remains at an early experimental stage and is still some way from widespread adoption.

Using computing power as a major credit-assessment factor currently presents several risks that need to be addressed, said Zeng Gang, president of the Tianfu Liyan Financial Research Institute.

One major challenge is auditing token consumption. The data is provided by platform operators, creating the possibility of inflated business scale through fake traffic. There is also a lack of independent third-party mechanisms for data verification. Moreover, token consumption does not necessarily translate into actual cash collections by a firm.

For cash flows to be considered reliable, computing-service contracts need to have sufficiently long terms, clearly defined rights and obligations as well as comprehensive mechanisms for handling defaults. Business stability is another concern. Demand for computing power fluctuates with cycles in the AI industry, while customer orders may be cancelled.

Dong also noted that the banking sector’s shift from relying on physical-asset collateral toward digital operating data is consistent with the development of the AI industry. However, broader adoption will require basic infrastructure for data standardization and authenticity verification, as well as risk-management models capable of distinguishing genuine business activity from fake traffic.

For now, such loans remain limited pilot programs. The model will need more time to mature, but it could eventually become an additional consideration for assessing the creditworthiness of technology firms.

Editor: Kim Taylor