Tokens have evolved into a fundamental building block of the AI large-model era, and their production, trading, and consumption are giving rise to an entirely new economic landscape. In recent days, several major banks, including the Agricultural Bank of China and Bank of China, have launched "Token Loan" products, marking the first time that an AI company's token consumption volume has been used as the core basis for credit approval. This innovation breaks away from the traditional collateral-based lending model, accelerating the realization of the financial value of data elements.
On August 18th, Wang Pengbo, chief analyst at Botong Consulting, said that the token and computing power industries are currently developing at a rapid pace, and many small and medium-sized AI enterprises have short-term funding needs for working capital. However, these companies typically operate on an asset-light model and lack traditional collateral such as real estate. By launching Token Loans, banks are adapting to the new financing demands of the industry, incorporating contract data and token consumption metrics into the credit evaluation process. This approach allows for a more accurate assessment of a company's actual business operations and helps alleviate the problem of high financing barriers.
Banks Intensify Focus on the AI Computing Power Sector
On August 17th, the Agricultural Bank of China's official WeChat account announced that its Shanghai Xuhui Science and Technology Sub-branch has innovatively launched a "Token Loan" exclusive financial service plan. This plan uses a company's token usage demand as a key basis for credit approval, addressing the funding difficulties enterprises face in model training and application scenarios, and helping AI companies improve their computing power efficiency. The product has recently been officially implemented in the "Super Entrepreneur Cluster" (Yuanli Community) in Shanghai's Xuhui District, supporting Shanghai Huizhi Xing Information Technology Development Co., Ltd. in its token procurement and daily operations. This marks an upgrade in AI financial services from traditional credit models to scenario-based, productized approaches.
On August 14th, Guangdong Province's first special financial product for the token economy, the "Token Loan," was launched, with Bank of China, China CITIC Bank, and Guangzhou Bank introducing similar products. Token Loans have rewritten traditional credit logic, no longer relying on physical collateral such as factories or equipment. Instead, they incorporate multi-dimensional indicators like "computing power resources" and "scientific innovation index" into the credit evaluation system, using token output and consumption as the core evaluation basis. Taking Bank of China Guangzhou Branch as an example, unlike traditional credit models, the product design uses a company's computing power token output and consumption, computing power service contract value, accounts receivable generated from computing power business, and token commission settlement volume as the core credit basis. It offers enterprises a total credit line of up to 30 million yuan for a maximum term of three years.
Previously, companies in the digital economy sector were characterized by light assets, weak financial statements, and scattered cash flows. The "Token Loan" innovates on collateral methods, primarily offering credit, accounts receivable pledge, and order financing as guarantees, while also providing combined guarantees and collateral options to further increase credit limits. Currently, Bank of China's Guangzhou Haizhu Sub-branch has provided over 400 million yuan in financing to computing power-related enterprises in Haizhu District. Since July of this year, leveraging the bank-enterprise matchmaking facilitated by the Haizhu District government, the region has implemented computing power Token Loan credit approvals exceeding 28 million yuan.
Meanwhile, Guangzhou's Haizhu District has released the "Several Measures for Promoting the Development of the Token Economy in Haizhu District," covering everything from token production, circulation, and application to going global, with individual project support reaching up to 5 million yuan. This includes collaborating with financial institutions to launch financial products like "Token Loans" and "Token Revenue Rights Financing," providing interest subsidies for special loans for enterprises conducting token-driven businesses, with a maximum of 2 million yuan per enterprise annually.
Zeng Gang, dean of the Tianfu Liyan Financial Research Institute, stated that the most core innovation of the Token Loan is incorporating token consumption volume, computing power service contract value, accounts receivable, and token commission settlement volume into the credit basis, with guarantee methods primarily being credit plus accounts receivable pledge plus order financing. The logical chain behind this framework is that sustained token consumption represents genuine business traffic; with business traffic comes cash inflow, and with cash inflow comes debt repayment capacity. This logic is directionally sound, attempting to capture the "operational vitality" of the enterprise rather than the "asset thickness" favored by traditional credit.
Dong Ximiao, chief researcher at China Merchants Union, noted that sustained growth in token consumption indicates real market demand for a company's products or services, providing a foundation for revenue growth and indirectly connecting to the "source of living water" for debt repayment capability. However, tokens represent "revenue potential" rather than "net profit." High consumption comes with high computing power costs; if a company's pricing or collections are weak, it may experience "increased revenue without increased profit." Therefore, this indicator needs to be cross-validated with data such as cash flow and gross margin; otherwise, it cannot be directly equated with debt repayment capacity.
"This is a substantial step toward making tokens a financial credit reference target. Tokens can now serve as a core basis for bank credit approval, realizing the financial value of data elements," Wang Pengbo said. However, he noted that this is currently just a credit evaluation indicator, not a standardized, freely tradable financial asset. It remains subject to existing policies and regulations and is still a phased exploration within local pilot programs.
A New Entry Point for Technology Finance
Dong Ximiao stated that AI companies are typically asset-light operations, and token consumption directly reflects the frequency and intensity of large model invocations. It is a core indicator for measuring customer activity, product market acceptance, and business continuity. It helps banks shift their risk control logic from static "assets and liabilities" to dynamic "operating cash flow," enabling more precise judgment of a company's development stage and growth potential. At the same time, this represents an entry point for banks to develop technology finance and serve technology-oriented enterprises.
Why lend? Zeng Gang believes that the current banking industry faces continuously narrowing interest margins. The dual pressures of multiple LPR cuts and deposit competition have nearly maxed out profit space in retail and small business operations. Against this backdrop, competing for emerging AI industry clients is both a strategic necessity for expanding incremental markets and a forward-looking move to build future comprehensive financial relationships. Today's AI application startups may be small and micro loan customers; tomorrow they could be high-quality comprehensive corporate clients. The imagination space of this customer lifecycle compensates, to a considerable extent, for the risk premium of early bank involvement.
It cannot be ignored that this new track also brings new risk control challenges. Dong Ximiao pointed out that the main risks are data distortion and operational disconnect. The first is manipulation risk—token data can easily be inflated through "fake orders" or invalid calls, and there is a lack of unified audit standards. The second is volatility risk—AI business token consumption often exhibits pulsed characteristics, dropping sharply when projects end, which can easily lead to post-lending misjudgment. The most critical is the profit disconnect risk, where a company has high consumption but thin profits. If banks over-rely on this single indicator, they may overestimate a company's repayment capacity, leading to increased credit risk.
Regarding the dynamic fluctuations in computing power usage, Wang Pengbo suggested that banks should implement dynamic post-lending management mechanisms. For example, systems can monitor a company's token consumption data in real time, set clear usage warning lines, and trigger manual due diligence once sustained declines occur to verify the company's project operations. Additionally, credit lines can be periodically reassessed, with limits adjusted based on real-time computing power demand. When necessary, credit lines can be reduced or additional guarantees required to avoid credit risks arising from business contraction.
Will data like computing power increasingly enter the bank credit system in the future? Dong Ximiao believes this is a trend, but the process requires caution. It signifies banks evolving from valuing hardware asset collateral to valuing digital operational data, aligning with the development patterns of the AI industry. However, large-scale promotion requires first solving infrastructure issues such as data standardization and authenticity verification, and banks must build risk control models capable of distinguishing effective business from fake traffic. Currently, these loans remain small-scale pilots, and mature models will take time. In the future, they may become a supplementary dimension for credit to technology-oriented enterprises.
From the logic of industrial evolution, Zeng Gang believes development will roughly follow two tracks. The first is the product deepening direction: moving from point-based credit to chain finance. As the token economy scales up, computing power finance will no longer be limited to token consumption loans for single enterprises, but may extend to supply chain finance. The second is the data infrastructure direction: moving from trust dependency to auditable credentials. For Token Loans to transition from experimental products to large-scale standardized applications, the fundamental solution lies in resolving data credibility issues.
Regarding the feasibility of replicating this nationwide, Wang Pengbo believes there is a foundation for nationwide replication, but implementation will certainly show significant divergence. Cities with concentrated computing power industries, complete infrastructure, and smooth data interoperability will see rapid adoption, while regions with weak industries and a lack of credible computing power data will find it difficult to advance. The common challenge for nationwide pilots is the absence of a unified evaluation standard for computing power value, with differences mainly concentrated in the data verification stage, where no unified industry norms currently exist.