A Guangzhou district is testing whether the digital exhaust of an AI business can help it qualify for a bank loan. Haizhu District has introduced Token Loan, a financial product that uses token consumption, API-call records, computing-power contracts, and receivables as evidence when banks assess credit for asset-light AI companies. The Haizhu District Government calls it Guangdong’s first financial product specifically aimed at the token economy.
The idea is simple, even if the underwriting is not. Many AI startups have few physical assets to pledge, particularly when their main costs are cloud capacity, model access, engineers, and sales efforts. Their token spending can reveal whether they are actually using AI services, while API logs and contracts can provide a view of activity and expected revenue. The district’s program treats those records as credit evidence, not as collateral.
The first reported borrower was Guangzhou Tengyuan Digital Technology, which received a three-year, 3 million yuan pure-credit loan from Bank of China. The government said Bank of China’s initial group of preliminary approvals under the program totaled 28 million yuan. The launch puts a local financing mechanism behind a broader national discussion about how China should define and finance the infrastructure of the token economy.
From cloud bill to credit evidence
A conventional bank loan often depends on property, machinery, inventory, a long operating record, or a personal guarantee. That framework is poorly suited to a young AI business whose value may lie in software, customer relationships, and a fluctuating cloud bill. Token Loan tries to bridge that gap by making operational data part of the credit picture.
The participating institutions include Bank of China, China CITIC Bank, and Bank of Guangzhou. According to the district government, lenders can consider monthly token use, API-call logs, computing-power contract values, and accounts receivable. Those signals can indicate that a business is not merely claiming to work in AI, but is actively consuming model services or delivering AI products to customers.
The approach is most relevant to sectors that require frequent model calls, including short-form drama production, e-commerce marketing, and AI tools. A company’s token footprint might show recurring demand, but it does not by itself prove profitability or repayment capacity. Banks still need to judge customer concentration, pricing power, cash conversion, and whether token spending reflects a sustainable business rather than an expensive experiment.
That nuance matters because the word “token” can mean different things. In this program, it refers to units of computation used by language and other AI models, not cryptoassets. EastFrontier explored China’s wider effort to make tokens part of technology financing in Chinese banks building AI loans around chips, compute, and models. Haizhu’s product takes that idea to a more concrete local level.
A local response to an asset-light financing problem
The government’s framing reflects a real constraint on China’s AI ecosystem. Early companies can have fast-growing usage but limited fixed assets, while the cost of cloud compute and model inference arrives before revenue is fully established. Traditional credit models can therefore underweight the operating signals that matter most to software and AI-service businesses.
Haizhu’s experiment is not the first attempt to modernize technology finance, but it is notable for using AI activity data directly. The district says Token Loan is a pure-credit product. That could expand financing options for eligible businesses that cannot or do not want to pledge assets, though it also increases the importance of data quality and risk controls.
The program’s reported first borrower provides an illustration rather than a proof of scale. A 3 million yuan facility can support working capital, model usage, or product development, but it does not demonstrate that banks will lend at large volumes across the sector. The 28 million yuan preliminary-approval figure is also an early program number, not evidence of a mature credit market.
China’s growing use of tokens is the context behind the policy. The National Data Bureau has promoted “ci yuan,” or token, as a Chinese term for AI computation, and officials have described token use as a way to measure activity in the AI economy. EastFrontier recently covered a Nanjing course that made token use part of software training, a reminder that tokens are becoming a practical concept for both developers and financiers.
For lenders, the promise is better information. For startups, the promise is a financing route that better reflects the economics of cloud-based AI. Neither promise is automatic. API volume can be bought, subsidized, or bursty, and usage metrics can be misleading if they are detached from revenue or customer retention.
The next test is whether lenders can price the risk
The policy’s real innovation lies in credit assessment, not simply the loan label. A bank must decide how to convert token activity into a view of repayment risk. That requires data systems capable of verifying usage, understanding whether contracts are recurring, and distinguishing a healthy customer base from a short-lived spike in model calls.
Haizhu’s framework suggests that banks will combine multiple indicators rather than rely on one number. Computing-power contracts may point to capacity needs, accounts receivable may show commercial relationships, and API records can indicate product activity. Together, they may offer a more useful picture than a balance sheet alone for a company that owns little physical equipment.
The model could be copied elsewhere if it produces low default rates and useful credit decisions. But wider adoption will depend on standardization. Different model providers, cloud platforms, and banks will need common ways to verify token data and protect commercially sensitive information. It will also require clear rules about which datasets lenders can use and how a borrower can challenge an inaccurate assessment.
Local officials naturally see a larger opportunity. The district describes the token economy as a potential foundation for a major AI industrial cluster. That is a development goal, not a current valuation. The immediate facts are more modest: a new local credit product, three participating banks, one disclosed borrower, and an early batch of preliminary approvals.
Those facts may still matter. China’s AI sector needs more than models and chips. It needs working capital for the businesses that turn those technologies into services. By treating token use as a financial signal, Haizhu is testing whether a record of computation can become part of the infrastructure for funding the AI economy.