Loan officers turn away young AI companies whose balance sheets are thin and whose income arrives unevenly. Five borrowers have been approved so far, for 28 million yuan between them.
Bank of China's Guangzhou arm has recently begun lending to AI companies on the strength of how many tokens they generate and burn through, with a single borrower able to draw a credit line of up to 30 million yuan (about $4.45 million) for as long as three years.
The facility, offered in the city's Haizhu district and called the BOC Computing Power Token Loan, also weighs what an applicant's computing service agreements are worth, the money owed to it from computing work, and fees settled according to token volumes. The bank calls it the first credit product in Guangdong province built around the token economy, covering how tokens are produced, how they move and how they are used, and channelling money toward businesses developing AI models.
Three of the five approved companies have drawn down 8 million yuan, spent mostly on computing power, and contracts covering a further 20 million yuan are being finalized.
Confirming that a borrower's usage is real remains manual work. The lender goes through the token settlement records that firms agree with the computing platforms they rely on, and sends staff to visit in person, where the client signs in and shows its consumption as it happens. The bank wants closer ties with the platforms instead: a company put forward by a platform would open a query ID with the bank, letting the lender inspect and track the token figures itself.
The product is one piece of a wider push into computing-power finance. Bank of China says that by the close of June 2026 it had served upward of 5,200 firms along the AI supply chain and directed more than 660 billion yuan of financing to them. Bank of Jiangsu runs a computing power loan of its own; SPD Bank's Beijing branch prices its credit off an index measuring green computing at data centers; and China Construction Bank's Shanghai branch, which lends to the large-model cluster in Xuhui district, counts state computing subsidies, incentive programs and intellectual property toward how much credit a borrower can get.
An industry analyst called the lending approach a genuine financial innovation, and put the demand down to what AI agents have made possible: companies of one or two people, financially weak but often growing fast, that need financing support all the same. The model is exploratory, the analyst cautioned, and building a new business model will take time. Spending on tokens, the analyst added, should not be equated with revenue: a business needs money flowing in as well as out.