Chinese Banks Pilot 'Token Loans' Backed by AI Compute Data

Commercial banks in China are pioneering 'Token loans,' using large language model processing volumes and computing power consumption as credit evaluation criteria for artificial intelligence startups. According to data from the National Data Bureau cited by People's Daily, daily token calls in China surpassed 140 trillion by March, up more than a thousandfold from early 2024. Major commercial lenders—including Bank of China, China Construction Bank, and China CITIC Bank in Guangdong—have begun issuing loans based on verified operational data. However, industry observers warn that the model faces challenges regarding data verification, profitability mismatches, and reliance on state subsidies.

The emergence of 'Token loans' illustrates how commercial lenders are attempting to adapt risk management frameworks to the digital economy. By translating computational activity into credit lines, banks hope to alleviate persistent financing bottlenecks for asset-light technology startups that lack traditional real estate or equipment collateral.

Commercial banks in China are introducing a new credit mechanism known as 'Token loans,' using computational activity and large language model consumption data as collateral criteria for technology startups, according to a report by People's Daily. Tokens—the basic units of text processed by large language models—have seen explosive demand as China accelerates its artificial intelligence buildout. Data from the National Data Bureau indicates that daily token queries in the country reached over 140 trillion by March, rising more than a thousandfold from approximately 100 billion at the beginning of 2024. Despite this surge, AI startups frequently face traditional financing barriers. Because these firms operate on asset-light models with limited fixed assets or established financial track records, conventional commercial underwriting frameworks have historically viewed them as high-risk, leaving many capital-constrained during critical development windows. In response, several state-backed and joint-stock lenders in Guangdong province have adjusted their evaluation models. The Guangzhou branch of Bank of China recently rolled out a computing-power token loan program with three sub-products aimed at firms of varying sizes. China CITIC Bank's Guangzhou branch has integrated token consumption volumes into its credit approvals, while the Guangdong branch of China Construction Bank introduced a specialized loan assessing token generation and consumption, service contract values, accounts receivable, and settlement flows. Fu Yifu, a special researcher at Sushang Bank, told People's Daily that the innovation essentially converts computing power—a core production factor in the AI sector—into a quantifiable credit metric. Fu noted that the shift helps resolve financing hurdles for asset-poor innovators while steering commercial banks toward customized vertical lending. Nevertheless, industry observers point to significant structural risks. First, verifying data remains difficult because token metrics are typically provided by cloud providers or self-reported by enterprises without unified auditing standards, opening the door to artificial traffic inflation. Second, high token usage reflects operational activity rather than net income; elevated computing costs can lead to top-line growth without generating positive cash flow. Finally, many existing products still rely on local government compute vouchers, risk compensation reserves, and interest subsidies, raising questions over whether the lending model can maintain commercial viability without ongoing state support.