Token Loans Arrive in China: Banks Use Token Consumption as Credit, First 30 Million Yuan Loan Lands in Guangdong

Forum topic · 小凯 · 2026-09-23

Summary

Chinese banks have begun using AI token consumption as a basis for credit approval. On September 22, 2026, China Construction Bank's Guangdong branch announced the first batch of Token Loan (Cidian Dai) business, granting a 30 million yuan pure credit loan to a national-level Little Giant enterprise in the Nansha Free Trade Zone. Credit is underwritten based on token consumption volume, computing service contract value, receivables, and token commission settlement—operational flow data rather than fixed-asset collateral. The shift reflects explosive growth in China's daily token usage, which rose from 100 billion in early 2024 to over 500 trillion by June 2026, a thousandfold increase in two years. Similar products rolled out nationwide within six months, including ABC's first Token Loan deal in Shanghai, a three-bank pilot in Guangzhou's Haizhu district offering up to 30 million yuan per borrower over 3 years, ICBC's computing power loan in Shenzhen, and a Beijing Economic-Development Zone program with nearly 2 billion yuan in total credit from six banks. Experts caution that token data lacks independent auditing, tokens differ in quality across models, and payback periods for different flagship models vary more than eightfold, so the model still needs standardization to mature.

Overview

On September 22, 2026, at the second Guangdong AI Application Matchmaking Conference's "AI + Finance" session, China Construction Bank (CCB) Guangdong Branch announced the first batch of its Token Loan product: a national-level specialized "Little Giant" enterprise in the Nansha Free Trade Zone received a 30 million yuan pure credit loan. Credit is underwritten on operational flow data—token consumption, computing service contract value, and receivables—rather than traditional fixed-asset collateral.

Treating tokens as credit is financial infrastructure's response to the industrial scaling of AI. China's daily token consumption grew from 100 billion in early 2024 to 100 trillion by end of 2025, 140 trillion in March 2026, and 500 trillion in June 2026—a thousandfold increase in two years.

Credit Logic

The product, launched by CCB Guangdong Branch with the Guangdong Token Trading and Service Center, shifts underwriting from "heavy on physical assets" to "heavy on operational flow," covering token suppliers, application developers, and service providers. Four core indicators replace traditional land, plant, collateral, and financial statements:

Token consumption volume
Computing service contract value
Receivables from computing business
Token commission settlement volume

Postal Savings Bank researcher Lou Feipeng noted the product breaks the static balance-sheet framework, using dynamic operational flows to identify growth potential and serve asset-light clients.

As of end-August 2026, CCB Guangdong's technology loan balance exceeded 540 billion yuan, serving over 38,000 tech enterprises.

Nationwide Rollout in Six Months

| Time | City | Banks | Key figures | |---|---|---|---| | 2026-06 | Shanghai Xuhui | ABC | First Token Loan service plan deal | | 2026-08-14 | Guangzhou Haizhu | BOC / CITIC / Guangzhou Bank | Guangdong's first batch; up to 30M yuan, 3-year term per borrower | | 2026-08 | Guangzhou Nansha | CCB Guangdong | First 30M yuan pure credit loan | | 2026-09 | Shenzhen Longgang | ICBC | 5M yuan credit line for Aizhihui Tech under "Computing e-Loan" | | 2026-09-13 | Beijing E-Town | ABC / CITIC / Industrial / Beijing / Minsheng / Huaxia | First batch nearly 2 billion yuan total credit |

The Guangzhou Haizhu pilot sized credit lines by contract value or token consumption; by late August, five firms had received 28 million yuan in approved credit, with 8 million yuan disbursed. In Beijing, Guoqi Zhikong and Shenzhou Guangda each received 30 million yuan lines.

Tokens as Capital—And Its Problems

In March 2026, the National Data Administration officially designated the Chinese name for Token and positioned it as a value anchor of the intelligent era. But challenges remain:

Tsinghua professor Zheng Weimin noted tokens are a new currency, but tokens are not equivalent—token factories do not equal high-quality token capacity.
Tencent Research Institute's Du Xiaoyu added that token "intelligence density" differs across models and tasks; a quality coefficient is needed for tokens to serve as a measure of value.
Market research firm Xinbao estimated the static payback period for the same AI server running different flagship open-source models differs by more than 8x, given official API pricing and infrastructure, power, and tax costs.
Sushang Bank researcher Wu Zewi pointed out token consumption and settlement data are mostly self-reported by enterprise platforms, lacking third-party independent audit mechanisms, risking data inflation—and data flow does not fully equal real cash flow.

Policy Backing

In July 2026, the People's Bank of China and eight other departments issued a notice pushing credit evaluation from "looking at collateral" toward "looking at data" and other dimensions—providing top-level cover for token loans. Bank products show high standardization: 3-year terms, up to 30 million yuan per borrower, and the same four underwriting indicators, enabling rapid replication across cities.

Outlook

| Dimension | Figure | |---|---| | Daily token consumption growth | >1,000x in two years (100B → 500T) | | Participating banks | BOC, ABC, ICBC, CCB, CITIC, Industrial, Beijing, Minsheng, Huaxia, Guangzhou Bank | | Beijing first-batch credit | Nearly 2 billion yuan | | Max per-borrower credit | 30 million yuan, 3-year term |

Token loans remain in a "pilot + rollout" stage. Whether they evolve from a credit instrument into a true value anchor depends on how standardized tokens become as a measure of value. As Gartner Managing VP Sun Xin summarized: "Computing demand is still growing fast, but demand growth does not mean every computing investment earns high returns. What truly differentiates returns is utilization, effective output, and the efficiency of the entire hardware-software stack."

> Source: zhichai.net forum post (September 2026)