Token Consumption as Basis for Loans to AI Companies: A Major Shift in Chinese Banks’ Credit Models

On August 14, 2026, Guangzhou’s Haizhu District announced a new financial product called the “token loan.” Tokens are the smallest units of text processed by AI models. In this product, the number of tokens consumed during AI input and response becomes the core metric for determining loan approval. This forces a fundamental shift in banks’ credit assessment logic from “does the borrower have assets that can be collateralized?” to “how stably and consistently is the borrower consuming tokens?”

The Reality and Mechanism of Token Loans

To understand the token loan, it is necessary to grasp why traditional banks found it difficult to lend to AI companies. AI companies are asset-light businesses without fixed assets like factories or equipment, and their books often appear as cash burn. However, real commercial value can exist behind this. The problem was that traditional banks’ credit models could not identify this value.

The “Computing Capability Token Loan” launched by the Bank of China’s Guangzhou branch covers three business scenarios: computing capability supply, application, and service. Credit is based on four points: the company’s computing capability token production and consumption volume, the value of computing capability service contracts, accounts receivable generated from the computing capability business, and the volume of token fee settlements. The credit limit is up to 30 million yuan, with terms as long as three years. Guarantee methods are mainly credit guarantees and accounts receivable pledges, allowing for flexible combinations.

Newly established companies may also qualify for loans. If the original operating company can provide proof of continuous operation, it can apply for financing based on orders. Entrepreneurial teams spun off from major enterprises may actually find it easier to obtain credit if their token consumption is large and sustained. This signifies a shift in the bank’s credit model from “what you own” to “what you consume.” The Bank of China’s Haizhu branch has already provided over 400 million yuan in loans to computing capability-related enterprises.

Development from Traffic Loans and

Alternative Data The token loan did not emerge suddenly. Its predecessor was the “traffic loan” launched by Haizhu District in June 2025. The traffic loan did not look at fixed assets but based credit decisions on traffic consumption volume and platform quality. Within one year, it facilitated over 1 billion yuan in new credit to 90 digital marketing companies, proving the viability of a “financing based on operational data” model.

The token loan expands this logic into the broader field of the AI industry. In the USA, there have been practices of alternative data credit like Square Loans since 2019. This mechanism determines credit limits based on merchants’ real-time operational data and deducts repayments proportionally from daily sales. The distinctiveness of the token loan lies in positioning an alternative data metric unique to the AI industry—token consumption. This is both an innovation and a venture.

The Industrial Logic of 30 Trillion Tokens

According to data from the National Data Administration, as of the end of June 2026, the daily domestic token consumption volume had exceeded 30 trillion. This represents an increase of over a thousandfold from the daily 100 billion at the beginning of 2024. In the industrial era, electricity consumption was watched closely; in the internet era, it was traffic volume. In the AI era, token consumption becomes the anchor of value in the age of intelligence.

Large and sustained increases in token consumption indicate that AI businesses are actively operating. It is evidence that customers are calling models, inference is being performed, and services are being delivered. This is the source of Haizhu District’s confidence; the number of general AI enterprises has already exceeded 8,000. Token consumption is data supported by real industrial synergy, not fabricated.

At the same time, an industrial policy known as the “Eight Articles on Tokens” has been established, supporting the entire chain from token production to finance. On the supply side, support includes up to 50% of R&D investment and up to 5 million yuan for application demonstrations. This policy combination leverages bank funds through financial products, reduces enterprise costs through fiscal subsidies, and cultivates market demand through industrial policy, constructing a complete closed loop.

Nationwide Financial Innovation Competition

The token loan is not a solo performance in Guangzhou. The Beijing Economic-Technological Development Area released the “Ten Articles on Tokens” on August 7, setting a goal to expand the intelligent economy scale to over 400 billion yuan by 2030. Anhui Province published a provincial-level action plan on August 5, explicitly encouraging the development of token loans and model loans. Chengdu executed its first purely credit-based computing capability loan of 1.14 million yuan on August 13.

Shanghai is taking a different path with data asset tokenization. The Shanghai Data Exchange has facilitated multiple data asset-backed financing deals, and the Bank of China’s Shanghai branch provided a 20 million yuan credit facility. The Everbright Bank’s Shanghai branch also achieved China’s first data asset credit enhancement financing. As various cities explore different approaches to supporting the AI industry with finance, the model using token consumption as an indicator is becoming a strong option.

Other examples, such as drone technology that transmits tokens, also demonstrate the transmission capability of tokens. The article Drone Direct Satellite Connection, Achieving Real-Time Video Transmission via Token Transmission introduces scenarios where tokens function as the unit of data transmission.

Editorial Opinion

In the short term, the introduction of token loans is likely to lower the fundraising threshold for AI startups within China. By being freed from traditional collateral requirements, asset-light emerging companies with strong technical capabilities can more easily obtain growth capital. Banks are also likely to accumulate expertise in lending to the AI industry and tap into new customer segments. However, the urgent tasks include establishing proper measurement methods for token consumption and implementing systems to prevent misuse. In the long term, a key challenge is whether token consumption will solidify as a universal metric for measuring the value of AI companies. This indicator is specific to the AI industry, limiting its transferability to other sectors. Furthermore, unless a positive correlation between token consumption and actual revenue or profits is empirically demonstrated, there is a risk of a bubble expanding. Financial authorities will likely be compelled to build regulatory frameworks from the perspectives of data security and consumer protection.

The editorial team is closely watching whether the attempt to make token consumption a financial indicator will promote the healthy growth of the AI industry or fuel speculative investments. Specifically, when token consumption stems from model development races or short-term increases in utilization rather than genuine business growth, its sustainability becomes a critical question.

References

• “银行风控逻辑开始”看Token不看砖头”了”, by 听风译码 — 钛媒体, 2026-08-18T08:10:22.000Z (ARR)
• Source URL: https://www.tmtpost.com/8106372.html