Summary
Chinese banks, telecom operators and a Guangzhou district are packaging AI tokens or model-use credits into card rewards, monthly service plans and business credit assessments. The developments point to the financialization of AI services, while also underscoring that these tokens are generally usage units rather than automatically tradable crypto-assets.
AI credits move beyond software subscriptions
Chinese banks, telecom operators and a district government in Guangzhou are experimenting with a new way to package artificial intelligence services: as units that can be rewarded, redeemed, billed and considered in credit reviews. A report by The Next Web describes arrangements involving a bank-linked credit card campaign, telecom AI packages and a lending initiative that uses a company’s token consumption as one input in financing decisions.
The developments suggest that AI is beginning to move from a conventional software product into a measurable service layer connected to payment and financial infrastructure. Yet the term “AI token” needs careful handling. In the cases described, it primarily refers to model-use credits or platform-denominated service units. It does not necessarily mean a blockchain token, an independently transferable asset or an instrument with a market price outside the issuing platform.
That distinction matters. A customer who receives AI credits may have the right to use a model for a specified period or up to a specified allowance, but may not own an asset that can be transferred or redeemed for cash. The operational and regulatory questions therefore concern service terms, billing, data records and credit risk as much as tokenization itself.
A credit card turns model access into a consumer benefit
Moonshot’s Kimi and Agricultural Bank of China have launched a campaign described on Kimi’s own page as an “AI-native credit card.” The product is available only in mainland China, and the gift campaign is scheduled to run through 30 September. First-time applicants who spend 5,888 yuan within three months can receive a co-branded plush charm and two months of premium membership. The campaign is limited to 1,000 people.
Structurally, the arrangement resembles a conventional card-acquisition or loyalty campaign with an AI service attached. The bank provides the payment and credit-card channel, while the AI company supplies membership access or related digital benefits. That is different from placing a separate financial asset in the hands of cardholders. The practical value of the benefit depends on the membership period, the models and features included, and the campaign’s eligibility terms.
The partnership also illustrates how AI services can enter established banking rewards systems. Card benefits have historically included merchandise, travel rewards and digital subscriptions. Model access and usage allowances can now be presented in a similar format. For consumers, however, the relevant questions are not limited to the headline reward. They also include whether the membership expires, whether usage is capped, whether access is limited by geography or account type, and what happens if the underlying service changes.
For banks and AI providers, the arrangement may create a familiar distribution channel for an emerging service category. It also raises ordinary but important issues around marketing clarity, data handling and service continuity. A benefit described as an AI token should make clear whether it is a membership, a usage allowance, a discount or another platform-specific entitlement.
Telecom operators apply a monthly-billing logic
China Telecom began trialling AI token packages on 17 May, with three tiers aimed at developers, small businesses and households. According to the carrier’s announcement, individual customers receive access to the operator’s Xingchen model and DeepSeek V3.2, while developers also receive GLM5. Industry reporting described the consumer entry tier as costing 9.9 yuan per month for 10 million tokens.
The carrier also plans a Tianyi Token system through which loyalty points can be redeemed for token packages. Management of the service is expected to take place through an operations platform called TokenHub. This connects AI resources to systems that telecom companies already use for account management, billing and loyalty programmes.
For households, the model resembles a monthly digital add-on. For developers and small businesses, it is closer to a prepaid budget for model inference or application development. The telecom structure could make AI access easier to understand because customers are already accustomed to recurring plans and account-based allowances.
The token count alone, however, does not fully describe the service. Different models may consume credits at different rates, and the effective allowance can depend on input length, output length, context windows and product features. Limits on concurrency, expiry, model switching or peak-period availability could also affect the customer’s actual access. If AI credits become a widespread billing unit, providers will need consistent definitions and plain-language disclosure comparable to the information telecom customers receive for data plans.
Guangzhou tests usage data as a lending signal
Haizhu district in Guangzhou went further by introducing Token Loan in August alongside eight supporting measures. The initiative focuses on the financing of young AI companies. Instead of relying primarily on plant and equipment, banks can bring token consumption, platform qualifications and payment-collection progress into the credit review. The report said Bank of China’s Guangzhou branch can set a credit limit based on a company’s contract or token usage.
The approach reflects a real challenge for AI businesses. A young model application company may own few machines or other conventional collateral while holding customer contracts, platform records and evidence of ongoing service demand. If those records are reliable and can be checked against contracts and incoming payments, they may provide lenders with additional information about a company’s operations.
Token consumption, though, is not the same as recurring revenue or repayment capacity. A company may generate high usage while testing a product, benefiting from subsidies or fulfilling a one-off project. Those conditions do not necessarily imply recurring cash flow. Model pricing changes, platform policy changes, service interruptions and customer concentration can all alter the meaning of usage data.
For that reason, consumption records are more appropriately treated as a supplementary credit signal than as a standalone basis for setting a facility. A prudent review would still need to examine the underlying contracts, the quality and concentration of customers, collection progress, cash flow and the company’s broader financial condition. The value of the data depends on whether the lender can verify who generated the usage, which service was consumed and whether the recorded activity is commercially meaningful.
Financialization creates a standards problem
The common thread across the three cases is the conversion of AI capacity into a recorded unit that can be connected to payments, loyalty programmes and financing workflows. This is a form of financialization, but it does not mean that AI tokens have become a single, standardized asset class. In practice, the word may describe different products: a membership entitlement, a model-call allowance, a prepaid service balance or an internal metric used in a credit assessment.
The important questions for institutional wallets, custody teams and corporate finance departments are therefore operational. Who issues and records the credits? Can they be transferred, or are they tied to a single account? What happens when a model is unavailable? Can a company reconcile consumption records with invoices and contracts? How can a lender audit the data? And how should the entitlement be described in accounting, procurement or credit documentation?
Without common definitions, two products carrying the same “token” label may have very different rights and risks. A usage credit can expire, depend on a particular provider or lose practical value when the underlying model changes. A financial institution assessing such data would need to distinguish platform activity from durable collateral and recurring income.
China’s initiatives remain examples of product trials and market experimentation. Their significance lies less in the label than in the infrastructure they are testing: AI services are being connected to card rewards, telecom billing, loyalty points and business credit assessment. Whether this model expands will depend on transparent measurement, clear customer rights and reliable records. It will also depend on whether lenders and service providers can use new operational data without treating AI consumption as a substitute for conventional due diligence.
Source: link